bench: concurrent reads re-measured after the read fixes

Idle tank at 408f69e, h5py re-run in the same session. Contiguous reads
went from 0.25x to 1.44x h5py (full) and 0.12x to 6.3x (256x256
hyperslabs) on one thread; deflate full reads at 8 threads 887 -> 2943
MB/s (h5py processes 3042). Full chunked reads at 16 threads are still
0.69x-0.76x h5py processes; the issue stays open.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

.gitea/workflows/ci.yml

@@ -22,6 +22,9 @@ jobs:
         run: rustup component add rustfmt clippy
       - name: Install thumbv7em-none-eabihf target
         run: rustup target add thumbv7em-none-eabihf
+      - name: Install wasm32-unknown-unknown target
+        # ci-test.sh builds the reader and clawhdf5-wasm for the browser.
+        run: rustup target add wasm32-unknown-unknown
       - name: Install Python interop dependencies
         # The interop suites used to skip silently when python3/h5py were
         # missing, so they never ran in CI. Install them and make a missing
@@ -31,12 +34,20 @@ jobs:
           # cmake builds libz-ng-sys for the opt-in `fast-deflate` (zlib-ng)
           # steps in ci-test.sh; rust:latest does not ship it. The default
           # build (pure-Rust zlib-rs) does not need it.
-          apt-get install -y --no-install-recommends python3 python3-venv cmake
+          # hdf5-tools: h5ls/h5stat/h5dump/h5diff, which the h5rs
+          # (clawhdf5-tools) interop tests compare against.
+          apt-get install -y --no-install-recommends python3 python3-venv cmake hdf5-tools
           python3 -m venv /opt/interop
-          /opt/interop/bin/pip install --no-cache-dir h5py numpy netCDF4 xarray
+          # maturin + pytest: ci-test.sh builds the Python package
+          # (crates/clawhdf5-py) and runs its tests against h5py.
+          /opt/interop/bin/pip install --no-cache-dir h5py numpy netCDF4 xarray hdf5plugin maturin pytest
           echo "/opt/interop/bin" >> "$GITHUB_PATH"
       - name: Show interop library versions
-        run: /opt/interop/bin/python -c "import h5py, netCDF4; print('h5py', h5py.__version__, 'HDF5', h5py.version.hdf5_version, 'netCDF4', netCDF4.__version__)"
+        # h5dump's version too: the h5rs dump test requires its exact output
+        # (checked against Debian's 1.14.5 in rust:latest and 1.14.6).
+        run: |
+          /opt/interop/bin/python -c "import h5py, netCDF4, hdf5plugin; print('h5py', h5py.__version__, 'HDF5', h5py.version.hdf5_version, 'netCDF4', netCDF4.__version__, 'hdf5plugin', hdf5plugin.version)"
+          h5dump --version
       - name: Run CI script
         env:
           # Name the interpreter outright rather than relying on $GITHUB_PATH

.gitea/workflows/conformance.yml

@@ -0,0 +1,56 @@
+name: Conformance
+# Nightly: read every file of the pinned public HDF5 corpora with clawhdf5 and
+# with h5py/libhdf5 and compare (conformance/run.sh; CONFORMANCE.md explains
+# the method). Fails on any panic, hang, crash or out-of-memory in clawhdf5,
+# and when the ok count drops below conformance/baseline.json or a file the
+# baseline lists as ok stops being ok. The report is printed into the job log;
+# nothing is uploaded (artifact actions are JavaScript, which rust:latest
+# cannot run — see CLAUDE.md).
+on:
+  schedule:
+    - cron: "17 3 * * *"
+  workflow_dispatch:
+jobs:
+  conformance:
+    runs-on: ubuntu-latest
+    container: rust:latest
+    timeout-minutes: 60
+    env:
+      CARGO_NET_RETRY: "10"
+    steps:
+      # Plain git, not actions/checkout (a JavaScript action; see ci.yml).
+      - name: Check out
+        run: |
+          git init -q .
+          git remote add origin "${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}.git"
+          for i in 1 2 3; do git fetch -q --depth 1 origin "${GITHUB_SHA}" && break; sleep 5; done
+          git checkout -q FETCH_HEAD
+      - name: Install h5py, h5dump and the probe's codec libraries
+        # hdf5-tools: h5dump for the CVE-corpus comparison. libaec-dev and
+        # pkg-config: the probe builds clawhdf5-format with `szip` (the core
+        # crates' default build needs neither).
+        run: |
+          apt-get update
+          apt-get install -y --no-install-recommends python3 python3-venv hdf5-tools libaec-dev pkg-config
+          python3 -m venv /opt/conformance
+          /opt/conformance/bin/pip install --no-cache-dir -r conformance/requirements.txt
+          /opt/conformance/bin/python -c "import h5py, hdf5plugin; print('h5py', h5py.__version__, 'HDF5', h5py.version.hdf5_version, 'hdf5plugin', hdf5plugin.version)"
+          h5dump --version
+      - name: Probe unit tests
+        run: cargo test --release --manifest-path conformance/probe/Cargo.toml
+        env:
+          CARGO_TARGET_DIR: conformance/.cache/target
+      - name: Sweep
+        # The corpora come from GitHub (pinned commits, conformance/corpus.txt),
+        # so this job needs a runner that reaches github.com.
+        env:
+          CLAWHDF5_PYTHON: /opt/conformance/bin/python
+        run: bash conformance/run.sh
+      - name: Report
+        if: always()
+        run: |
+          if [ -f CONFORMANCE.md ]; then cat CONFORMANCE.md; else echo "no report was generated"; fi
+          if [ -f conformance/.cache/results/summary.md ]; then
+            echo; echo "---- per-file detail (conformance/.cache/results/summary.md) ----"
+            cat conformance/.cache/results/summary.md
+          fi

.gitignore

@@ -5,3 +5,5 @@ benchmarks/longmemeval/*.json
 # Local model weights (MiniLM etc.) — large, not committed
 weights/
 .venv
+__pycache__/
+.pytest_cache/

BENCHMARKS.md

@@ -243,6 +243,30 @@ index asked for ~16 000 candidates, where scanning the few hundred or thousand
 allowed records is exact and cheap. Re-ranking a 3k candidate pool and
 confidence rejection add about 3%.
 
+### Signed checkpoints
+
+Measured 2026-09-25 on tank (AMD Ryzen 7 7800X3D). A default store (float16,
+int8 index), 384-dim; each checkpoint rewrites the whole file, as every
+checkpoint does. Medians of five checkpoints and three verifies; three runs
+agreed to within the ranges shown.
+
+```bash
+cargo run --release -p clawhdf5-bench --bin search_harness -- --signing-study --full
+```
+
+| N | checkpoint, unsigned | checkpoint, signed | signing adds | `verify` | file size added |
+|---:|---:|---:|---:|---:|---:|
+| 1 000 | 5.4 ms | 6.4 ms | 0.7–1.0 ms | 2.1 ms | 0.03 MiB |
+| 10 000 | 46 ms | 55 ms | 8.1–9.4 ms | 18.6 ms | 0.31 MiB |
+| 100 000 | 495 ms | 598 ms | 89–112 ms | 247 ms | 3.05 MiB |
+
+Signing costs about 20% of a checkpoint: every record is rehashed (SHA-256)
+and the Merkle root recomputed each time; the Ed25519 signature itself is
+microseconds. Caching per-record hashes between checkpoints would cut this to
+the records that changed. The per-record hashes stored for locating edits are
+32 bytes each (4% of a 100K float16 store). `verify` reads and rehashes the
+whole checkpoint.
+
 ### float16 embedding storage (`MemoryConfig::float16`)
 
 Measured 2026-09-23 on tank (AMD Ryzen 7 7800X3D). The same clustered
@@ -458,6 +482,180 @@ The rows and columns of the uncompressed layouts are within 20% (chunked
 column 0.45 -> 0.49 ms, contiguous column 2.55 -> 2.61 ms). This run does not
 explain the slower windows.
 
+## Concurrent reads
+
+### Results after the read fixes (2026-09-26, tank, `408f69e`)
+
+Same machine, files and commands as the first run below, re-run on an idle
+tank (load average 1.60 at the start; the 1-minute figure rose to about 5
+during the clawhdf5 runs, mostly their own threads) after two fixes:
+contiguous reads back their output with transparent huge pages and copy
+hyperslabs run by run, and full chunked reads no longer queue behind a
+one-thread rayon pool. h5py was re-run in the same session.
+
+Each read decoding on its calling thread (`--decode-threads 1`, like h5py):
+
+| layout | mode | threads | clawhdf5 MB/s (eff) | h5py threads MB/s (eff) | h5py processes MB/s (eff) |
+|---|---|---:|---:|---:|---:|
+| deflate | distinct | 1 | 606 (1.00) | 432 (1.00) | 421 (1.00) |
+| deflate | distinct | 4 | 1816 (0.75) | 428 (0.25) | 1654 (0.98) |
+| deflate | distinct | 8 | 2943 (0.61) | 428 (0.12) | 3042 (0.90) |
+| deflate | distinct | 16 | 2142 (0.22) | 375 (0.05) | 3083 (0.46) |
+| deflate | same | 1 | 154 (1.00) | 130 (1.00) | 129 (1.00) |
+| deflate | same | 4 | 599 (0.98) | 129 (0.25) | 499 (0.97) |
+| deflate | same | 16 | 1592 (0.65) | 128 (0.06) | 1399 (0.68) |
+| contiguous | distinct | 1 | 13665 (1.00) | 9490 (1.00) | 8781 (1.00) |
+| contiguous | distinct | 16 | 12674 (0.06) | 2285 (0.02) | 6942 (0.05) |
+| contiguous | same | 1 | 31991 (1.00) | 5087 (1.00) | 5078 (1.00) |
+| contiguous | same | 16 | 237151 (0.46) | 4304 (0.05) | 35772 (0.44) |
+
+With the default rayon pool: deflate `distinct` 2117 MB/s at 1 thread (4.9x
+h5py), 3163 at 4, 2341 at 16 (0.76x h5py processes); deflate `same` 1439 MB/s
+at 16; contiguous as above within a few percent.
+
+Before -> after for clawhdf5 (`--decode-threads 1` unless noted):
+contiguous full read at 1 thread 2495 -> 13665 MB/s (0.25x -> 1.44x h5py);
+contiguous 256 x 256 hyperslabs at 1 thread 624 -> 31991 MB/s (0.12x ->
+6.3x); deflate full reads at 8 threads 887 -> 2943 MB/s; deflate
+hyperslabs at 16 threads 1244 -> 1592 MB/s.
+
+Read with care:
+- `contiguous same` reads 1024 slabs of one 64 MiB dataset over and over, so
+  it mostly measures copies out of the CPU's caches (the 7800X3D has 96 MiB
+  of L3); the per-call overhead is what differs (h5py's is about 50 us).
+- At 16 threads every tool dropped in this run (h5py threads on contiguous
+  data from 8002 to 2285 MB/s, processes from 12846 to 6942), so the
+  16-thread rows are noisier than the others.
+- Still behind: full reads of chunked data at 16 threads (0.69x-0.76x h5py
+  processes). See `docs/known-issues.md`.
+
+### First run, before the read fixes (2026-09-26, tank, `91644d8`)
+
+Measured on tank (AMD Ryzen 7 7800X3D, 8 cores / 16 threads, 61 GiB, Linux
+7.0) at commit `91644d8`, load average 1.84 when the run started (the
+1-minute figure rose to 3.7 during the runs; that is mostly the benchmark's
+own threads). Warm page cache. clawhdf5 2.7.0 (workspace), h5py 3.16.0 on
+HDF5 2.0.0. Commands exactly as in the **Run** box below; files at their
+defaults (64 datasets of 16384 x 1024 `f32`, 64 MiB each; deflate chunks
+256 x 256, level 4). MB/s is decoded data, the median of the repetitions;
+eff is scaling efficiency against the same tool's 1-thread row.
+
+Each read decoding on its calling thread (`--decode-threads 1`, like h5py):
+
+| layout | mode | threads | clawhdf5 MB/s (eff) | h5py threads MB/s (eff) | h5py processes MB/s (eff) |
+|---|---|---:|---:|---:|---:|
+| deflate | distinct | 1 | 421 (1.00) | 433 (1.00) | 421 (1.00) |
+| deflate | distinct | 4 | 890 (0.53) | 428 (0.25) | 1651 (0.98) |
+| deflate | distinct | 16 | 880 (0.13) | 427 (0.06) | 4424 (0.66) |
+| deflate | same | 1 | 151 (1.00) | 130 (1.00) | 129 (1.00) |
+| deflate | same | 4 | 490 (0.81) | 129 (0.25) | 497 (0.96) |
+| deflate | same | 16 | 1244 (0.52) | 128 (0.06) | 1402 (0.68) |
+| contiguous | distinct | 1 | 2495 (1.00) | 9789 (1.00) | 9169 (1.00) |
+| contiguous | distinct | 16 | 8083 (0.20) | 8096 (0.05) | 12272 (0.08) |
+| contiguous | same | 1 | 624 (1.00) | 5022 (1.00) | 5172 (1.00) |
+| contiguous | same | 16 | 4778 (0.48) | 4411 (0.05) | 37138 (0.45) |
+
+With the default rayon pool decoding inside each read, deflate `distinct`
+is 912 MB/s at 1 thread (2.1x h5py) and 2824 MB/s at 16 (6.6x h5py threads,
+0.64x h5py processes); the other rows are within a few percent of the table
+above. Full tables (2, 4, 8 threads, both decode modes) come from
+`compare_concurrent_read.py` on the JSON files.
+
+What this shows:
+- **h5py threads do not scale** (flat at about 430 MB/s on deflate, every
+  thread count): libhdf5's global lock.
+- **clawhdf5 threads on one `File` do, for hyperslab reads of compressed
+  data:** 1244 MB/s at 16 threads, 9.7x h5py threads and 0.89x h5py
+  processes, without a process pool.
+- **Where clawhdf5 is behind** (open performance bugs, see
+  `docs/known-issues.md`):
+  - *Full reads of chunked datasets stop scaling at about 4 threads*
+    (about 880 MB/s) while h5py processes reach 4424 MB/s. Hyperslab
+    reads, which bypass the `File`'s chunk cache, keep scaling, so the
+    cache (one mutex and one 16 MiB budget per `File`, thrashed by 64 MiB
+    datasets) is the suspect. The cause of the `--decode-threads 1`
+    ceiling was not the cache: every full read queued its chunks for the
+    pool's single rayon worker. That case was fixed after these
+    measurements (2026-09-26, not yet re-measured here). With the default
+    pool the gap to h5py processes remains (see `docs/known-issues.md`).
+  - *Contiguous reads are slow*: 2.5 GB/s for a single-threaded full read
+    against h5py's 9.8 GB/s (0.25x), and 0.12x for 256 x 256 hyperslabs.
+    Threads close the gap (about 1.0x h5py at 16), but single-thread
+    contiguous I/O is a real deficit.
+
+The question: libhdf5's threadsafe build serialises every API call under one
+global mutex, and h5py holds a global lock around every call too, so threads
+reading through h5py cannot decode in parallel; h5py users scale with
+processes. A clawhdf5 `File` is `Send + Sync`, and nothing on the read paths
+this harness uses (`read_f32`, `read_f32_selection`) takes a library-wide
+lock: the one mutex is the `File`'s chunk cache (keyed per dataset), taken by
+full reads of chunked datasets for each chunk's O(1) lookup and insert, never
+across a decode; hyperslab reads do not use the cache. How does
+decoded throughput scale with threads on one open file, against h5py threads
+and h5py processes on the same files?
+
+Workload (`crates/clawhdf5-bench/src/bin/concurrent_read.rs`; the h5py script
+mirrors it): `<dir>/deflate.h5` and `<dir>/contiguous.h5`, each with 64 `f32`
+datasets of 64 MiB decoded (`[16384, 1024]`; the deflate file chunked
+`256 x 256`, level 4), written by clawhdf5 on first use and reused while
+`manifest.json` matches. The data is a slowly varying ramp plus 8 bits of
+noise per element, every value exact in `f32`, so both harnesses check what
+they read; it deflates about 3.1x (128 MiB -> 40.7 MiB for two 64 MiB
+datasets). For each layout and thread count
+(1, 2, 4, 8, 16; fixed total work per repetition, split among the threads):
+
+- `distinct`: every dataset read in full once, thread `t` taking datasets
+  `t, t + T, ...`;
+- `same`: 1024 random `256 x 256` hyperslabs of `d00` in total, from a seeded
+  splitmix64 stream that both harnesses generate identically.
+
+Reported per row: MB/s of decoded (selected) data from the median of the
+repetitions, and scaling efficiency `MB/s(T) / (T x MB/s(1))`. Each worker
+times itself from a start barrier; a repetition spans the earliest start to
+the latest finish. Page cache: warm by default (each file is read once before
+timing); `--cold` evicts the files with `posix_fadvise(POSIX_FADV_DONTNEED)`
+before every repetition (no root needed; best effort). clawhdf5 opens one
+`File` per repetition, shared by all threads; h5py threads share one
+`h5py.File`; h5py processes (spawned before timing) each open the file inside
+the timed region.
+
+Decode inside a single clawhdf5 read is itself parallel in this binary
+(clawhdf5-format's `parallel` feature, enabled here through clawhdf5-agent;
+it is off in the facade's default features), so a 1-thread clawhdf5 full read
+of the deflate file already uses the whole rayon pool. Run both
+`--decode-threads 1` (each read decodes on its calling thread, like h5py —
+this isolates the API's own scaling) and the default pool.
+
+> **Run** (from the repository root). The default files take about 5.4 GiB
+> of disk (4 GiB contiguous + about 1.3 GiB deflate). Generating them is
+> memory-hungry because `FileBuilder` holds a whole file in memory: peak RSS
+> was 676 MB for `--datasets 2 --mib 64` (2026-09-25, tank,
+> `/usr/bin/time -f %M`), about 5x one file's decoded size, so expect about
+> 21 GB at the defaults (once; later runs reuse the files). Put `--dir` on a
+> real disk, not tmpfs, if `--cold` is to mean anything.
+>
+> ```bash
+> DIR=/path/on/disk/concurrent-read
+> BENCH=crates/clawhdf5-bench/scripts
+> PY=.venv/bin/python   # h5py 3.16 / HDF5 2.0 in this repo
+> cargo build --release -p clawhdf5-bench --bin concurrent_read
+> B=target/release/concurrent_read
+> $B --dir $DIR --json claw-pool.json                    # generates on first run
+> $B --dir $DIR --decode-threads 1 --json claw-1.json
+> $PY $BENCH/concurrent_read_h5py.py --dir $DIR --executor threads   --json h5py-threads.json
+> $PY $BENCH/concurrent_read_h5py.py --dir $DIR --executor processes --json h5py-procs.json
+> $PY $BENCH/compare_concurrent_read.py claw-1.json    h5py-threads.json h5py-procs.json
+> $PY $BENCH/compare_concurrent_read.py claw-pool.json h5py-threads.json h5py-procs.json
+> ```
+>
+> Cold page cache: add `--cold` to every harness command. Smoke test (seconds):
+> `$B --dir /tmp/cr --datasets 4 --mib 1 --threads 1,2,4 --slabs 16 --reps 1`
+> and the same `--threads/--slabs/--reps` to the h5py script.
+
+Other flags (both harnesses): `--threads`, `--reps`, `--slab`, `--slabs`,
+`--seed`, `--modes distinct,same`, `--layouts deflate,contiguous`; sizes
+(`--datasets`, `--mib`) only on the Rust harness, which writes the files.
+
 ## Search harness baseline (v2.3.0)
 
 Produced by `cargo run --release -p clawhdf5-bench --bin search_harness -- --full`

CHANGELOG.md

@@ -2,7 +2,451 @@
 
 ## Unreleased
 
+### Concurrent reads (2026-09-26)
+- **Full reads of chunked datasets scale with threads again when rayon's
+  pool has one thread.** Each full read handed its chunks to rayon to
+  decode; with a one-thread pool (`RAYON_NUM_THREADS=1`, or
+  `concurrent_read --decode-threads 1`) every thread reading through a
+  `File` queued behind that single worker, so N readers decoded on one core
+  and throughput stopped at about 2x one thread. Such reads, and
+  `verify_provenance`'s uncached reader, now decode on the calling thread
+  (`clawhdf5_format::parallel_read::pool_can_parallelise`). The `File`'s
+  chunk cache, the suspect in `docs/known-issues.md`, was not the cause:
+  reads of datasets larger than its budget already skipped inserting, and
+  its lookups cost a few percent at 16 threads. Throughput with the default
+  pool is unchanged, and still short of an h5py process pool.
+
+### Contiguous read speed (2026-09-26)
+- **Large read buffers are backed by transparent huge pages.** A full read
+  of a contiguous dataset was one `memcpy` from the mapped file, yet ran at
+  a quarter of h5py's speed on one thread: the fresh output `Vec` took a
+  page fault (and a kernel page clear) for every 4 KiB page it was written
+  to, 16384 of them for 64 MiB, and those cost several times the copy.
+  numpy, and so h5py, asks for transparent huge pages on every allocation of
+  4 MiB or more; clawhdf5-format's read buffers now do too
+  (`madvise(MADV_HUGEPAGE)` on Linux, `libc` added as a Linux-only
+  dependency; a no-op elsewhere or when THP is disabled). It applies to the
+  typed readers' output (`read_f32`, `read_f64`, `read_i32`, `read_i64`,
+  `read_u64`, both byte orders), the raw contiguous read and the chunk
+  assembly buffer. Values are unchanged; new h5py comparison
+  `crates/clawhdf5/tests/contiguous_read_interop.rs` covers every 1-8-byte
+  integer and float type in both byte orders, ranks 1-4, and datasets past
+  the 4 MiB threshold.
+- **Hyperslab and point reads of contiguous data copy runs, not elements.**
+  A 256 x 256 hyperslab of a contiguous `f32` dataset read at an eighth of
+  h5py's speed: the selection's bounding box was copied out of the file,
+  then walked element by element (a recursive call and two bounds checks per
+  element) into a second buffer, which `read_f32_selection` converted into
+  a third. Selections of contiguous data are now copied straight from the
+  file, one `memcpy` per run of elements that is contiguous in the file
+  (a block along the last dimension, blocks that touch, and whole rows when
+  the inner dimensions are selected in full, merged), with no zero-filled
+  intermediate; a selection covering most of the dataset no longer makes a
+  full copy first. The typed selection readers (`read_f32_selection`,
+  `read_f64_selection`, `read_i32_selection`, `read_i64_selection`) copy
+  directly into their output when the dataset stores that type natively,
+  and convert as before otherwise (big-endian, other widths). The chunked
+  paths use the same run-based extraction. New public
+  `clawhdf5_format::data_read::read_selection_native` and the sealed
+  `NativeElement` trait (also used by the `read_as_*` fast paths, which
+  gained one for native `u64`). Values are unchanged: checked against h5py
+  by `contiguous_read_interop.rs` (strided, blocked, adjacent-block and
+  whole-row hyperslabs, points, empty selections; every type, both byte
+  orders, ranks 1-4).
+
+### Variable-length data (2026-09-26)
+- **VL values in files with 4-byte offsets** (`sizeof_addr = 4`). A VL
+  string attribute came back as `AttrValue::Raw`, a VL member of a compound
+  failed with `GlobalHeapObjectNotFound`, and VL datasets failed with a
+  size mismatch. Two causes: `Datatype::type_size()` reported 16 for every
+  VL type (the element is 4 + offset size + 4 bytes: 12 here), and the
+  global heap was parsed without the padding libhdf5 puts after its
+  collection and object headers (`H5HG_SIZEOF_HDR`/`H5HG_SIZEOF_OBJHDR`
+  round up to 8), so with 4-byte lengths every object was looked up 4
+  bytes early. `Datatype::VariableLength` now carries the element `size`
+  stored in the datatype message (**breaking** for code that builds or
+  exhaustively destructures that variant; patterns with `..` are
+  unaffected), and it is written back as stored. Tested against h5py
+  (`crates/clawhdf5/tests/vl_offset4_interop.rs`).
+- **Wrong data: VL strings with an embedded NUL, and VL elements whose heap
+  object has the wrong size.** libhdf5 hands VL strings over as C strings,
+  so h5py reads `"a\0b"` as `"a"`; `read_vl_strings` returned the NUL and
+  what followed. An element whose heap object is not exactly
+  `length × base size` bytes is refused by libhdf5 ("Expected global heap
+  object size does not match"); we returned the object cut or padded to
+  the length. Both now behave as libhdf5, and a heap address of 0 is a null
+  element (empty) whatever its length. The new
+  `clawhdf5_format::vl_data::VlResolver` does this and parses each global
+  heap collection once per read: `read_vl_strings` parsed the whole
+  collection again for every element. `vl_data::check_element_size` refuses
+  a VL datatype whose stored size is not 4 + offset size + 4 (libhdf5
+  ignores the stored size). The conformance probe resolves VL elements
+  with `VlResolver` too; conformance unchanged at 575 of 697.
+- **VL data through the facade.** VL-string datasets (h5py's default `str`
+  dtype) failed `read_string` with "type mismatch: expected String, got
+  VariableLength". `Dataset::read_string` now reads fixed- and
+  variable-length strings; new `read_string_bytes` (a VL string's exact
+  bytes, as h5py's `Dataset[()]` returns them), `read_string_selection`,
+  `read_vlen::<T>()` / `read_vlen_selection::<T>()` for VL sequences of
+  numbers (`T` = `f64`, `f32`, `i64`, `i32`, `u64`; converted like the
+  other typed readers), and `File::decode_strings` / `decode_string_bytes`
+  / `decode_vlen` for VL values in compound fields and `AttrValue::Raw`
+  attributes. `MmapDataset` and `LazyDataset` gain `read_string` for VL
+  strings, `read_string_bytes` and `read_vlen`. Checked against h5py with
+  8- and 4-byte offsets: scalar and 1-/2-D, ASCII and UTF-8, empty strings,
+  contiguous, compact, chunked with gzip/shuffle, never-written and
+  partly written chunks, hyperslab selections, VL members of compound
+  datasets and attributes (`crates/clawhdf5/tests/vl_data_interop.rs`).
+  NetCDF-4 `string` variables now read through
+  `clawhdf5_netcdf4::Variable::read_string` (checked against netCDF4-python
+  in `crates/clawhdf5-netcdf4/tests/interop_tests.rs`).
+
+- **Crafted global heaps could exhaust memory.** `VlResolver` kept an owned
+  copy of every object of every heap collection it read, so collections
+  nested inside each other's object data made a 744 KB file take 1.58 GB
+  (and `read_vl_strings` before it did the same). The cache now records
+  where objects lie instead of copying them, is dropped past a 32 MiB
+  budget, and a collection overlapping one already read is an error
+  (libhdf5 never writes one). New `GlobalHeapCollection::parse_index`
+  locates a collection's objects without copying them; `parse` and
+  `parse_index` refuse a collection running past the end of the file or an
+  object running past its collection. Conformance unchanged at 575 of 697
+  (`crates/clawhdf5-format/tests/vl_heap_bounds.rs`).
+
+- **Every reader resolves VL data the same way.** `h5rs` (`dump`, `ls`,
+  `diff`, `check --data`) had its own lenient VL decoder: a heap object
+  longer than the element's length was cut to it (h5py refuses it), a null
+  string printed `""` where h5dump prints `NULL`, the stored element size
+  was trusted, and each heap collection was kept as a copy for the whole
+  run. It now resolves through `VlResolver`, so `dump` matches h5dump byte
+  for byte on VL strings (`"a\0b"` as `"a"`, null as `NULL`), VL sequences
+  and 4-byte-offset files, `dump --json` gives h5py's values, and
+  `check --data` reports any heap object whose size is not exactly the
+  element's length × base size. `clawhdf5-wasm` already resolved VL strings
+  with `read_vl_strings`; it now uses `VlResolver` and refuses a VL type
+  whose stored element size disagrees with the file, as `File` does
+  (`crates/clawhdf5-tools/tests/h5rs_interop.rs`,
+  `crates/clawhdf5-wasm/tests/vl_strings.rs`). New
+  `VlResolver::element` / `string_element` resolve one element in place.
+
+- **A VL element at the undefined heap address is an error**, as in
+  libhdf5 ("addr undefined"). One of length 0 read as `""` in every reader
+  (`File`, `h5rs`, `clawhdf5-wasm`, `read_vl_strings`, `read_vl_bytes`).
+  libhdf5 writes a null element with heap address 0, which still reads as
+  empty, and h5py writes `""` as a zero-size heap object at a real address,
+  so no file libhdf5 or h5py writes is affected
+  (`a_vl_element_at_the_undefined_heap_address_fails_like_h5py` in
+  `crates/clawhdf5/tests/vl_data_interop.rs`). `read_vl_bytes` now also
+  treats address 0 as null whatever the length, as `VlResolver` does.
+
+### Writer: groups and links (2026-09-26)
+- **Nested groups, to any depth.** `FileWriter`/`FileBuilder` wrote the root
+  group plus one level, and refused path-like names. Now a name may be a path
+  (`create_dataset("a/b/x")`, `create_group("a/b")`, a leading `/` at the
+  root) and missing intermediate groups are created, as h5py does; groups
+  also nest through the new `GroupBuilder::create_group`/`add_group`. A group
+  added at a path that already holds a group is merged into it (h5py's
+  `require_group`); a name used twice otherwise, an empty or `"."`
+  component (`"a//b"`, `"a/"`) or an absolute path below the root is an
+  error. Datasets, attributes, dense attribute storage and dense link
+  storage work at every level.
+- **Soft, hard and external links at any depth:** `add_soft_link(name,
+  target)` (h5py's `SoftLink`; the target may dangle),
+  `add_hard_link(name, target)` (h5py's `f[name] = f[target]`; the target
+  path is resolved when the file is written, may go through other hard
+  links, and a missing target, a soft link on the way or a cycle of
+  hard-link paths is an error) and `add_external_link`, on `FileWriter`,
+  `FileBuilder` and `GroupBuilder`. An object with several hard links gets
+  an Object Reference Count message, so libhdf5 can delete one of the links
+  without freeing the object.
+- **Link creation order:** `track_order(true)` on a `GroupBuilder`, or on
+  `FileWriter`/`FileBuilder` for every group that does not set its own,
+  tracks and indexes link creation order (h5py's `track_order=True`): the
+  Link Info message carries the flags, each link its order, and a dense
+  group a creation-order B-tree (type 6). h5py then lists members in
+  insertion order. Attribute creation order is not tracked.
+- A group holds at most 65 535 links (its link index is one B-tree leaf),
+  and in a group of more than 8 links (dense storage) each link message
+  must be at most 65 515 bytes (one fractal heap block; huge heap objects
+  are not written); more is an error. Measured at the limit: 65 535 links
+  with 100-byte names (a 7 MB heap) read in h5py, h5dump and clawhdf5, and
+  h5py can add to the group. `GroupBuilder`'s fields changed (they were
+  crate-private); `FinishedGroup` is unchanged for callers.
+- Files that use one level of groups and no new link kinds are laid out as
+  before: byte-identical to the writer with the Group Info fix below
+  (compared on simple, mixed dense/chunked/compact/external-link and paged
+  files). Tests: h5py and clawhdf5 read the same
+  tree (every path, attribute and value) from a 5-level file; soft, hard,
+  external and cyclic hard links; 10 000, 20 000 and 65 535 links in one
+  group, with and without creation order; libhdf5 adding and deleting links
+  in our groups;
+  `h5rs check` passes and `h5rs dump` equals h5dump
+  (`crates/clawhdf5/tests/writer_groups_interop.rs`,
+  `crates/clawhdf5-tools/tests/h5rs_interop.rs`).
+- **Big dense groups and attribute sets were unreadable.** The fractal heap
+  holding dense links or attributes wrote every doubling-table row as
+  direct blocks, but past the 512 KiB the root's direct blocks hold, rows
+  are child indirect blocks, and libhdf5 and `h5rs check` read them as
+  such: a group with 20 000 links of 20-byte names was written without
+  error and h5py could not list it ("incorrect metadata checksum"); 150
+  dense attributes of up to 56 KB could not be opened. This was in 2.7.0
+  too. The heap writer now writes child indirect blocks, nested as deep as
+  needed. Found on the way: an object bigger than the next block's space
+  was cut off (it now goes in the first block big enough), and h5py adding
+  a link to a heap over 64 KiB overwrote its first block (the header's
+  next-block offset was 0).
+- **h5py crashed adding a link to a group of more than about 47 700
+  links** (35 000 with creation order tracked). The link index leaf's node size gave libhdf5 room for more than
+  65 535 records, which overflows the leaf's 2-byte count. The node is now
+  capped at 65 535 records. Dense attributes use the same index builder:
+  more than 65 535 on one object used to be written with the count modulo
+  65 536, and are now an error.
+- **A dense link or attribute message over 65 515 bytes** (e.g. a soft link
+  with a long target in a group of more than 8 links) was written cut off,
+  and libhdf5 could not list the group ("object overruns end of direct
+  block"). It is now an error.
+- **Chained hard links took exponential time to resolve.** A hard-link
+  target going through other hard links resolved them again on every path
+  through them: 26 links whose targets each named the previous one twice
+  took 46 s. Each hard link is now resolved once, and a cycle is reported
+  by the link's name.
+- **A dataset attribute set twice read back as its first value**, as for
+  groups below (h5py listed the name twice). The later value now replaces
+  the earlier one; a hand-set attribute named like a provenance attribute
+  is replaced by the computed one.
+- **A group attribute set twice read back as its first value.** Setting a
+  group (or root) attribute again wrote a second attribute message with the
+  same name, and h5py returned the first value. The later value now replaces
+  the earlier one, as `attrs[name] = v` does in h5py — also when a group is
+  merged from two builders.
+- **Non-ASCII link names were marked ASCII.** A group or dataset name such as
+  `größe` was written with the ASCII character set flag (h5py reported
+  `cset` 0 for it); it is now flagged UTF-8, as h5py writes it.
+- **libhdf5 could not add links to groups we wrote.** h5py in `"r+"` mode
+  failed with "Unable to create link (message type not found)" on every
+  group `FileWriter` wrote: libhdf5 reads a group's Group Info message before
+  inserting a link, and none was written. Every group now carries one
+  (version 0, default thresholds: 6 more bytes per group header, so files
+  are not byte-identical to earlier versions). Regression test:
+  `crates/clawhdf5/tests/writer_groups_interop.rs`.
+
+### Python bindings (2026-09-26)
+- **Panic: selections of v4 implicit-index chunked datasets** (pre-existing,
+  facade `Dataset::read_selection`, Rust callers too). A hyperslab whose
+  bounding box covered more than half of a chunked dataset with the implicit
+  index (`libver='latest'`, early allocation, no filters) panicked with
+  "index out of bounds" in `generate_implicit_chunks`: the fallback in
+  `data_read::read_raw_data_selection` passed the layout's chunk dimensions,
+  element-size dimension included, and then decoded the whole dataset
+  anyway. That arm now decodes and extracts directly, for every chunk index.
+  `crates/clawhdf5/tests/v4_chunk_index_selection.rs` reads small and large
+  hyperslabs of all five v4 indexes (single chunk, implicit, fixed array,
+  extensible array, B-tree v2) and compares them with h5py; it panicked
+  before. The Python bindings made this easy to reach (`ds[0:3]` on
+  libhdf5's `h5fc_ext*.h5` test files).
+- **`pip install` / `maturin develop` now gives `import clawhdf5`.** The
+  distribution in `crates/clawhdf5-py/pyproject.toml` was still called
+  `rustyhdf5` while the extension module was `clawhdf5`, and the package's
+  tests imported `rustyhdf5`, so they failed at collection. Distribution,
+  module and tests now all say `clawhdf5`, and the module has
+  `__version__`.
+- **h5py-style reads that read the selection, not the dataset.** `ds[...]`
+  used to read the whole dataset and slice it in numpy, and knew six
+  dtypes. Now
+  integers (negative from the end), slices with positive steps, `...`, one
+  increasing list of integers per key and compound field names map onto the
+  facade's hyperslab selection (a list is read one group of neighbouring
+  chunks at a time and picked from in memory), with h5py's results (numpy scalar for an all-integer
+  key, 0-d array for `scalar[...]`) and h5py's errors for everything else
+  (negative steps, `None`, boolean masks, out-of-range indices).
+  `Dataset.dtype` is the numpy dtype h5py reports, for every integer and
+  IEEE float width (incl. `float16`) in either byte order, `bool`, enums
+  (base integer with `metadata['enum']`), complex (`r`/`i` compounds),
+  fixed strings (`S<n>`), variable-length strings (`object` of `bytes`, as
+  h5py), variable-length sequences (`object` of arrays), opaque (`V<n>`),
+  HDF5 array types and compounds (numpy structured, offsets and padding
+  kept, nested). The bytes the library returns become the numpy array's
+  buffer without a copy. Types the mapping cannot describe exactly
+  (references, bitfields, time, non-IEEE floats, integers with padding
+  bits, variable-length members inside compounds) raise `TypeError` rather
+  than return guessed data. Attributes come back as h5py returns them
+  (numpy scalars and arrays with the stored dtype, `str` for
+  variable-length strings, `numpy.bytes_` for fixed ones — **a change**:
+  string attributes written by this package are fixed-length and used to
+  come back as `str` — and `clawhdf5.Empty` for a null dataspace, which
+  datasets return too). `Group`/`File` gain `get`, `values`, `items`,
+  iteration, `len`, `name`, absolute and relative paths (`g['/a/b']`,
+  `g['c/d']`, `f['/']`); `Dataset` gains `ndim`, `size`, `maxshape`,
+  `name`, `len()` and `numpy.asarray(ds)`. File access and decoding run
+  with the GIL released, so Python threads read in parallel.
+  `crates/clawhdf5-py/tests/test_read_vs_h5py.py` compares every read with
+  h5py 3.16 (HDF5 2.0) on a file h5py writes. One difference is h5py's:
+  it returns variable-length sequences of big-endian floats unswapped; this
+  package returns the stored values.
+- **A panic in the library is an ordinary Python exception.** PyO3 turns a
+  Rust panic into `PanicException`, a `BaseException` that `except
+  Exception` does not catch. Every call from the bindings into the library
+  is now guarded and a panic becomes `clawhdf5.InternalError` (a
+  `RuntimeError`) naming the object; with the implicit-index panic above
+  restored, `ds[0:30]` raises it.
+- **Wrong data: uninitialised padding in compound results of index lists.**
+  `ds[[0, 3, 6]]` joined one read per run with `np.concatenate`, which
+  copies structured dtypes field by field into an `np.empty` result, so the
+  padding bytes held whatever was in memory (pointers were seen) and leaked
+  through `tobytes()`, hashes and write-backs. The runs' bytes are now joined
+  in Rust, whole elements at a time, so the result carries the bytes read
+  from the file (h5py's, zero for files it wrote) and stays zero-copy.
+  The h5py comparisons now also compare every byte of structured values
+  (`test_compound_padding_bytes_match_h5py` and `assert_same`).
+- **Index lists no longer decode the same chunks once per run.** A list
+  index was one uncached hyperslab read per run of consecutive indices, so
+  on a chunked, compressed dataset every run decoded its chunk again:
+  `d[list(range(0, 200000, 40))]` over 20 gzip chunks took 8 s (h5py:
+  0.014 s). The list is now read in groups — for a chunked dataset a group
+  ends only where a whole chunk holds no selected index, so each chunk is
+  decoded once; otherwise at a gap of more than 64 KiB — and the selected
+  rows are picked from each group in Rust. The same read now takes 3.8 ms
+  (h5py 4.1 ms; release build on tank, best of 5).
+  `test_a_long_index_list_decodes_each_chunk_once` compares 1-D, 2-D and
+  contiguous cases with h5py under a 2 s bound (5.8 s before, debug build).
+- **Groups and datasets remember where they are.** Every `ds[...]`, and
+  every `g[k]`, resolved its path from the root again (two or three times
+  per open), and in a large group each resolution scans the group's links,
+  so visiting a group was quadratic: 4000 scalar datasets in one group took
+  39 s (`libver='earliest'`) and 131 s (`'latest'`) to list, read and
+  re-read in `test_big_groups_are_not_quadratic`; now 0.3 s each (debug
+  build). A `Dataset` keeps its object's address, and a `Group` (and the
+  file's root) its address and, once listed, its link table. New facade
+  API: `File::dataset_at(address)` opens a dataset without resolving a
+  path. libhdf5's `h5stat_newgrat.h5` (35001 members in the root): listing
+  takes 0.03 s and 2000 opens 1 ms (h5py: 0.022 s).
+- **`ds[np.array(1)]` is an integer index**, as in h5py; a 0-d integer
+  array went down the index-list path and raised a confusing `TypeError`.
+  The h5py comparison keys now include 0-d arrays on every axis.
+- **Tests that would notice a held GIL, and our extra errors.**
+  `test_reads_release_the_gil` times a Python thread spinning while another
+  reads: with the read made to hold the GIL it stalls for the whole read
+  (0.062 s of a 0.064 s read) and the test fails; released, its longest
+  stall is about 3 ms. (The existing threads test only checked values.)
+  `test_errors_match_h5py` now also requires that every key h5py reads
+  reads here too, with the same result, and covers more keys (0-d arrays,
+  repeated and empty lists, `()`, `...`).
+- **Docs say when a selection reads more than itself.** The README and
+  the package README said `ds[...]` reads only the selected elements,
+  without condition. The library decodes the whole dataset when the
+  selection's bounding box covers more than half of it, and for compact,
+  virtual, unwritten and non-default-fill chunked datasets; the READMEs,
+  the facade's `read_selection` docs and `docs/known-issues.md` now say so.
+- **CI builds and tests the Python package.** It was excluded from CI.
+  `scripts/ci-test.sh` now lints `clawhdf5-py`, builds the wheel with
+  maturin, unpacks it under `target/` and runs the pytest suite; skipped
+  without maturin/pytest in `$CLAWHDF5_PYTHON`, a failure then under
+  `CLAWHDF5_REQUIRE_INTEROP=1`. The CI interop venv installs both.
+
+### Plugin filters (2026-09-26)
+- **LZF, bitshuffle, bzip2 and Blosc read and write, in pure Rust.** Files
+  written by h5py with `compression="lzf"`, or with hdf5plugin's
+  `Bitshuffle`, `BZip2` and `Blosc`, failed with `UnsupportedFilter`. New
+  `clawhdf5-format`/`clawhdf5` features: `lzf` (32000, **on by default**, no
+  dependencies), `bitshuffle` (32008: transpose only, LZ4 and Zstandard
+  modes), `bzip2` (307), `blosc` (32001: Blosc 1 frames with BloscLZ,
+  LZ4/LZ4HC, Snappy, Zlib and Zstandard codecs and byte/bit shuffle;
+  BloscLZ is decoded by a port of c-blosc 1.21's decoder, and cannot be
+  written), and `plugin-filters` for all four. None compiles C: Zstandard is
+  ruzstd, bzip2 is libbz2-rs-sys. Write with `DatasetBuilder::with_lzf()`,
+  `with_bitshuffle(..)`, `with_bzip2(..)`, `with_blosc(..)` or
+  `with_plugin_filter(PluginFilter::..)`; `ChunkOptions` gains a `plugin`
+  field (**breaking** for code that builds `ChunkOptions` with a struct
+  literal and no `..Default::default()`). Tested both ways against h5py 3.16
+  + hdf5plugin 7.1 over 1-3-D shapes with partial edge chunks, 1-8-byte
+  types in both byte orders and incompressible data
+  (`crates/clawhdf5/tests/plugin_filters_interop.rs`). Conformance: 573 of
+  697 files ok (was 569) — h5ex_d_lzf/bshuf/bzip2/blosc.
+- **Filter registry.** Filters are looked up by ID in
+  `clawhdf5_format::filter_registry` instead of a `match`: the built-in
+  table (per build), then codecs registered at run time with
+  `register_filter(id, codec)` — a decoding closure or a `FilterCodec` that
+  can also encode. Built-in IDs cannot be overridden; a registered decoder's
+  output is held to the chunk-size bound. Unknown IDs still fail with
+  `UnsupportedFilter(id)`, whose message now names known filters and the
+  missing feature ("unsupported filter: 32026 (Blosc2, not implemented by
+  clawhdf5)").
+- **Not implemented:** Blosc2 (32026) and ZFP (32013) remain a clear error.
+- **Wrong data: a chunk that decodes short read as zeros** (pre-existing, every
+  filter). HDF5 stores every chunk at the full chunk size, so a filter
+  pipeline that decodes to fewer bytes means a corrupt chunk; every chunk
+  reader (full, cached, selection, parallel, partial) padded it with zeros.
+  It is now an error naming the chunk ("chunk at [16] decoded to 16 bytes,
+  expected 32"), via the new `filters::decompress_chunk_exact`. libhdf5
+  returns the rest of such a chunk uninitialised, or fails when the filter
+  checks. A Blosc frame declaring no data for a non-empty chunk is an error
+  too. Legitimate edge chunks are unaffected (they are stored full-size,
+  filtered or not); conformance is unchanged at 573 of 697, with no file
+  changing class.
+- **Crash: a hostile Blosc chunk panicked** in builds with overflow checks
+  (debug builds, `cargo test`, `maturin develop`): a frame size below the
+  16-byte header underflowed. It is now an error. Every new decoder (LZF,
+  bitshuffle, bzip2, Blosc/BloscLZ) is fuzzed with random and mutated frames
+  in the unit tests.
+- **`register_filter(32023, ..)` works with the `pcodec` feature.** 32023 is
+  Granular BitRound's ID; the built-in entry there only reads clawhdf5
+  <= 2.7.0's pcodec chunks (filter name `"pcodec"`), so a registered codec now
+  handles every other chunk with that ID, and writes. It was refused as
+  "built in".
+
 ### Upgrade Notes
+- **HDF5 correctness audit (2026-09-25).** A sweep of 686 public files (the
+  libhdf5 test files, the HDF Group's CVE reproducers, pyfive, netcdf-c,
+  netcdf4-python, h5wasm, h5py and xarray corpora), a 567-case read matrix and
+  a 96-case write matrix against HDF5 1.10–2.0 found bugs that returned wrong
+  values with no error, and files we wrote that libhdf5 rejects. The fixes are
+  listed under Correctness and Interop. What changes for callers:
+  - **Chunked datasets whose max shape is larger than their current shape**,
+    or whose unlimited dimension is not the first, were indexed by the current
+    shape instead of the max shape, both when read and when written. Files from
+    libhdf5 now read correctly. Files clawhdf5 wrote with such a max shape were
+    laid out wrongly and now read the way libhdf5 always read them — rewrite
+    them. Agent stores and ClawBrainHub files have no max shape and are
+    unaffected.
+  - Integer reads (`read_i32`/`read_i64`/`read_u64`/...) of float data now
+    convert (truncate toward zero, saturate at the type's range, NaN reads as
+    0) instead of returning the IEEE bit pattern, and out-of-range integers
+    saturate instead of keeping the low bits.
+  - `FileWriter::finish()` now returns an error instead of writing a corrupt
+    file for: a header message over 64 KiB (e.g. an attribute larger than
+    ~64 KiB), a group/dataset/link name that is empty, `.` or contains `/`
+    (nested paths were written as one literal link), a max shape smaller than
+    the shape, a page size outside 512 B–1 GiB, and more than 65 535 chunks in
+    a dataset with several unlimited dimensions.
+  - **Breaking (format crate):** `ObjectHeaderWriter::serialize`,
+    `BatchObjectHeaderWriter::compute_sizes`/`serialize_all` and
+    `build_chunked_data_from_precompressed` return `Result`;
+    `read_fixed_array_chunks`/`read_extensible_array_chunks` take `max_dims`;
+    `build_fixed_array_at`/`ea_writer::build_extensible_array_at` take one
+    `Option<WrittenChunk>` per index slot; `fill_value::dataset_fill_value`
+    returns `UnresolvedSharedMessage` for a shared message it cannot resolve
+    instead of `None`. `FillTime::default()` is `IfSet` (libhdf5's default;
+    default files are byte-identical).
+- **ZeroClaw does not use clawhdf5.** The project described itself as
+  ZeroClaw's memory backend ("imported as a `clawhdf5` Cargo feature"). Checked
+  against ZeroClaw v0.8.5 (the latest release), the `osobh/zeroclaw` fork and
+  their full history: no such feature or backend has ever existed. And
+  `clawhdf5-migrate`'s "ZeroClaw layout" (`memory_chunks`, `sessions`,
+  `entities`, `relations`) is not ZeroClaw's schema — ZeroClaw uses a single
+  `memories` table — so the migrator cannot read a ZeroClaw database. The
+  claims are withdrawn; the migrator's layout is documented as its own.
+- **OpenClaw is not supported, and never was.** The docs described a
+  "drop-in" OpenClaw memory backend enabled with `memory.backend = "clawhdf5"`.
+  That config was never valid in any OpenClaw release (v2026.2–v2026.7
+  accepted only `builtin`/`qmd` and rejected unknown keys, so a Gateway given
+  it refuses to start; OpenClaw 2.0 removed the key), no plugin was ever built,
+  and `@redclaw/clawhdf5` was never published. The integration docs
+  (`openclaw-integration.md`, `openclaw-config.md`, `migration-guide.md`) are
+  removed; `docs/openclaw.md` explains the status and what a real plugin would
+  need against OpenClaw v2026.9.6. `ClawhdfBackend` stays as a library API.
+- **Breaking:** `MemoryError` is now `#[non_exhaustive]` and gained
+  `SigningKeyRequired`; a `match` on it needs a wildcard arm. Future variants
+  will no longer be breaking.
 - **Breaking:** `clawhdf5-agent`'s `agent` feature is removed. It enabled
   nothing — the agent layer is always built — but the README and guides told
   people to pass it; drop `agent` from `features = [...]`.
@@ -60,6 +504,88 @@
   `quantized_index = false`, or pass `create --f32-index` to the CLI, to opt
   out. The CLI's `--quantized-index` is still accepted but is now a no-op.
 
+### Tools
+- New crate **`clawhdf5-tools`** with the binary **`h5rs`**: HDF5
+  command-line tools without libhdf5, built only on the `clawhdf5` facade
+  and `clawhdf5-format` (no C, so it also builds as a static musl binary).
+  - `h5rs ls [-r] [-v] FILE[/path]` lists objects like h5ls (its first two
+    columns are h5ls's text on the test files) plus the datatype; `-v` adds
+    address, link count, layout and chunk index, chunk size, storage,
+    filters, datatype and attributes.
+  - `h5rs dump [--json] [-A] [-p] [-d PATH] FILE` prints DDL text that is
+    byte-identical to h5dump 1.14.6's (and to Debian's 1.14.5, which CI
+    uses) on the test files (all layouts and
+    chunk indexes, v1/v2 groups, compound, enum, strings, links, named
+    types, attributes; null-padded strings show their NULs at any depth),
+    or JSON in the HDF Group's hdf5-json layout (schema in the crate
+    README). Nested compounds print inline and `long double` values as
+    errors (exit 1); both are listed in the README.
+  - `h5rs stat FILE` reports h5stat's object, link, rank, layout, filter,
+    attribute, raw-data and file-size figures (equal to h5stat's on the test
+    files); metadata space is one figure, not broken down.
+  - `h5rs diff [-r] [-q] [-n N] [-d D] [-p R] [--follow-symlinks] A B [OBJ1
+    [OBJ2]]` (option names as h5diff's: `-c` is `--compare`, the count is
+    `-n`/`--count=N`) compares objects, kinds, datatypes, shapes, attributes, values and link
+    targets; exit status 0/1/2 as h5diff's. Soft links are compared by
+    target path, as h5diff's default, or with `--follow-symlinks` by the
+    objects they lead to (external links are never followed). Every path is
+    compared, including every name of a hard-linked object and the members
+    of a hard-linked group; with a `-d`/`-p` tolerance, integers are
+    compared exactly in integer arithmetic (no loss above 2^53), and a `-p`
+    below the f64 epsilon compares exactly, as h5diff's. Objects that cannot
+    be compared count as a difference (h5diff exits 0 for them), and NaN
+    equals NaN.
+  - `h5rs check [--data] FILE` is a structural validator: it walks every
+    object, parses every header message, verifies the checksums of every
+    version 2+ structure it meets (superblock, object headers and
+    continuation chunks, v2 B-tree nodes, fractal heap headers and — which
+    the library's reads do not — every direct and indirect heap block, and
+    extensible/fixed array chunk indexes), checks each chunk index against
+    its dataset (aligned, in-extent, unique, plausibly sized chunks), and
+    that raw data lies inside the file without overlaps. Every problem is
+    printed with its address; exit 1 when there are any. libhdf5's h5check
+    reads only the 1.8 format. On the conformance corpus it passes all 418
+    files that both clawhdf5 and h5py read in full, and `check --data` flags
+    135 of the 150 CVE and fuzzer files of the `cve_hdf5` corpus (tank,
+    2026-09-26). `--data` also follows variable-length data into its global
+    heap collections and reports a damaged one at its address. It inherits
+    the library's tolerance, though: 8 of the 15 it passes are files h5dump
+    1.14.6 rejects (see `docs/known-issues.md`, header checks).
+  - Values over `--max-bytes` (default 1 GiB) are reported instead of read;
+    a panic is caught and reported as an internal error (exit 3).
+    `scripts/h5rs-fuzz.sh` runs every subcommand over a corpus (default the
+    CVE reproducers, optionally with byte-flipped copies) with overflow
+    checks, a timeout and a memory limit, and fails on any panic, crash or
+    hang; `scripts/h5rs-check-ok-files.sh` runs `check --data` over the
+    fully-read conformance files.
+  - Because the library does not verify fractal heap block checksums when
+    it reads a dense group's links or dense attributes, `h5rs` verifies a
+    heap's blocks before reading from it and refuses a damaged one, as
+    libhdf5 does, instead of printing what the damaged block holds.
+
+### Signing
+- `clawhdf5-agent`: **Ed25519-signed checkpoints** — the README's
+  "cryptographically verifiable memory", now true. With
+  `HDF5Memory::set_signing_key(key)`, every checkpoint stores a signed
+  manifest: a SHA-256 per record (text, embedding as stored, channel,
+  timestamp, session, tags, deleted flag, activation) in a Merkle tree, plus
+  hashes of the settings (and WAL mark), sessions and knowledge graph, with
+  the per-record hashes in `/integrity/record_hashes`.
+  `HDF5Memory::verify(path, &public_key)` recomputes everything from the file
+  and reports which part changed and which records (`changed_records`); a
+  forged manifest fails the signature. The key is never persisted; a signed
+  store refuses to checkpoint without it (`MemoryError::SigningKeyRequired`),
+  and `remove_signature()` is the deliberate way back to unsigned. Saves still
+  in the WAL are not covered (`wal_entries_unsigned`). Tests include every
+  kind of edit, and an edit made with h5py in place, which verify pinpoints.
+  Cost: ~20% of a checkpoint, 32 bytes per record (`BENCHMARKS.md`, "Signed
+  checkpoints"). New dependencies `ed25519-dalek`, `sha2`, `rand_core` — pure
+  Rust; the no-C check still passes.
+- `clawhdf5-cli`: `keygen --out <file>` (owner-only key file),
+  `--signing-key <file>` / `CLAWHDF5_SIGNING_KEY` on writing commands
+  (`create` signs immediately), `verify --public-key <hex|file>` (JSON report;
+  exit status 2 if not valid), and `signed` in `create`/`stats` output.
+
 ### Migration
 - `clawhdf5-migrate`: writes through the agent's own API (`HDF5Memory::create`
   / `open`, `save_batch`, the session cache and knowledge graph), so there is
@@ -100,6 +626,13 @@
   activation of the `k` results it returns, not of the whole `3k` candidate
   pool it re-ranks.
 
+### Documentation
+- OpenClaw claims withdrawn across the README, QUICKSTART, USE_CASES, ROADMAP
+  (Track 7 marked withdrawn) and the `openclaw` module docs; the dead
+  `github.com/redclawsystems/openclaw` link is gone. The Node package is
+  marked unpublished and broken (now `"private": true` so it cannot be
+  published by accident), with its bugs recorded in `docs/known-issues.md`.
+
 ### Benchmarks
 - Every undated or pre-September section of `BENCHMARKS.md` re-run on one
   machine on one day (tank, 2026-09-24, commit 5c8323c), with the command for
@@ -113,8 +646,39 @@
   README claimed but nothing measured.
 - `footprint_bench` reports whether it built `float16` or `f32` stores and
   takes `--f32`; it had kept printing "f32" after the default changed.
+- New `concurrent_read` harness, with an h5py counterpart
+  (`crates/clawhdf5-bench/scripts/concurrent_read_h5py.py`, threads or
+  processes) and `compare_concurrent_read.py`: decoded read throughput and
+  scaling efficiency at 1-16 threads on one open file, full reads of distinct
+  datasets and random hyperslabs of one dataset, deflate and contiguous, warm
+  or `--cold` page cache, JSON output. Not yet measured — `BENCHMARKS.md`
+  ("Concurrent reads") has the commands and no numbers.
 
 ### Interop
+- **h5py could not open chunked datasets we wrote with a chunk dimension
+  from 65 536 to 16 777 215.** A version-4 layout must store its chunk
+  dimensions in the fewest bytes that hold the largest (3 for 70 000);
+  the writer rounded 3 up to 4, and HDF5 2.0.0 (h5py 3.16) refuses that
+  ("stored chunk dimension encoding length does not match value calculated
+  from chunk dimensions"). Newer libhdf5 and clawhdf5 read those files; new
+  files use the exact width. Test: `we_write_chunk_dimensions_in_the_fewest_bytes`.
+- **Conformance sweep in the repo** (`conformance/`, report in
+  `CONFORMANCE.md`). `conformance/run.sh` fetches eight public HDF5 corpora
+  pinned by commit (libhdf5's test files, the HDF Group's CVE reproducers,
+  pyfive, netcdf-c, netcdf4-python, h5wasm, h5py, xarray-data) into a
+  gitignored cache, reads every file with clawhdf5 and with h5py/libhdf5 (and
+  the CVE files with h5dump) under a timeout and memory limit, compares them
+  object by object and regenerates the report — about 30 s once the corpus is
+  cached. A nightly Gitea job (`.gitea/workflows/conformance.yml`) runs it and
+  fails on any panic, hang, crash or out-of-memory, or when a file in
+  `conformance/baseline.json` stops reading identically. First report, on
+  42b81d9: 467 of 697 files identical to h5py, 123 our-error, 15 mismatch
+  (2 of them an h5py bug), 92 that libhdf5 cannot read, no panics, hangs or
+  crashes. Compared with the ad-hoc audit sweep, the probe now compares
+  N-Bit floats (and integers with a bit offset) as the values libhdf5
+  converts them to rather than raw file bytes — 8 files that were reported as
+  mismatches read identically — and the reference side no longer flips
+  between runs when libhdf5 aborts while freeing h5py objects.
 - `clawhdf5-format`: **every `f32` dataset was unreadable by h5py and
   libhdf5.** The float datatype encoder hard-coded the sign bit's position to
   63, correct only for `f64`; libhdf5 validates it and refused the dataset. It
@@ -129,6 +693,37 @@
   `float16` rounding matches numpy's bit for bit on 4 020 probe values,
   including ties, subnormals and the overflow boundary), and an agent store —
   `f32` and `float16` — opened by h5py with every dataset decoded.
+- `clawhdf5-format` filters, checked against libhdf5 + hdf5plugin:
+  - **LZ4 (32004) now uses the registered HDF5 LZ4 format** (8-byte BE size,
+    4-byte BE block size, BE-length-prefixed blocks). Our old framing (4-byte
+    LE size + one block) was readable only by clawhdf5, and we could not read
+    libhdf5's (`h5ex_d_lz4.h5`). Old clawhdf5 LZ4 chunks still read; they are
+    told apart unambiguously (a registered chunk starts with four zero bytes).
+  - **Zstd (32015) frames now record the content size**, which libhdf5's zstd
+    plugin needs; h5py could not read our zstd datasets.
+  - **Pcodec moved from filter ID 32023 to 480.** 32023 is registered to
+    Granular BitRound, whose decode is a pass-through — libhdf5 with that
+    plugin would have returned compressed bytes as data. Pcodec has no
+    registered ID; 480 is in the registry's private range (256–511) and only
+    clawhdf5 can read it. Chunks written under 32023 with the filter name
+    `pcodec` (clawhdf5 ≤ 2.7.0) still read.
+  - **SZIP decode matches libhdf5.** It returned garbage or zeros with no
+    error for libhdf5-written files (the 4-byte size prefix, 32/64-bit
+    byte-plane interleaving, reference interval, scanline padding and byte
+    order were all handled wrongly) and rejected 64-bit data.
+  - N-Bit honours libhdf5's "need not compress" flag (multi-filter pipelines
+    such as `tfilters.h5` failed) and reads enum/no-op members.
+  - Scale-offset `float` decode uses libhdf5's single-precision arithmetic
+    (was 1 ULP off for some values).
+  - A pipeline with Fletcher32 ahead of the compressor (h5py
+    `set_fletcher32()` then `set_deflate()`) no longer fails with "deflate:
+    output exceeds size limit".
+- `clawhdf5-format`: **HDF5 1.4/1.6-era files are readable.** Data Layout
+  message versions 1 and 2 (compact, contiguous, and chunked through the
+  version-1 B-tree) failed with `InvalidLayoutVersion` — 84 of the 686 files in
+  the 2026-09-25 audit sweep, 205 datasets. They now read as libhdf5 does;
+  checked byte for byte against h5py on HDF5's own test files
+  (`tests/legacy_format_interop.rs`).
 
 ### Storage
 - `clawhdf5-format`: **half-precision datasets.**
@@ -150,6 +745,32 @@
   infinity; batches are all or nothing. CLI: `create --float16`. See
   `BENCHMARKS.md`, "float16 embedding storage".
 
+### Browser (WebAssembly)
+- **New crate `clawhdf5-wasm`:** the reader compiled to
+  `wasm32-unknown-unknown` with a wasm-bindgen JavaScript API —
+  `open(bytes)`, `list`, `info`, `attrs`, `read`, `readHyperslab` — returning
+  typed arrays of the stored width (`BigInt64Array` for 64-bit integers),
+  string arrays for strings and enums, and a thrown `Error` for types with no
+  typed-array form (compound, reference, opaque, VL sequences) or filters the
+  build lacks (Zstd, SZIP). Read-only; the file is held in memory.
+- **`examples/wasm-viewer/`:** a drop-a-file HDF5/NetCDF-4 viewer page (tree,
+  type/shape/attributes, values paged as hyperslabs; `?file=&path=` opens a
+  URL). `build.sh` produces the package; `test/run.sh` checks it under Node
+  (251 checks against values h5py/libhdf5 read back from an h5py- and a
+  netCDF4-written file) and renders the page in headless Chromium. Size,
+  measured 2026-09-26 on tank (`gzip -9 -n`): 627,501 B of wasm, 191,639 B
+  gzipped, plus 21,826 B (4,487 B) of JS glue; h5wasm 0.10.3's embedded wasm
+  is 3,544,184 B (907,096 B) — full libhdf5, so not equal functionality. See
+  `examples/wasm-viewer/README.md`.
+- The facade's read path already built for `wasm32-unknown-unknown` (nothing
+  needed gating); `ci-test.sh` now builds it (`--no-default-features`) and
+  lints `clawhdf5-wasm` for that target, and CI installs the target. The Node
+  and browser tests run in `ci-test.sh` only where `node` and `wasm-bindgen`
+  exist (not the CI container); CI checks the same expectations natively
+  (`clawhdf5-wasm`'s `h5py_interop` test).
+- `Dataset::raw_datatype()` (facade) returns the full stored datatype, for
+  decoding `read_selection` bytes with `clawhdf5_format::data_read`.
+
 ### Build
 - **Pure-Rust default.** `clawhdf5-format`, `clawhdf5-filters` and the
   `clawhdf5` facade default to the `zlib-rs` deflate backend; `fast-deflate`
@@ -166,6 +787,281 @@
 - CI keeps zlib-ng building and tested; the arm64 job no longer needs cmake.
 
 ### Correctness
+- **Corrupt files libhdf5 refuses are now refused instead of read.** On the
+  HDF Group's CVE reproducers, 18 objects that libhdf5 (HDF5 2.0, through
+  h5py) refuses to open were read by clawhdf5, some as wrong data (a chunk
+  dimension of 0 read as all fill values; chunks read at offsets off the
+  chunk grid). The
+  parser now makes libhdf5's checks, with libhdf5's error text:
+  - object headers (`FormatError::InvalidObjectHeader`): every message of a
+    v1 chunk is read and more than the prefix's count is refused (the rest
+    used to be dropped); v1 message sizes must be multiples of 8 and a v1
+    chunk cannot end in a gap; a message running past its chunk is an error
+    (it used to end the chunk quietly); contradictory message flags; a
+    message of a class that cannot be shared flagged shareable; a
+    reference-count message in a v1 header; malformed continuation,
+    reference-count and modification-time messages; unknown v2 header
+    flags.
+  - datatypes (`FormatError::InvalidDatatype`): size 0; integer bits outside
+    the type; float exponent/mantissa outside the type, empty or
+    overlapping; a compound with no members, a member outside the compound,
+    a duplicate name or overlapping members; an enum whose size differs from
+    its base type's or with an empty name; array rank over 32 or a zero
+    dimension; an opaque tag length that is not a multiple of 8; in a
+    version-1 (unchecksummed) header, a numeric type that leaves more than
+    half its bits unused (`Datatype::parse_in_header`,
+    `Datatype::check_unused_bits`). A v1/v2 float's class bit 6 was read as
+    VAX byte order; libhdf5 ignores it before version 3, and so does this.
+    The overlap check measures each earlier member by its stored size, as
+    libhdf5 does, so a variable-length member (4 + offset size + 4 bytes)
+    in a file with 4-byte offsets does not overlap the member after it.
+  - chunked layouts (`FormatError::InvalidChunkDimensions`): a zero chunk
+    dimension, a chunk rank that does not match the dataspace, a chunk of
+    4 GiB or more indexed by a v1 B-tree (layout version 3 or earlier;
+    0x80000000-sized chunks hung the reader — layout versions 4 and 5 allow
+    larger chunks, and HDF5 2.0 writes them), an element size in the
+    layout that differs from the datatype's stored size (the chunks were
+    laid out with the wrong element size), and v1 B-tree
+    chunk keys whose offsets are not multiples of the chunk dimensions,
+    including the keys that only bound a node
+    (`chunked_read::collect_chunk_info_checked`).
+  - truncated files (`FormatError::TruncatedFile`, `Superblock::data_end`):
+    a file shorter than the end of file its superblock records is refused
+    ("truncated file"), and nothing past that end is read. Every reader
+    does this: `File`, `LazyFile` and `MmapFile`, and in `clawhdf5-io`
+    `NativeVol` (at `open`, and on read for `from_bytes`),
+    `AsyncHDF5File` and `MpiVol` (the MPI path is not built in CI: it
+    needs an MPI installation).
+  - the writer: `FileWriter::finish()` / `FileBuilder::finish()` refuse a
+    datatype the reader would refuse (`FormatError::SerializationError`,
+    "datatype cannot be written: ..."), such as a compound with a repeated
+    field name or no fields, or an enum member with an empty name
+    (`CompoundTypeBuilder` and `EnumTypeBuilder` build them without
+    complaint). These were never valid HDF5 — h5py refuses them — and
+    clawhdf5 wrote them, which made files it could not read back.
+
+  Checks newer libhdf5 releases make but HDF5 2.0 does not (bit-field
+  offsets, the variable-length kind, array sizes) are left out, so files
+  h5py opens still open. Two libhdf5 checks are skipped on purpose because
+  clawhdf5 up to v2.7.0 wrote files that fail them without being wrong:
+  the sign bit of every float at position 63, and a size-0 string type for
+  an empty-string attribute (new fixtures written by v2.7.0 guard this).
+  Conformance: 569 -> 571 ok (h5stat_err_refcount.h5,
+  h5clear_fsm_persist_less.h5), and 17 of the 18 CVE objects now fail as in
+  libhdf5 (see `docs/known-issues.md` for the one left), as do 10 files
+  h5py refuses as truncated. Tests:
+  `header_validation_interop.rs` (h5py writes, the test damages a copy, both
+  libraries must refuse it), `legacy_writer_files.rs`, and unit tests next
+  to each check. **Breaking (format crate):** `FormatError` gained
+  `InvalidObjectHeader`, `InvalidDatatype`, `InvalidChunkDimensions` and
+  `TruncatedFile`; an exhaustive `match` on it needs the new arms.
+- **Chunked datasets whose chunk dimensions take 3, 5, 6 or 7 bytes did not
+  open.** A version-4 layout (`libver="latest"`) stores each chunk dimension
+  in the fewest bytes that hold the largest one, so a chunk dimension from
+  65 536 to 16 777 215 (e.g. h5py `chunks=(70000,)`) takes 3 bytes; only 1, 2,
+  4 and 8 were read, and the rest failed with `UnexpectedEof`. Widths 1-8 are
+  read now, and 0 or more than 8 is refused as libhdf5 refuses it. A width
+  larger than needed is accepted: HDF5 2.0.0 refuses one ("stored chunk
+  dimension encoding length does not match"), but libhdf5 since
+  HDFGroup/hdf5@e124c36 (2026-06-05) reads it, and clawhdf5 itself wrote such
+  layouts.
+- `clawhdf5-format` VDS: variable-length and reference data from a source in
+  another file is refused. Those elements are global-heap IDs and object
+  addresses in the source file; copied into the virtual dataset they would
+  be decoded against the wrong file and name another object.
+- `clawhdf5-agent`: a store whose `/meta` has an attribute that cannot be
+  decoded fails to open (`MemoryError::Schema`). With `attrs()` now leaving
+  unreadable attributes out, it would otherwise have opened with defaults in
+  place of its settings (`float16`, `compression`, the WAL mark, ...).
+- `clawhdf5-format` reader: an old-style group whose local heap has a free
+  list pointing outside the heap was listed with names read from the broken
+  heap (garbage names on `cve-2021-36977.h5` once its user block was
+  applied). libhdf5 refuses such a heap ("bad heap free list"); so do we now,
+  with `FormatError::InvalidLocalHeapFreeList`. As in libhdf5 the free list
+  is checked when the first name is read (`LocalHeap::validate_free_list`,
+  new), so an empty group with a damaged heap still lists as empty.
+- **Files with a user block** (`h5py.File(..., userblock_size=N)`, `h5jam`;
+  the superblock at 512, 1024, …) could not be read: every address in the
+  file is relative to the superblock, but it was applied from byte 0
+  (`InvalidObjectHeaderVersion` on the root group). `File` (mmap, buffered,
+  `from_bytes`), `MmapFile`, `LazyFile`, `AsyncHDF5File`, the VOL readers,
+  the HNSW loader and external VDS sources now view the file from the
+  superblock on, using the signature's position as the base address as
+  libhdf5 does; `user_block_size()` reports the user block (h5py's
+  `userblock_size`), and `as_bytes()` returns the bytes from the superblock
+  on. **Breaking (format crate):** `Superblock::parse` refuses a non-zero
+  signature offset with `FormatError::UserBlockNotStripped`, since the
+  addresses it returns would be applied to the wrong bytes; pass the slice
+  from `signature::split_user_block` (new) and parse at offset 0.
+- `clawhdf5-format` reader: version-1 shared messages (HDF5 1.6-era files,
+  e.g. a dataset using a committed datatype in libhdf5's `tcompound.h5`)
+  read the heap-offset field of the embedded symbol-table entry as the
+  target address and failed with `InvalidObjectHeaderVersion`. The address
+  is now read after it, as libhdf5 does. **Breaking (format crate):**
+  `shared_message::parse_shared_ref` takes `length_size`. A reference whose
+  target header has no message of the referenced type is now
+  `FormatError::SharedMessageTargetMissing` instead of returning the first
+  other message found there (which decoded as garbage).
+- `clawhdf5-format` reader: array members of version-1 compound datatypes
+  (HDF5 1.6-era files, e.g. libhdf5's `tcompound.h5`) were read as a single
+  element: a `[4] i32` member came back as one `i32`, with the wrong size.
+  The legacy per-member dimension fields are now decoded into an array type,
+  as libhdf5 does; more than four dimensions, or a zero-sized one, is an
+  error.
+- `clawhdf5-format` virtual datasets (VDS), checked against HDF5 2.0 through
+  h5py (`crates/clawhdf5/tests/vds_interop.rs`):
+  - **Wrong data:** elements no mapping supplies — unmapped regions, and
+    mappings whose source file or dataset is missing — read as 0 instead of
+    the virtual dataset's fill value (e.g. h5py `fillvalue=-1`). Assembly moved
+    to the new `vds` module: `vds::read_virtual_dataset` takes the fill value
+    and a resolver that can refuse a name (`VdsFileResolver`), and `File`
+    passes the dataset's fill value. A missing source *dataset* read as an
+    error; it is fill now, as in libhdf5. Source datasets are read with their
+    own fill value for unallocated chunks, and a source whose datatype differs
+    from the virtual dataset's is an error (libhdf5 converts; we do not).
+    `File` now refuses a source name that leaves the virtual file's directory
+    (`../x.h5`, absolute paths), or any external source of a `File::from_bytes`
+    file, with an error — these used to read as fill.
+    **Behaviour change:** the raw-read API (`read_raw_data_full*`), which has
+    no fill value, now returns an error for a virtual dataset with unmapped
+    elements instead of zeros.
+  - Unlimited and printf-style mappings are supported (all 7 VDS files in the
+    libhdf5 test set are such mappings, e.g. Eiger/Percival detector layouts).
+    `%b` in a source file or dataset name is the block number and `%%` a
+    literal `%` (other `%` sequences are an error, as in libhdf5); block `j`
+    is read from the source named with `j`, probing from 0 up to the first
+    missing source. Unlimited source/virtual selections cover as much as the
+    source's current extent fills, including a partial last block. As
+    libhdf5 does on `H5Dget_space`, the extent is recomputed from the sources
+    present (default "last available" view, printf gap 0) —
+    `vds::virtual_dataset_extent`, used by `Dataset::shape()` — so e.g.
+    `vds-eiger.h5` is `[5, 10, 10]`, not its stored `[20, 10, 10]`. A source
+    stored in the other byte order is byte-swapped (libhdf5 converts);
+    other type conversions remain an error.
+  - Hyperslab selection versions 1 and 2 were refused ("only version-3
+    hyperslab selections are supported"). Version 1 is what libhdf5 writes for
+    every VDS created with the default format bounds (h5py's default), so
+    those could not be read at all; version 2 is its encoding of an unlimited
+    selection. Both are decoded now, as are irregular hyperslabs (a union of
+    blocks, read in row-major order as libhdf5 iterates them).
+    `SerializedSelection` exposes the raw form, including unlimited counts.
+  - The version-1 mapping list HDF5 2.0 writes (low version bound 2.0) was
+    misparsed: each entry's flags byte was read as the start of the source
+    file name, and names shared with an earlier entry (stored as that entry's
+    index) were not followed. Now decoded as `H5D__virtual_load_layout` does.
+- `clawhdf5-format` reader — **values returned wrong with no error:**
+  - Fixed Array and Extensible Array chunk indexes were laid out by the
+    dataset's current shape instead of its max shape (23 libhdf5 test files,
+    and any h5py file with e.g. `maxshape=(10, None)` or `(20, 10)` under
+    `libver='latest'`).
+  - Files with 4-byte offsets: unfiltered chunked datasets read as zeros.
+    Chunk B-tree keys store offsets in 8 bytes whatever the file's offset
+    size.
+  - A chunk's filter mask skipped the whole pipeline when any bit was set;
+    only the flagged filters are skipped now.
+  - Float data read as an integer returned the bit pattern; narrowing integer
+    reads kept the low bits; bfloat16 was decoded as IEEE half. Floats are now
+    decoded from their datatype fields (bf16, FP8 E4M3/E5M2, IEEE half, single
+    and double).
+  - `vl_data::read_vl_bytes` truncated sequences of non-byte base types.
+  - A shared fill-value message read as zero fill; it is resolved now,
+    including from the file's shared-message (SOHM) table, which could never
+    resolve because its index version byte was skipped.
+  - Two threads reading two chunked datasets through one `File` could get each
+    other's chunks (the shared chunk cache was switched between datasets
+    across separate lock acquisitions). The cache is now keyed by dataset.
+  - Compound datatype version 1 members with legacy array dimensions (HDF5
+    before 1.4, which had no array class) were read as a single scalar at
+    the member's offset; they are now array members, as in libhdf5
+    (`tarrold.h5`, `tcompound.h5`). Only reachable once layout versions 1/2
+    were readable, since the files that use it are that old.
+- `clawhdf5-format` reader — errors on valid files: a version-1 shared
+  message (a committed datatype in HDF5 1.4/1.6-era files) was read as if the
+  object header address followed the reserved bytes; it follows a link-name
+  offset (the reference is an old-style symbol table entry), so the reader
+  followed the name offset and failed with `InvalidObjectHeaderVersion`
+  (`tcompound.h5`). New `shared_message::parse_shared_ref_sized` takes the
+  superblock's length size; `parse_shared_ref` assumes it equals the offset
+  size.
+- `clawhdf5-format` reader — errors on valid files: enum and bool datasets
+  through the numeric readers; the "don't filter partial edge chunks" layout
+  flag; Fletcher32 ahead of deflate (NetCDF-4's order). Unknown-message flags
+  follow libhdf5 (`tbogus.h5`): "fail if unknown" is refused, "fail if unknown
+  and writing" is ignored by a reader.
+- `clawhdf5-format` reader — dense groups and attributes (links or
+  attributes kept in a fractal heap indexed by a v2 B-tree):
+  - A link heap larger than the root indirect block's direct rows (512 KiB
+    with libhdf5's defaults: a few thousand long link names, or ~20 000 short
+    ones) could not be listed: child indirect blocks were given the wrong
+    number of rows, so every link stored in one was unreachable.
+  - v2 B-trees of depth 3 or more (a dense group of ~22 000+ links) were
+    misparsed: internal-node child pointers were read with widths from an
+    estimate instead of libhdf5's per-depth record capacities, and the
+    listing failed. The same B-tree code indexes dense attributes, shared
+    messages and chunks.
+  - Fractal-heap "huge" objects (larger than the heap's managed-object
+    limit, 4 KiB by default — e.g. an 8 KiB dense attribute or a link with a
+    very long name) and "tiny" objects are now read; the ID type was taken
+    from the wrong bits (6-7, the version, instead of 4-5), so a huge object
+    failed and took every attribute on its object down with it (NetCDF-4
+    files such as netcdf4-python's `issue671.nc`). Huge objects are found
+    directly from the ID or through the huge-object v2 B-tree, filtered or
+    not.
+  - Heaps with an I/O filter pipeline (a group created with a filter on its
+    creation property list compresses its link heap) are now read: the
+    header's pipeline was skipped with the wrong size, so its checksum was
+    looked for in the wrong place, and filtered direct blocks were read raw.
+  - A user-defined link (link class 65-255, e.g. 187 in libhdf5's
+    `tall.h5`/`tudlink.h5`) made its whole group unlistable. Such links
+    cannot be followed without the application that registered the class, so
+    they are now left out of `datasets()`/`groups()` and path lookup, as h5py
+    leaves out links it cannot open; reserved link types are still an error.
+- `clawhdf5` — soft links are listed, as h5py lists them: `datasets()` and
+  `groups()` on `Group`/`MmapGroup`/`LazyGroup` include each soft link under
+  its own name as the kind of object it resolves to, and `dataset(name)` /
+  `group(name)` open through it. Relative targets resolve from the group
+  holding the link. Dangling or cyclic soft links, external links and
+  user-defined links are left out (h5py lists their names but cannot open
+  them). Previously soft links were missing from the listings, and in
+  old-style (symbol table) groups a soft link made the listing fail. New
+  `group_v2::resolve_group_children` / `resolve_path_from` and
+  `group_v1::v1_soft_links` in `clawhdf5-format`.
+- `clawhdf5` — one unreadable attribute no longer fails `attrs()` for every
+  attribute on its object: it is left out of the map, and the new
+  `attrs_with_errors()` (on every group and dataset handle) returns the map
+  plus one error per attribute left out. Returned values are always complete.
+  An error in the attribute index itself (attribute info message, dense heap
+  header or B-tree) still fails the call. `clawhdf5-format` gains
+  `attribute::extract_attributes_tolerant`; `extract_attributes_full` stays
+  strict.
+- `clawhdf5-format` reader — files with shared object header messages
+  (SOHM, `H5Pset_shared_mesg_index`): a datatype, dataspace, filter pipeline
+  or attribute stored in the file's SOHM heap failed with "invalid shared
+  message version: 2" — only shared fill values loaded the SOHM table — so
+  such files' datasets and attributes could not be read.
+  `shared_message::resolve_shared_message` now loads the table when a
+  reference needs it (36 cases of the audit's read matrix).
+- `clawhdf5-format` writer — **files libhdf5 rejects or reads wrong:**
+  - Extensible Array (one unlimited dimension): chunks from index 244 on were
+    written but never indexed and read as 0, by libhdf5 and by us.
+  - Fixed Array: more than 1 024 chunks gave checksum errors (data blocks
+    were never paged).
+  - A finite max shape larger than the shape gave libhdf5 "addr overflow"; an
+    unlimited dimension that is not the first scrambled the data; several
+    unlimited dimensions (`(None, None)`) broke the whole file. These now
+    write the index libhdf5 writes (swizzled Extensible Array, or a B-tree v2
+    index for several unlimited dimensions).
+  - Header messages over 64 KiB (the size field is 16 bits) and compact
+    datasets at 65 534–65 535 bytes produced corrupt files.
+  - Reference, Opaque, BitField and Time datatypes were written as empty
+    messages; they now encode as HDF5 2.0 does.
+  - `with_page_size` wrote a nonexistent superblock version 4; it now writes
+    the v3 superblock and File Space Info message libhdf5 writes.
+  - `FillTime` values were rotated on disk (NEVER was written as ALLOC, and so
+    on). New `DatasetBuilder::with_fill_value`.
+  - An empty-string attribute got a zero-size datatype, which made every
+    attribute on the object unreadable in libhdf5.
+  - `maxshape` equal to the shape no longer forces chunked layout.
 - `clawhdf5-format`: **a truncated deflate chunk read back short, with no
   error.** The deflate filter used flate2's streaming reader, which returns the
   bytes it has when the input runs out before the end-of-stream marker. It now

CLAUDE.md

@@ -1,29 +1,31 @@
 # clawhdf5
 
 ## Purpose
-Pure-Rust HDF5 format implementation with HNSW vector search, WAL-backed persistence, agent memory storage, and GPU-accelerated I/O. Used by ZeroClaw as its persistent memory and knowledge graph backend.
+Pure-Rust HDF5 format implementation with HNSW vector search, WAL-backed persistence, agent memory storage, and GPU-accelerated vector search. A standalone library. Its one verified consumer is ClawBrainHub (`.brain` files); no agent framework integrates it (OpenClaw and ZeroClaw claims were withdrawn on 2026-09-25 — neither was ever true).
 
 ## Architecture
 
-Cargo workspace with 16 crates under `crates/` (plus `libaec-sys`, an internal FFI bindings crate for the optional `szip` feature):
+Cargo workspace with 18 crates under `crates/` (plus `libaec-sys`, an internal FFI bindings crate for the optional `szip` feature):
 
 | Crate | Role |
 |-------|------|
 | `clawhdf5-format` | HDF5 binary spec parser (superblock, B-tree, heap) — also holds shared type definitions and physical constants |
 | `clawhdf5-io` | Read/write implementation |
-| `clawhdf5-filters` | Compression filters (gzip, LZ4, Zstd, Blosc) |
+| `clawhdf5-filters` | Deflate backends (zlib-rs, zlib-ng, Apple Compression); the HDF5 filter pipeline, the filter registry (`clawhdf5_format::filter_registry`) and the other codecs (LZ4, Zstd, SZIP, N-Bit, scale-offset, pcodec, and the pure-Rust plugin filters LZF, bitshuffle, bzip2, Blosc 1) live in `clawhdf5-format`. No Blosc2 or ZFP. |
 | `clawhdf5-derive` | Proc-macro derive for HDF5-serializable structs |
 | `clawhdf5` | Main facade crate |
 | `clawhdf5-netcdf4` | NetCDF-4 compatibility layer |
 | `clawhdf5-ann` | HNSW approximate nearest-neighbor vector index |
 | `clawhdf5-agent` | Agent memory, session history, knowledge graph storage |
-| `clawhdf5-gpu` | GPU-accelerated I/O via wgpu (hand-written WGSL compute shaders) |
+| `clawhdf5-gpu` | GPU vector distance computation via wgpu (hand-written WGSL compute shaders) — not dataset I/O |
 | `clawhdf5-accel` | CPU SIMD acceleration path |
 | `clawhdf5-migrate` | SQLite → HDF5 agent-memory migration |
 | `clawhdf5-android` | Android JNI bindings |
-| `clawhdf5-cli` | Command-line interface |
+| `clawhdf5-cli` | Command-line interface (agent memory) |
+| `clawhdf5-tools` | `h5rs`: pure-Rust HDF5 tools — `ls`, `dump` (DDL / hdf5-json), `stat`, `diff`, `check` (structural + checksum validator) |
 | `clawhdf5-napi` | Node.js native addon bindings |
 | `clawhdf5-py` | PyO3 Python bindings |
+| `clawhdf5-wasm` | WebAssembly (wasm-bindgen) reader for the browser; demo in `examples/wasm-viewer/` |
 | `clawhdf5-bench` | Benchmark suite |
 
 ## Key Features
@@ -104,11 +106,34 @@ Cargo workspace with 16 crates under `crates/` (plus `libaec-sys`, an internal F
   the allowed records whenever cheaper than `pool × M` index distance
   evaluations, and as the fallback when the pool comes back short), fusion,
   activation scaling, optional re-ranking and confidence rejection.
-  `hybrid_search`/`hybrid_search_with` are thin wrappers; the OpenClaw
-  backend is `search` with re-rank + confidence on. Measure changes with
+  `hybrid_search`/`hybrid_search_with` are thin wrappers; `ClawhdfBackend`
+  (the `openclaw` module) is `search` with re-rank + confidence on.
+- **OpenClaw is not supported** (decided 2026-09-25): clawhdf5 is not an
+  OpenClaw memory plugin and never was — the old `memory.backend = "clawhdf5"`
+  config was never valid. Don't reintroduce OpenClaw claims; `docs/openclaw.md`
+  records what a real plugin would need.
+- **ZeroClaw does not use clawhdf5** (checked 2026-09-25 against upstream
+  v0.8.5 and the `osobh/zeroclaw` fork, and their full history): no
+  `clawhdf5` feature or backend exists; ZeroClaw's memory backends are
+  sqlite/lucid/postgres/qdrant/markdown/none behind its own `Memory` trait.
+  `clawhdf5-migrate`'s default SQLite layout (`memory_chunks`, `sessions`,
+  `entities`, `relations`) is not ZeroClaw's schema either (ZeroClaw's is a
+  `memories` table). Don't reintroduce integration claims without an
+  integration and a test against the real consumer. Measure changes with
   `search_harness --options-study`.
 - `MemoryConfig::compression` is off by default; when on, embeddings are
   deflate-compressed, or Zstd with the agent's `zstd` feature (links libzstd).
+- Signed checkpoints (`clawhdf5-agent` `signing` module): with
+  `HDF5Memory::set_signing_key` every checkpoint stores an Ed25519-signed
+  manifest (SHA-256 per record in a Merkle tree + settings/sessions/graph
+  hashes; per-record hashes in `/integrity/record_hashes`);
+  `HDF5Memory::verify(path, &pk)` locates edits. The hashes must cover exactly
+  what the file persists in the form the loader returns it (strings lose
+  trailing NULs; an empty WAL mark is not written) or untouched stores stop
+  verifying — `tests/signed_store.rs` round-trips awkward strings. The key is
+  never persisted; a signed store refuses to checkpoint without it
+  (`MemoryError::SigningKeyRequired`, and `MemoryError` is `#[non_exhaustive]`).
+  WAL entries after the checkpoint are not covered.
 - `Dataset::verify_provenance()` (clawhdf5 facade, `provenance` feature, on by
   default) recomputes a dataset's SHA-256 and compares it against the
   `_provenance_sha256` attribute written automatically on save when
@@ -125,7 +150,15 @@ Cargo workspace with 16 crates under `crates/` (plus `libaec-sys`, an internal F
   Alerts never block a save — drain them with `HDF5Memory::take_anomaly_alerts`.
   `MemorySource` for this bookkeeping is inferred from the caller-supplied
   `source_channel` string (a heuristic, not an authenticated trust boundary).
-- GPU-accelerated batch I/O for large dataset processing
+- GPU-accelerated vector distance computation (`clawhdf5-gpu`, wgpu); HDF5 I/O itself is CPU-only
+- Browser: `clawhdf5-wasm` (wasm-bindgen, read-only, file held in memory;
+  no Zstd/SZIP since they link C) and the `examples/wasm-viewer/` page.
+  `examples/wasm-viewer/test/run.sh` builds the package (needs the
+  `wasm-bindgen` CLI at the crate's exact version) and tests it under Node
+  and headless Chromium (a Playwright download in `~/.cache/ms-playwright`
+  on tank); the CI container has neither, so CI runs the native
+  `clawhdf5-wasm` `h5py_interop` test on the same fixture. Size numbers are
+  in the example's README.
 - Python and Node.js bindings for cross-language use
 - NetCDF-4 compatibility for scientific data interop
 
@@ -165,6 +198,16 @@ cargo run -p clawhdf5-cli -- --help
 # create, save, search, recall, stats, flush-wal, agents-md, export, snapshot subcommands
 ```
 
+### HDF5 tools (`h5rs`, crate `clawhdf5-tools`)
+```bash
+cargo run -p clawhdf5-tools -- ls -r file.h5    # also dump [--json], stat, diff, check
+bash scripts/h5rs-fuzz.sh                       # every subcommand over the CVE corpus: no panic/crash/hang
+bash scripts/h5rs-check-ok-files.sh --data      # check passes every fully-read conformance file
+```
+Its interop tests compare against h5ls/h5stat/h5dump/h5diff (Debian
+`hdf5-tools`, installed in CI); `dump` must stay byte-identical to h5dump on
+the test files.
+
 ### Python bindings
 ```bash
 cd crates/clawhdf5-py
@@ -173,4 +216,12 @@ python -c "import clawhdf5; print(clawhdf5.__version__)"
 ```
 
 ## Integration
-ZeroClaw imports this as a Cargo feature (`clawhdf5` feature flag) to persist agent memory with HNSW vector search for context retrieval.
+- **ClawBrainHub** (`clawverse/clawbrainhub` on git.redclaw.dev) is the one
+  verified consumer: `cbh-core` reads and writes `.brain` files through the
+  facade (`File`, `FileBuilder`, `AttrValue`, `Selection`), `cbh-scanner`
+  uses the facade, and `cbh-cli` uses `clawhdf5_agent::bm25::BM25Index`. It
+  depends on this repo by path (`../clawhdf5`), so it builds against whatever
+  is checked out — changes to those APIs reach it directly. Verified
+  2026-09-25 against main: builds, and its 204 tests pass.
+- OpenClaw and ZeroClaw were both described as consumers; neither integrates
+  clawhdf5 (see Key Features and `docs/openclaw.md`).

CONFORMANCE.md

@@ -0,0 +1,298 @@
+# clawhdf5 conformance report
+
+Every HDF5 file of eight public corpora (pinned by commit) is read twice — by
+clawhdf5 (`conformance/probe`, the same `clawhdf5-format` calls the facade
+makes) and by h5py/libhdf5 (`conformance/ref.py`) — and the two readings are
+compared object by object: the set of hard-linked objects, each dataset's and
+attribute's shape, and a SHA-256 of its values in a canonical encoding. The
+CVE corpus is also run through `h5dump`. Each side runs under a timeout and an
+address-space limit, so a hang, crash or runaway allocation is recorded, not
+fatal. This file is generated by `conformance/run.sh`; do not edit it by hand.
+
+## Run
+
+| | |
+|---|---|
+| date | 2026-09-26 14:18 UTC |
+| clawhdf5 commit | `73a01f1256fb9bf1b1e7601f755af9e8273cec4e` |
+| machine | `tank`: AMD Ryzen 7 7800X3D 8-Core Processor, 16 CPUs, 61 GiB, Linux 7.0.0-34-generic x86_64 |
+| command | `conformance/run.sh --no-fetch --update-baseline` |
+| rustc | rustc 1.98.1 (48a229cea 2026-09-01) |
+| reference | h5py 3.16.0, HDF5 2.0.0, numpy 2.5.3, hdf5plugin 7.1.0, Python 3.14.4 |
+| h5dump | Version 1.14.6 (CVE corpus only) |
+| limits | 20 s timeout (SIGKILL), 4096 MiB address space, per process; 16 files in parallel |
+| runtime | 23 s probing + comparing (0 s fetch/build before it) |
+
+## Results
+
+A file's class is the first that applies:
+
+- **panic / hang / crash / oom** — clawhdf5 panicked (caught per object or not), hit the timeout, died on a signal, or failed an allocation. The CI gate fails on any of these.
+- **h5py-cannot-read** — libhdf5 could not open the file (or itself crashed or hung). Nothing to compare against; most are the deliberately malformed CVE reproducers.
+- **our-error** — clawhdf5 returned an error for something h5py reads.
+- **mismatch** — both read it, but the shapes, values, object set or attribute set differ.
+- **ok** — every object h5py reads, clawhdf5 reads identically.
+
+| corpus | files | ok | our-error | mismatch | h5py-cannot-read | panic | hang | crash | oom |
+|---|---|---|---|---|---|---|---|---|---|
+| NCAS-CMS_pyfive | 33 | 32 | 0 | 1 | 0 | 0 | 0 | 0 | 0 |
+| cve_hdf5 | 147 | 100 | 6 | 9 | 32 | 0 | 0 | 0 | 0 |
+| h5py_data | 4 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
+| hdf5 | 466 | 392 | 4 | 10 | 60 | 0 | 0 | 0 | 0 |
+| netcdf-c | 20 | 20 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
+| netcdf4-python | 18 | 18 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
+| usnistgov_h5wasm | 5 | 5 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
+| xarray-data | 4 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
+| **all** | **697** | **575** | **10** | **20** | **92** | **0** | **0** | **0** | **0** |
+
+2 of the 20 mismatches are a known h5py bug, not ours (see *Known not-our-bug*).
+
+Corpora (fetched by `conformance/fetch-corpus.sh` into the gitignored `conformance/.cache/`):
+
+| corpus | source | commit |
+|---|---|---|
+| hdf5 | https://github.com/HDFGroup/hdf5 | `a3cf1ea82cc7` |
+| cve_hdf5 | https://github.com/HDFGroup/cve_hdf5 | `3fd1f5ae3869` |
+| netcdf-c | https://github.com/Unidata/netcdf-c | `beb7b9585273` |
+| NCAS-CMS_pyfive | https://github.com/NCAS-CMS/pyfive | `8cf07b874913` |
+| usnistgov_h5wasm | https://github.com/usnistgov/h5wasm | `02f6336527d2` |
+| netcdf4-python | https://github.com/Unidata/netcdf4-python | `6e67576d39ae` |
+| xarray-data | https://github.com/pydata/xarray-data | `a35297e9da2c` |
+| h5py_data | https://github.com/h5py/h5py (`h5py/tests/data_files`) | `b2f0347c4200` |
+
+## Panics, hangs, crashes, out-of-memory
+
+None.
+
+## Our-error root causes
+
+Grouped by normalised error message. *files* counts files whose class this cause affects.
+
+| files | objects | error | examples |
+|---:|---:|---|---|
+| 3 | 3 | `DataSizeMismatch { expected: N, actual: N }` | `cve_hdf5/cvefiles/cve-2020-18494.h5`, `cve_hdf5/cvefiles/cve-2024-32623.h5`, `cve_hdf5/cvefiles/cve-2025-2309.h5` |
+| 2 | 2 | `ChunkedReadError("…")` | `cve_hdf5/cvefiles/cve-2025-2308.h5`, `hdf5/test/testfiles/bad_nbit_parms_walk.h5` |
+| 2 | 2 | `UnsupportedFilter(N)` | `hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_blosc2.h5`, `hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_zfp.h5` |
+| 1 | 1 | `UnexpectedEof { expected: N, available: N }` | `cve_hdf5/cvefiles/cve-2019-9151.h5` |
+| 1 | 1 | `MissingMessage(Dataspace)` | `cve_hdf5/cvefiles/cve-2024-33874.h5` |
+| 1 | 1 | `InvalidObjectHeaderVersion(N)` | `hdf5/tools/test/testfiles/h5clear_mdc_image.h5` |
+
+## Mismatch root causes
+
+| files | objects | cause | examples |
+|---:|---:|---|---|
+| 13 | 14 | `missing-object` | `cve_hdf5/cvefiles/cve-2019-8397.h5`, `cve_hdf5/cvefiles/cve-2019-8398.h5`, `cve_hdf5/cvefiles/cve-2021-46243.h5` (+10 more) |
+| 2 | 6 | `extra-attr` | `cve_hdf5/cvefiles/cve-2018-17438`, `cve_hdf5/cvefiles/cve-2018-17439` |
+| 1 | 1 | `attr-values: ours=vlen(>u8) h5py=object layout=- filters=-` | `NCAS-CMS_pyfive/tests/data/attr_datatypes.hdf5` |
+| 1 | 4 | `extra-object` | `cve_hdf5/cvefiles/cve-2021-46244.h5` |
+| 1 | 1 | `values: ours=<f4 h5py=float32 layout=chunked filters=-` | `cve_hdf5/cvefiles/cve-2025-44904.h5` |
+| 1 | 1 | `values: ours=>i2 h5py=>i2 layout=chunked filters=[6]` | `cve_hdf5/cvefiles/cve-2025-44905.h5` |
+| 1 | 1 | `values: ours=>f4 h5py=>f4 layout=chunked filters=[2]` | `cve_hdf5/cvefiles/cve-2025-44905.h5` |
+| 1 | 1 | `values: ours=<f4 h5py=float32 layout=chunked filters=[2]` | `cve_hdf5/cvefiles/cve-2025-44905.h5` |
+| 1 | 1 | `values: ours=((<i4)[6, 3])[4] h5py=(('<i4', (6, 3)), (4,)) layout=contiguous filters=-` | `hdf5/tools/test/testfiles/tarray3.h5` |
+| 1 | 1 | `values: ours=vlen({r:>f4,i:>f4}8) h5py=object layout=contiguous filters=-` | `hdf5/tools/test/testfiles/tcomplex_be.h5` |
+
+## CVE corpus: clawhdf5 vs h5dump vs h5py
+
+The 147 files of [HDFGroup/cve_hdf5](https://github.com/HDFGroup/cve_hdf5) — reproducers for
+published libhdf5 CVEs and fuzzer finds. *read* = produced output (possibly with per-object
+errors), *error* = refused cleanly. h5dump exits non-zero on any error anywhere in a file, so
+its read/error split is not comparable with the other two rows; the panic, crash, hang and oom
+columns are.
+
+| tool | read | error | panic | crash | hang | oom |
+|---|---:|---:|---:|---:|---:|---:|
+| clawhdf5 | 140 | 7 | 0 | 0 | 0 | 0 |
+| h5dump 1.14.6 | 16 | 129 | 0 | 2 | 0 | 0 |
+| h5py 3.16.0 / HDF5 2.0.0 | 115 | 31 | 0 | 1 | 0 | 0 |
+
+<details><summary>Per-file outcomes</summary>
+
+| file | h5dump | h5py | clawhdf5 | class |
+|---|---|---|---|---|
+| cvefiles/cve-2016-4330.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2016-4331.h5 | error exit | read 25 obj, 1 errors | read 25 obj, 1 errors | ok |
+| cvefiles/cve-2016-4332-mtime-new.h5 | error exit | read 25 obj, 1 errors | read 25 obj, 1 errors | ok |
+| cvefiles/cve-2016-4332-mtime.h5 | error exit | read 4 obj, 3 errors | read 4 obj, 3 errors | ok |
+| cvefiles/cve-2016-4332-stab.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2016-4333.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2017-17505.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2017-17506.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2017-17507.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2017-17508.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2017-17509.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-11202.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-11203.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-11204.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-11205.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-11206-new.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-11206-old.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-11207.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-13866.h5 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2018-13867.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2018-13868.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-13869.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2018-13870.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2018-13871.h5 | error exit | read 2 obj | read 2 obj | ok |
+| cvefiles/cve-2018-13872.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2018-13873.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2018-13874.h5 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2018-13875.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-13876.h5 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2018-14031.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-14033.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-14034.h5 | error exit | read 1 obj, 2 errors | read 1 obj | ok |
+| cvefiles/cve-2018-14035.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-14460.h5 | error exit | read 3 obj, 2 errors | read 3 obj, 2 errors | ok |
+| cvefiles/cve-2018-15671.h5 | ok | read 1 obj | read 1 obj | ok |
+| cvefiles/cve-2018-15672.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-16438.h5 | error exit | read 1 obj, 1 errors | read 1 obj | ok |
+| cvefiles/cve-2018-17233.h5 | error exit | read 6 obj, 1 errors | read 6 obj, 1 errors | ok |
+| cvefiles/cve-2018-17234.h5 | error exit | read 6 obj, 1 errors | read 6 obj, 1 errors | ok |
+| cvefiles/cve-2018-17237.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2018-17432.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-17433 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2018-17434.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-17435.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-17436 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2018-17437.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2018-17438 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | mismatch |
+| cvefiles/cve-2018-17439 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | mismatch |
+| cvefiles/cve-2019-8396.h5 | error exit | read 3 obj, 2 errors | read 3 obj, 2 errors | ok |
+| cvefiles/cve-2019-8397.h5 | error exit | read 3 obj, 2 errors | read 2 obj, 1 errors | mismatch |
+| cvefiles/cve-2019-8398.h5 | error exit | read 3 obj, 2 errors | read 2 obj, 1 errors | mismatch |
+| cvefiles/cve-2019-9151.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 2 errors | our-error |
+| cvefiles/cve-2019-9152.h5 | error exit | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2020-10809 | error exit | open error | open error | h5py-cannot-read |
+| cvefiles/cve-2020-10810.h5 | error exit | open error | read 2 obj | h5py-cannot-read |
+| cvefiles/cve-2020-10811.h5 | error exit | read 25 obj, 1 errors | read 25 obj, 1 errors | ok |
+| cvefiles/cve-2020-10812.h5 | error exit | open error | read 2 obj | h5py-cannot-read |
+| cvefiles/cve-2020-18232.h5 | error exit | read 3 obj, 2 errors | read 3 obj, 2 errors | ok |
+| cvefiles/cve-2020-18494.h5 | ok | read 2 obj | read 2 obj, 1 errors | our-error |
+| cvefiles/cve-2021-36977.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2021-37501.h5 | error exit | read 18 obj, 1 errors | read 18 obj, 1 errors | ok |
+| cvefiles/cve-2021-45829.h5 | error exit | read 1 obj, 2 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2021-45830.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2021-45833.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2021-46242.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2021-46243.h5 | error exit | read 3 obj, 2 errors | read 2 obj, 1 errors | mismatch |
+| cvefiles/cve-2021-46244.h5 | error exit | read 2 obj, 1 errors | read 6 obj, 4 errors | mismatch |
+| cvefiles/cve-2024-29157.h5 | error exit | read 4 obj, 7 errors | read 4 obj, 7 errors | ok |
+| cvefiles/cve-2024-29158.h5 | ok | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2024-29159.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2024-29160.h5 | error exit | read 4 obj, 1 errors | read 4 obj, 1 errors | ok |
+| cvefiles/cve-2024-29161.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2024-29162.h5 | error exit | read 17 obj, 4 errors | read 17 obj, 4 errors | ok |
+| cvefiles/cve-2024-29163.h5 | error exit | read 7 obj, 1 errors | read 7 obj, 1 errors | ok |
+| cvefiles/cve-2024-29164.h5 | ok | read 3 obj | read 3 obj | ok |
+| cvefiles/cve-2024-29165.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2024-29166.h5 | error exit | read 17 obj, 2 errors | read 17 obj | ok |
+| cvefiles/cve-2024-32605.h5 | ok | read 6 obj, 1 errors | read 6 obj, 1 errors | ok |
+| cvefiles/cve-2024-32606.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2024-32607-1.h5 | ok | read 10 obj | read 10 obj | ok |
+| cvefiles/cve-2024-32607-2.h5 | error exit | read 9 obj, 1 errors | read 9 obj, 1 errors | ok |
+| cvefiles/cve-2024-32608.h5 | error exit | read 6 obj, 1 errors | read 6 obj, 1 errors | ok |
+| cvefiles/cve-2024-32609.h5 | error exit | SIGSEGV | read 3 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2024-32610.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2024-32611.h5 | ok | read 6 obj | read 6 obj | ok |
+| cvefiles/cve-2024-32612.h5 | ok | read 3 obj | read 3 obj | ok |
+| cvefiles/cve-2024-32613.h5 | error exit | read 7 obj, 1 errors | read 7 obj, 1 errors | ok |
+| cvefiles/cve-2024-32614.h5 | error exit | read 25 obj, 2 errors | read 25 obj, 2 errors | ok |
+| cvefiles/cve-2024-32615.h5 | error exit | read 4 obj, 1 errors | read 4 obj, 1 errors | ok |
+| cvefiles/cve-2024-32616.h5 | error exit | read 10 obj, 7 errors | read 10 obj, 6 errors | ok |
+| cvefiles/cve-2024-32617.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2024-32618.h5 | error exit | read 4 obj, 2 errors | read 3 obj, 1 errors | mismatch |
+| cvefiles/cve-2024-32619.h5 | error exit | read 3 obj, 2 errors | read 3 obj, 2 errors | ok |
+| cvefiles/cve-2024-32620.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2024-32621.h5 | ok | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2024-32622.h5 | ok | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2024-32623.h5 | ok | read 6 obj | read 6 obj, 1 errors | our-error |
+| cvefiles/cve-2024-32624.h5 | error exit | read 6 obj, 1 errors | read 6 obj | ok |
+| cvefiles/cve-2024-33873.h5 | error exit | read 4 obj, 1 errors | read 4 obj, 1 errors | ok |
+| cvefiles/cve-2024-33874.h5 | ok | read 6 obj, 1 errors | read 6 obj, 2 errors | our-error |
+| cvefiles/cve-2024-33875.h5 | ok | read 2 obj | read 2 obj | ok |
+| cvefiles/cve-2024-33876.h5 | ok | read 3 obj, 1 errors | read 3 obj, 1 errors | ok |
+| cvefiles/cve-2024-33877.h5 | error exit | read 8 obj, 1 errors | read 8 obj, 1 errors | ok |
+| cvefiles/cve-2025-2153.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2308.h5 | error exit | read 25 obj, 1 errors | read 25 obj, 2 errors | our-error |
+| cvefiles/cve-2025-2309.h5 | ok | read 6 obj, 1 errors | read 6 obj, 1 errors | our-error |
+| cvefiles/cve-2025-2310.h5 | error exit | read 24 obj, 8 errors | read 24 obj, 8 errors | ok |
+| cvefiles/cve-2025-2912.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2913.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2914.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2915.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2923.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-2924.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-2925.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-2926.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-44904.h5 | error exit | read 25 obj, 1 errors | read 25 obj, 1 errors | mismatch |
+| cvefiles/cve-2025-44905.h5 | error exit | read 25 obj, 3 errors | read 25 obj, 3 errors | mismatch |
+| cvefiles/cve-2025-6269-1.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6269-2.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6269-3.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6269-4.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6270-1.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6270-2.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6270-3.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6516.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6750.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6816.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6817.h5 | error exit | open error | read 1 obj | h5py-cannot-read |
+| cvefiles/cve-2025-6818.h5 | error exit | open error | read 1 obj | h5py-cannot-read |
+| cvefiles/cve-2025-6856.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-6857.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-6858.h5 | SIGSEGV | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-7067.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2025-7068.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2025-7069.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| cvefiles/cve-2026-26200.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| cvefiles/cve-2026-34734.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/cve-2026-92627.h5 | error exit | read 2 obj, 1 errors | read 2 obj, 1 errors | ok |
+| cvefiles/unknown-1.h5 | error exit | read 11 obj, 1 errors | read 11 obj, 1 errors | ok |
+| fuzzerfiles/gh-4431-poc-03.h5 | error exit | read 1 obj | read 1 obj | ok |
+| fuzzerfiles/gh-4432-poc-05.h5 | SIGSEGV | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| fuzzerfiles/gh-4433-poc-08.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| fuzzerfiles/gh-4434-poc-09.h5 | error exit | open error | read 1 obj, 1 errors | h5py-cannot-read |
+| fuzzerfiles/gh-4435-poc-10.h5 | error exit | read 1 obj, 1 errors | read 1 obj, 1 errors | ok |
+| fuzzerfiles/gh-4585.h5 | error exit | open error | open error | h5py-cannot-read |
+| fuzzerfiles/gh_2649_flawed.h5 | error exit | read 9 obj, 1 errors | read 9 obj, 1 errors | ok |
+| fuzzerfiles/gh_2649_plain_model.h5 | ok | read 10 obj | read 10 obj | ok |
+
+</details>
+
+## Known not-our-bug
+
+- **h5py big-endian variable-length sequences.** h5py returns the elements of a VL sequence
+  whose base type is big-endian with the file's big-endian bytes but a native (little-endian)
+  numpy dtype, so the values it reports are byte-swapped garbage; `h5dump` prints the values
+  clawhdf5 reads. Reproducer: `h5py.vlen_dtype(np.dtype('>f4'))` dataset holding `[1.0, 2.0]`
+  reads back in h5py as `[4.6e-41, 9.0e-44]`. Affected here: `NCAS-CMS_pyfive/tests/data/attr_datatypes.hdf5`, `hdf5/tools/test/testfiles/tcomplex_be.h5`.
+- **Non-IEEE floats and partial-precision integers (N-Bit).** libhdf5 converts a float whose
+  bit layout is not IEEE (e.g. `H5Tset_precision` for the N-Bit filter) or an integer with a
+  bit offset / reduced precision into the plain numpy type of the same size. The probe
+  compares such values as converted numbers, not raw file bytes (before 2026-09-25 it compared
+  raw bytes, which reported every N-Bit float dataset as a mismatch).
+- **Types h5py widens.** Where h5py reads a type into a numpy type of a different size
+  (FP8 -> float16, bfloat16 -> float32, x87 long double -> float128) the values are not
+  compared (shape and presence still are): dataset file type size 1 -> numpy float16 (2) (15x), attr file type size 1 -> numpy float16 (2) (15x), dataset file type size 2 -> numpy float32 (4) (2x), dataset file type size 8 -> numpy float128 (16) (1x), dataset file type size 12 -> numpy float128 (16) (1x), attr file type size 2 -> numpy float32 (4) (1x), dataset file type size 2 -> numpy >f4 (4) (1x), attr file type size 2 -> numpy >f4 (4) (1x).
+- **References** are compared by presence only (`R`), not by target.
+
+## Objects h5py fails on but clawhdf5 reads
+
+- 19 x `OSError: Can't synchronously read data (no appropriate function for conversion path)`
+- 1 x `TypeError: unhandled dtype kind M (dtype('…'))`
+- 1 x `TypeError: No NumPy equivalent for TypeTimeID exists`
+- 1 x `KeyError: "…"`
+- 1 x `ValueError: Insufficient precision in available types to represent (N, N, N, N, N)`
+
+## Reproduce
+
+```sh
+# needs: Rust, python3 with h5py numpy hdf5plugin (conformance/requirements.txt), h5dump (hdf5-tools), git
+CLAWHDF5_PYTHON=/path/to/venv/bin/python conformance/run.sh
+```
+
+The corpus (about 450 MB of sparse checkouts) is cached in `conformance/.cache/`; results for
+every file, both sides' raw JSON and stderr, are in `conformance/.cache/results/`.
+`conformance/baseline.json` holds the ok files the nightly CI job (`.gitea/workflows/conformance.yml`)
+must keep; `conformance/run.sh --update-baseline` rewrites it.

Cargo.toml

@@ -16,6 +16,8 @@ members = [
     "crates/clawhdf5-cli",
     "crates/clawhdf5-napi",
     "crates/clawhdf5-bench",
+    "crates/clawhdf5-tools",
+    "crates/clawhdf5-wasm",
     "crates/libaec-sys",
 ]
 resolver = "2"
@@ -34,3 +36,12 @@ tempfile = "3"
 criterion = { version = "0.5", features = ["html_reports"] }
 half = "2.7"
 serde = { version = "1", features = ["derive"] }
+
+# The browser build of clawhdf5-wasm (examples/wasm-viewer/build.sh): size
+# over speed, whole-program optimisation. Native profiles are unaffected.
+[profile.wasm-release]
+inherits = "release"
+opt-level = "s"
+lto = true
+codegen-units = 1
+panic = "abort"

README.md

@@ -8,7 +8,7 @@
 [![LongMemEval](https://img.shields.io/badge/LongMemEval__s-Turn--Level%20Hit@5%2081.4%25%20hybrid-blue.svg)](BENCHMARKS.md#longmemeval-results)
 [![Footprint](https://img.shields.io/badge/on--disk-~820%20B%2Frecord%20float16%2C%20synthetic%20text-lightgrey.svg)](BENCHMARKS.md#memory-footprint-1)
 
-ClawHDF5 is a pure-Rust HDF5 implementation combined with a research-grade agent memory engine. It gives AI agents persistent, searchable, integrity-checked memory — all stored in a single portable file.
+ClawHDF5 is a pure-Rust HDF5 implementation combined with a research-grade agent memory engine. It gives AI agents persistent, searchable, cryptographically verifiable memory (Ed25519-signed checkpoints) — all stored in a single portable file.
 
 > **Two things live here:**
 > - **A general-purpose, pure-Rust HDF5 library** — zero C dependencies, NetCDF-4 support, SIMD/GPU acceleration. See the **[Crate Map](#crate-map)** and **[BENCHMARKS.md](BENCHMARKS.md)** for the libhdf5 head-to-head numbers.
@@ -71,10 +71,11 @@ breaking change, are in [CHANGELOG.md](CHANGELOG.md).
   100K). It no longer rebuilds BM25 or rewrites the store per query, and the
   HNSW graph is persisted (v2.4.0).
 - Default fusion weights are now the measured 0.4 / 0.6 (v2.5.0). Re-ranking had
-  been discarding the retrieval score, costing the OpenClaw backend 40.6pp of
+  been discarding the retrieval score, costing the Markdown backend 40.6pp of
   Hit@1; fixed in v2.6.0.
-- Selection reads decode only the chunks they touch (a 64×64 window: 105 ms to
-  0.39 ms), and full reads are 1.2–1.9× faster (v2.5.0).
+- Selection reads whose bounding box covers at most half the dataset decode
+  only the chunks they touch (a 64×64 window: 105 ms to 0.39 ms), and full
+  reads are 1.2–1.9× faster (v2.5.0).
 
 **Memory**
 - A loaded store holds ~30% less (embeddings stored once, v2.6.0), and the
@@ -93,7 +94,7 @@ breaking change, are in [CHANGELOG.md](CHANGELOG.md).
   identical LongMemEval retrieval on real embeddings.
 - `HDF5Memory::search` with `SearchOptions`: filter by source channel (exact
   filtered top-k, never slower than unfiltered), and opt-in re-ranking and
-  confidence rejection, which used to be OpenClaw-only.
+  confidence rejection, which used to be reachable only through `ClawhdfBackend`.
 
 **Tooling**
 - CI now runs the h5py/netCDF4 interop suites for real (they had been skipping
@@ -113,7 +114,7 @@ Every AI agent needs memory. Today that means scattered Markdown files, SQLite d
 | Memory consolidation | Manual pruning | Hippocampal-inspired automatic tiers |
 | Temporal queries | Custom code | Native temporal index (622 ns range query over 10K) |
 | Multi-modal | Multiple stores | Unified cross-modal search (exact scan: 842 µs over 1K records) |
-| Integrity | Hope for the best | Chained-CRC WAL, checksummed chunk indexes, write-anomaly alerts, opt-in SHA-256 dataset provenance |
+| Integrity | Hope for the best | Ed25519-signed checkpoints that pinpoint any edited record, chained-CRC WAL, checksummed chunk indexes, write-anomaly alerts |
 | Portability | Config + DB + files | **One `.h5` file. Copy it anywhere.** |
 
 ---
@@ -228,7 +229,7 @@ is for. The weights matter more than the stages: a sweep of `vector_weight` from
 0.0 to 1.0 found the old `0.7/0.3` default is **strictly dominated** by
 `0.4/0.6` — better on Hit@1, Hit@5, Hit@10 and MRR at both granularities. Since
 v2.5.0 `0.4/0.6` is the default (`hybrid::DEFAULT_FUSION`, used by
-`unified_search`, `hybrid_search_with` and the OpenClaw backend); callers that
+`unified_search`, `hybrid_search_with` and `ClawhdfBackend`); callers that
 pass weights to `hybrid_search` explicitly choose their own. Use `0.3/0.7` if
 rank-1 precision matters most. Reciprocal rank fusion is selectable
 (`hybrid::Fusion::Rrf`) but measured worse than the weighted sum. See
@@ -329,7 +330,7 @@ ClawhDF5's agent memory engine draws on 15+ recent papers on agentic memory syst
                 │ × √(Hebbian activation)            │
                 └─────────────────┬──────────────────┘
                                   │  opt-in (SearchOptions);
-                                  │  the OpenClaw backend turns both on
+                                  │  ClawhdfBackend turns both on
                 ┌─────────────────▼──────────────────┐
                 │ Multi-factor re-ranking            │
                 │ relevance · recency · authority ·  │
@@ -365,13 +366,14 @@ directly; the store persists the records, sessions and graph they work over.
 | **`knowledge`** | Entity/relation graph with BFS traversal, spreading activation, fuzzy (Levenshtein) entity resolution |
 | **`consolidation`** | Three-tier memory (Working → Episodic → Semantic) with importance scoring, novelty, and time-decay |
 | **`hybrid`** | Vector + BM25 fusion. Default is a min-max-normalised weighted sum, vector 0.4 / keyword 0.6 (`hybrid::DEFAULT_FUSION`, tuned on LongMemEval); RRF is available via `Fusion::Rrf` / `hybrid_search_with`. The vector stage uses the HNSW index by default (`hnsw` feature); disable with `--no-default-features --features float16` for an exact linear scan |
-| **`reranker`** | Multi-factor re-ranking: retrieval relevance (leads, weight 1.0), temporal recency, source authority, activation weight. Opt-in via `SearchOptions::with_rerank`; on in the OpenClaw backend |
-| **`confidence`** | Low-confidence rejection — suppresses spurious recalls when nothing matches. Opt-in via `SearchOptions::with_confidence`; on in the OpenClaw backend |
+| **`reranker`** | Multi-factor re-ranking: retrieval relevance (leads, weight 1.0), temporal recency, source authority, activation weight. Opt-in via `SearchOptions::with_rerank`; on in `ClawhdfBackend` |
+| **`confidence`** | Low-confidence rejection — suppresses spurious recalls when nothing matches. Opt-in via `SearchOptions::with_confidence`; on in `ClawhdfBackend` |
 | **`temporal`** | Sorted timestamp index, session DAG, entity timeline, temporal query hints |
 | **`multimodal`** | Cross-modal search across text/image/audio/video embeddings |
+| **`signing`** | Ed25519-signed checkpoints: SHA-256 per record in a Merkle tree, plus hashes of settings, sessions and the knowledge graph; `HDF5Memory::verify` names any edited record |
 | **`provenance`** | Source attribution and an unkeyed FNV-1a content hash per record, held in memory for the session, for detecting accidental corruption (not tamper-proof) |
 | **`anomaly`** | Write rate limiting, 15 injection-pattern detectors, source-distribution analysis. Alerts never block a save; drain them with `take_anomaly_alerts` |
-| **`openclaw`** | OpenClaw integration: MemoryBackend trait, Markdown ↔ HDF5 conversion |
+| **`openclaw`** | `ClawhdfBackend`: a Markdown-oriented backend (ingest by section, search, read back by path, export). Named for OpenClaw, but **not an OpenClaw plugin** — see [docs/openclaw.md](docs/openclaw.md) |
 | **`vector_search`** | Flat cosine, pre-normed, SIMD, BLAS, GPU, parallel search paths |
 | **`ivf` / `pq`** | Standalone IVF and IVF-PQ indexes (benchmarked to 100K vectors); not used by `HDF5Memory`, whose ANN index is HNSW |
 | **`bm25`** | Incremental Okapi BM25 inverted index, kept for the life of the store; optional stemming |
@@ -406,6 +408,79 @@ let values = ds.read_f64()?;
 assert_eq!(values, vec![22.5, 23.1, 21.8]);
 ```
 
+### Groups and links
+
+```rust
+use clawhdf5::{AttrValue, FileBuilder};
+
+let mut b = FileBuilder::new();
+// A path creates its missing intermediate groups, as in h5py.
+b.create_dataset("run/2026/temps").with_f64_data(&[22.5, 23.1]);
+// Builders nest; a group added at an existing path is merged into it.
+let mut run = b.create_group("run");
+run.set_attr("operator", AttrValue::String("ana".into()));
+let mut cal = run.create_group("calibration");
+cal.track_order(true); // h5py lists members in insertion order
+cal.create_dataset("offset").with_f64_data(&[0.1]);
+run.add_group(cal.finish());
+b.add_group(run.finish());
+b.add_soft_link("latest", "/run/2026");    // h5py.SoftLink
+b.add_hard_link("temps", "/run/2026/temps"); // f["temps"] = f["run/2026/temps"]
+b.add_external_link("raw", "raw.h5", "/data");
+b.write("groups.h5")?;
+```
+
+A group holds at most 65 535 links; more is an error, as is a link over
+65 515 bytes (a very long soft-link target) in a group of more than 8 links.
+
+### Python
+
+`crates/clawhdf5-py` is a Python package (PyO3 + numpy) that reads HDF5 with
+an h5py-shaped API and no libhdf5. It is not on PyPI; build it with
+[maturin](https://www.maturin.rs) into a virtualenv:
+
+```bash
+python -m venv .venv && . .venv/bin/activate
+pip install maturin numpy
+maturin develop --release -m crates/clawhdf5-py/Cargo.toml
+python -c "import clawhdf5; print(clawhdf5.__version__)"
+```
+
+```python
+import numpy as np
+import clawhdf5
+
+with clawhdf5.File("data.h5", "r") as f:
+    print(list(f.keys()))              # sorted member names, like h5py
+    ds = f["group/temperatures"]       # relative or absolute ("/group/...") paths
+    print(ds.shape, ds.dtype)          # dtype is the numpy dtype h5py reports
+    block = ds[100:200, ::4]           # a small selection reads only its chunks
+    row = ds[-1]                       # integers drop the axis
+    picked = ds[[1, 5, 9], :]          # one increasing index list per key
+    units = ds.attrs["units"]          # attributes come back as h5py returns them
+    everything = np.asarray(ds)
+
+    records = f["table"]               # compound -> numpy structured array
+    ids = records["id"]                # one field
+```
+
+Reads cover integers and IEEE floats of every width in either byte order,
+`bool`, enums, complex, fixed and variable-length strings, variable-length
+sequences, opaque, HDF5 array types and compounds; other types (references,
+bitfields, ...) raise `TypeError` instead of returning guessed data. Keys
+follow h5py (negative steps, `None` and boolean masks are refused). The
+read itself runs with the GIL released, so Python threads read in parallel.
+A selection whose bounding box covers at most half the dataset decodes only
+the chunks (or contiguous rows) that box overlaps; a larger one — including
+a strided slice across the whole dataset — decodes the whole dataset, as
+do datasets that are compact, virtual, unwritten, or chunked with a
+non-default fill value (`docs/known-issues.md`). An index list is read one
+group of neighbouring chunks at a time.
+Writing (`File(path, "w")`, `create_dataset`, `create_group`, `attrs[...] =`)
+covers `float64`, `float32`, `int64`, `int32` and `uint8` arrays. The tests
+in `crates/clawhdf5-py/tests` compare every read with h5py; run them with
+`pip install pytest h5py && pytest crates/clawhdf5-py/tests`.
+
 ### Agent Memory
 
 ```rust
@@ -447,7 +522,7 @@ let work = memory.search(
 );
 
 // Re-rank by relevance, recency, source authority and activation, then drop
-// low-confidence results — the pipeline the OpenClaw backend runs.
+// low-confidence results — the pipeline ClawhdfBackend runs.
 let careful = memory.search(
     &query_embedding,
     "user preferences",
@@ -457,6 +532,35 @@ let careful = memory.search(
 );
 ```
 
+### Signed Checkpoints
+
+```rust
+use clawhdf5_agent::signing;
+
+// Once, somewhere safe: keep the secret key, publish the public key.
+let key = signing::generate_key();
+let public = key.verifying_key();
+
+// Every checkpoint is signed from now on. The key is never written to disk;
+// a signed store refuses to checkpoint without it.
+memory.set_signing_key(key);
+memory.flush_wal()?;
+
+// Anyone holding the public key can check the file, e.g. after copying it.
+let report = HDF5Memory::verify(std::path::Path::new("agent.h5"), &public)?;
+assert!(report.is_valid());
+// On a tampered file: report.changed_records lists the records that differ.
+```
+
+The signature covers every record (text, embedding as stored, channel,
+timestamp, session, tags, deleted flag, activation), the store's settings,
+its sessions and its knowledge graph — a change made with any tool is caught.
+It covers checkpoints, not saves still in the WAL
+(`report.wal_entries_unsigned` counts those). CLI: `clawhdf5-cli keygen`,
+`--signing-key <file>` on writing commands, and `verify --public-key`.
+Signing adds about 20% to a checkpoint and 32 bytes per record to the file
+([BENCHMARKS.md § Signed checkpoints](BENCHMARKS.md#signed-checkpoints)).
+
 ### Knowledge Graph
 
 ```rust
@@ -527,7 +631,13 @@ let ids = index.range_query(1700000000.0, 1700010800.0);
 let recent = index.latest(10);
 ```
 
-### OpenClaw Integration
+### Markdown Backend
+
+`ClawhdfBackend` ingests Markdown by section and searches it with the full
+pipeline. It is a library API — clawhdf5 is **not** an OpenClaw memory plugin
+([docs/openclaw.md](docs/openclaw.md)). Sections stored this way carry no
+embedding, so their search is keyword-only unless you save records with
+vectors through `save_entry`.
 
 ```rust
 use clawhdf5_agent::openclaw::*;
@@ -551,14 +661,14 @@ let exported = backend.export_markdown("MEMORY.md")?;
 ## Crate Map
 
 ```
-clawhdf5 workspace (16 crates, ~86K lines of Rust in src/, ~104K with tests
+clawhdf5 workspace (17 crates, ~86K lines of Rust in src/, ~104K with tests
                      and benches; plus libaec-sys, an internal FFI bindings
                      crate for the optional szip feature)
 │
 ├── Core HDF5
 │   ├── clawhdf5-format      — Binary parser/writer (no_std-capable), shared type definitions
 │   ├── clawhdf5-io          — I/O abstraction (file/memory readers; optional mmap, async, HSDS, MPI)
-│   ├── clawhdf5-filters     — Fast deflate path (zlib-ng); lz4/zstd/pcodec/szip filters live in clawhdf5-format
+│   ├── clawhdf5-filters     — Fast deflate path (zlib-ng); the filter registry and the lz4/zstd/pcodec/szip/LZF/bitshuffle/bzip2/Blosc filters live in clawhdf5-format
 │   ├── clawhdf5-derive      — Proc macros
 │   ├── clawhdf5             — High-level API
 │   ├── clawhdf5-netcdf4     — NetCDF-4 support
@@ -574,7 +684,8 @@ clawhdf5 workspace (16 crates, ~86K lines of Rust in src/, ~104K with tests
 │
 ├── Bindings
 │   ├── clawhdf5-py          — Python (PyO3)
-│   └── clawhdf5-napi        — Node.js (napi-rs)
+│   ├── clawhdf5-napi        — Node.js (napi-rs)
+│   └── clawhdf5-wasm        — Browser (WebAssembly, wasm-bindgen; read-only)
 │
 └── Tooling
     └── clawhdf5-bench       — Benchmark suite
@@ -660,10 +771,26 @@ stores keep their setting. Opt out with `float16 = false` or
 | `fast-checksum` | no | crc32fast-accelerated checksums |
 | `lz4` | no | LZ4 block compression filter (id 32004) |
 | `zstd` | no | Zstandard compression filter (id 32015) |
-| `pcodec` | no | Pcodec lossless numerical codec (id 32023, via `pco` crate) |
+| `pcodec` | no | Pcodec lossless numerical codec (via `pco` crate). Private, unregistered filter id 480: **only clawhdf5 can read these datasets** (h5py/libhdf5 cannot). Files from clawhdf5 <= 2.7.0 used id 32023, which is registered to Granular BitRound; they still read. |
 | `system-zlib` | no | System zlib backend for deflate (C) |
 | `blake3_hash` | no | BLAKE3 content hashing for provenance |
 | `szip` | no | SZIP filter (id 4) via libaec (C, through the internal `libaec-sys` crate) |
+| `lzf` | **yes** | LZF filter (id 32000), h5py's built-in `compression="lzf"`: read and write. No dependencies |
+| `bitshuffle` | no | Bitshuffle filter (id 32008) with its LZ4 and Zstandard modes: read and write. Pure Rust (lz4_flex, ruzstd) |
+| `bzip2` | no | bzip2 filter (id 307): read and write. Pure Rust (the `bzip2` crate's libbz2-rs-sys backend compiles no C) |
+| `blosc` | no | Blosc 1 filter (id 32001): reads BloscLZ, LZ4/LZ4HC, Snappy, Zlib and Zstandard frames with byte or bit shuffle; writes LZ4, Snappy, Zlib or Zstandard (not BloscLZ). Pure Rust |
+| `plugin-filters` | no | All four above |
+
+Blosc2 (32026) and ZFP (32013) are not implemented: reading them fails with
+`UnsupportedFilter`, whose message names the filter. Any other filter can be
+supplied at run time with `filter_registry::register_filter` (a decoder
+closure, or a `FilterCodec` that also encodes). The facade (`clawhdf5`)
+forwards `lzf`, `bitshuffle`, `bzip2`, `blosc` and `plugin-filters`. Write
+with `DatasetBuilder::with_lzf()`, `with_bitshuffle(..)`, `with_bzip2(..)`
+and `with_blosc(..)`; h5py + hdf5plugin read the result (tested both ways in
+`crates/clawhdf5/tests/plugin_filters_interop.rs`). The pure-Rust Zstandard
+encoder has one level (about zstd's level 1); no speed or ratio claims are
+made for these codecs.
 
 ### `clawhdf5-ann`
 
@@ -782,7 +909,9 @@ clawhdf5-migrate --sqlite old.db --hdf5 memory.h5 --agent-id my-agent --embedder
 
 The output is an ordinary `clawhdf5-agent` store, written through the agent's
 own API: open it with `HDF5Memory::open` (or `clawhdf5-cli --path memory.h5 …`)
-and search it straight away. What carries over from the ZeroClaw tables:
+and search it straight away. The source must use the `memory_chunks` / `sessions` / `entities` / `relations` layout (names are
+configurable with `--*-table`); note that this is not ZeroClaw's schema, and
+ZeroClaw does not use clawhdf5. What carries over:
 
 | SQLite | Agent store |
 |--------|-------------|
@@ -821,10 +950,10 @@ See [ROADMAP.md](ROADMAP.md) for the full implementation tracker.
 - ✅ Temporal reasoning with sub-µs queries
 - ✅ Memory security + anomaly detection
 - ✅ Multi-modal memory (text/image/audio/video)
-- ✅ OpenClaw integration layer
+- ✅ Markdown ingest/export backend (`ClawhdfBackend`); an OpenClaw plugin was never built — see [docs/openclaw.md](docs/openclaw.md)
 - ✅ Comprehensive Criterion benchmarks
 
-**Phase 2** — MemoryArena and LongMemEval academic benchmarks are done (see [BENCHMARKS.md](BENCHMARKS.md), reproduced on a second machine); remaining: publish the OpenClaw TypeScript bridge to npm, crates.io/PyPI publishing.
+**Phase 2** — MemoryArena and LongMemEval academic benchmarks are done (see [BENCHMARKS.md](BENCHMARKS.md), reproduced on a second machine); remaining: crates.io/PyPI publishing. The Node bindings are unpublished and known to be broken ([known issues](docs/known-issues.md)).
 
 ---
 

ROADMAP.md

@@ -105,24 +105,30 @@
 
 ---
 
-## Track 7: OpenClaw Integration
-**Status:** 🟢 Complete
+## Track 7: OpenClaw Integration — withdrawn (2026-09-25)
+**Status:** ⚪ Withdrawn (the items below were library work; no OpenClaw integration shipped)
 **Priority:** Critical (for adoption)
 **Crates:** `clawhdf5-agent`, `clawhdf5-napi`
 
 - [x] **7.1** Memory backend trait — MemoryBackend with search/get/write/ingest/export/stats
 - [x] **7.2** Hybrid retrieval pipeline — ClawhdfBackend wires RRF → reranker → confidence rejection
 - [x] **7.3** Markdown import/export — MarkdownParser + MarkdownExporter with line tracking + metadata
-- [x] **7.4** memory_search tool — backed by full hybrid retrieval pipeline
-- [x] **7.5** memory_get tool — get() with path + line range support
+- [x] **7.4** `search()` — backed by the full hybrid retrieval pipeline (a Rust method; no OpenClaw tool was ever registered)
+- [x] **7.5** `get()` — read back by path, with a line slice (not an OpenClaw tool either)
 - [x] **7.6** Compaction integration — run_compaction() (decay + compact + WAL flush), run_consolidation() (hippocampal engine), tick_session(), flush_wal()
-- [x] **7.7** Config surface — `memory.backend = "clawhdf5"` schema documented in docs/openclaw-config.md
-- [x] **7.8** Documentation + migration guide — docs/migration-guide.md, docs/openclaw-integration.md (architecture, full API reference, code patterns)
+- [ ] **7.7** ~~Config surface — `memory.backend = "clawhdf5"`~~ — never valid OpenClaw config; docs removed
+- [ ] **7.8** ~~Documentation + migration guide~~ — removed: they described an integration that never worked
 
-**Node.js bridge:** `clawhdf5-napi` (napi-rs) → `@redclaw/clawhdf5` npm package with full TypeScript types.
+**Node.js bridge:** `clawhdf5-napi` (napi-rs) and a TypeScript wrapper in `packages/clawhdf5-node` exist but are unpublished, untested in CI and known to be broken (docs/known-issues.md).
 
 ---
 
+> **Withdrawn.** None of this track produced a working OpenClaw integration: no
+> plugin was built, the documented `memory.backend = "clawhdf5"` config was never
+> valid in any OpenClaw release, and the Node package was never published. The
+> Rust `ClawhdfBackend` remains as a library API. Not pursued for now; see
+> [docs/openclaw.md](docs/openclaw.md) for what a plugin would need today.
+
 ## Track 8: Benchmarking & Validation
 **Status:** 🟢 Complete
 **Priority:** High
@@ -142,7 +148,7 @@
 
 **Phase 1:** ~~Tracks 1, 2, 3 — core memory intelligence~~ 🟢 Complete
 **Phase 2:** ~~Track 4 (temporal) + Track 5 (security)~~ 🟢 Complete
-**Phase 3:** ~~Track 6 (multi-modal) + Track 7 (OpenClaw integration)~~ 🟢 Complete
+**Phase 3:** ~~Track 6 (multi-modal)~~ 🟢 Complete; Track 7 (OpenClaw integration) withdrawn
 **Phase 4:** ~~Track 8 (benchmarking + validation)~~ 🟢 Complete
 
 All 8 tracks delivered. 1,650+ tests passing, zero clippy warnings.

conformance/.gitignore

@@ -0,0 +1,3 @@
+/.cache/
+# pin the probe's dependencies (the workspace lock is not committed)
+!/probe/Cargo.lock

conformance/README.md

@@ -0,0 +1,39 @@
+# Conformance sweep
+
+Reads every HDF5 file of eight public corpora with clawhdf5 and with
+h5py/libhdf5, compares the two readings object by object, and writes
+[`CONFORMANCE.md`](../CONFORMANCE.md).
+
+```sh
+CLAWHDF5_PYTHON=/path/to/venv/bin/python conformance/run.sh      # ~30 s once the corpus is cached
+conformance/run.sh --update-baseline                             # after an intended change in results
+```
+
+Needs Rust, `git`, `h5dump` (Debian/Ubuntu `hdf5-tools`), `libaec` (for the
+probe's `szip` feature; `libaec-dev`), and a Python with the packages in
+`requirements.txt`. The first run downloads about 450 MB of sparse checkouts.
+
+| file | role |
+|---|---|
+| `corpus.txt` | the corpora: git URL, pinned commit, swept root, sparse-checkout patterns |
+| `fetch-corpus.sh` | shallow, sparse, blob-filtered checkout of each pinned commit into `.cache/src/` (gitignored); no-op when already there |
+| `list_files.py` | which files are probed (HDF5/netCDF-4 extensions minus netCDF classic, plus the CVE reproducers) |
+| `probe/` | the clawhdf5 side: a standalone crate (outside the workspace, so `cargo test --workspace` never builds it) that walks a file with `clawhdf5-format` and prints canonical JSON |
+| `ref.py` | the h5py side: the same JSON from h5py |
+| `run_one.sh` | runs both sides on one file (and `h5dump` on the CVE corpus) under a timeout and an address-space limit |
+| `compare.py` | classifies each file (ok / our-error / mismatch / h5py-cannot-read / panic / hang / crash / oom) and groups root causes |
+| `report.py` | writes `CONFORMANCE.md` |
+| `check.py` | the gate: fails on any panic/hang/crash/oom, on an ok count below `baseline.json`, or on a baseline-ok file that is no longer ok |
+| `baseline.json` | the ok files the gate holds the line on |
+| `requirements.txt` | pinned h5py / numpy / hdf5plugin / netCDF4 |
+
+Results for every file (both sides' JSON and stderr, `results.csv`,
+`results.json`, `summary.md`) are left in `.cache/results/`.
+
+The nightly job is `.gitea/workflows/conformance.yml`; it prints the report
+into the job log.
+
+The canonical value encoding both sides hash is documented at the top of
+`probe/src/main.rs`. Values are compared as libhdf5 presents them: a float
+with a non-IEEE bit layout (N-Bit) or an integer with a bit offset is compared
+as the converted number, not as raw file bytes.

conformance/baseline.json

@@ -0,0 +1,624 @@
+{
+ "comment": "conformance/run.sh fails if the ok count drops below `ok` or a file in `ok_files` stops being ok. Regenerate with `conformance/run.sh --update-baseline` after an intended change.",
+ "commit": "73a01f1256fb9bf1b1e7601f755af9e8273cec4e",
+ "date": "2026-09-26 14:18 UTC",
+ "reference": "h5py 3.16.0 / HDF5 2.0.0",
+ "files": 697,
+ "ok": 575,
+ "counts": {
+  "h5py-cannot-read": 92,
+  "mismatch": 20,
+  "ok": 575,
+  "our-error": 10
+ },
+ "per_corpus": {
+  "NCAS-CMS_pyfive": {
+   "mismatch": 1,
+   "ok": 32
+  },
+  "cve_hdf5": {
+   "h5py-cannot-read": 32,
+   "mismatch": 9,
+   "ok": 100,
+   "our-error": 6
+  },
+  "h5py_data": {
+   "ok": 4
+  },
+  "hdf5": {
+   "h5py-cannot-read": 60,
+   "mismatch": 10,
+   "ok": 392,
+   "our-error": 4
+  },
+  "netcdf-c": {
+   "ok": 20
+  },
+  "netcdf4-python": {
+   "ok": 18
+  },
+  "usnistgov_h5wasm": {
+   "ok": 5
+  },
+  "xarray-data": {
+   "ok": 4
+  }
+ },
+ "ok_files": [
+  "NCAS-CMS_pyfive/tests/compact.hdf5",
+  "NCAS-CMS_pyfive/tests/data/btreev2.hdf5",
+  "NCAS-CMS_pyfive/tests/data/chunked.hdf5",
+  "NCAS-CMS_pyfive/tests/data/cmip_bad_eg.nc",
+  "NCAS-CMS_pyfive/tests/data/compressed.hdf5",
+  "NCAS-CMS_pyfive/tests/data/compressed_v1.hdf5",
+  "NCAS-CMS_pyfive/tests/data/dataset_datatypes.hdf5",
+  "NCAS-CMS_pyfive/tests/data/dataset_multidim.hdf5",
+  "NCAS-CMS_pyfive/tests/data/dim_scales.hdf5",
+  "NCAS-CMS_pyfive/tests/data/earliest.hdf5",
+  "NCAS-CMS_pyfive/tests/data/enum_h5variable.hdf5",
+  "NCAS-CMS_pyfive/tests/data/enum_variable.hdf5",
+  "NCAS-CMS_pyfive/tests/data/enum_variable.nc",
+  "NCAS-CMS_pyfive/tests/data/enums_from_netcdf.nc",
+  "NCAS-CMS_pyfive/tests/data/fillvalue_earliest.hdf5",
+  "NCAS-CMS_pyfive/tests/data/fillvalue_latest.hdf5",
+  "NCAS-CMS_pyfive/tests/data/filter_pipeline_v2.hdf5",
+  "NCAS-CMS_pyfive/tests/data/fletcher32.hdf5",
+  "NCAS-CMS_pyfive/tests/data/fractal_heap_no_mci_rlat.nc",
+  "NCAS-CMS_pyfive/tests/data/groups.hdf5",
+  "NCAS-CMS_pyfive/tests/data/h5netcdf_test.hdf5",
+  "NCAS-CMS_pyfive/tests/data/issue23_A.nc",
+  "NCAS-CMS_pyfive/tests/data/issue23_A_contiguous.nc",
+  "NCAS-CMS_pyfive/tests/data/issue23_B.nc",
+  "NCAS-CMS_pyfive/tests/data/latest.hdf5",
+  "NCAS-CMS_pyfive/tests/data/netcdf4_classic.nc",
+  "NCAS-CMS_pyfive/tests/data/new_style_groups.hdf5",
+  "NCAS-CMS_pyfive/tests/data/noy_AERmonZ_UKESM1-0-LL_piControl_r1i1p1f2_gnz_200001-200012.nc",
+  "NCAS-CMS_pyfive/tests/data/references.hdf5",
+  "NCAS-CMS_pyfive/tests/data/resizable.hdf5",
+  "NCAS-CMS_pyfive/tests/opaque_datetime.hdf5",
+  "NCAS-CMS_pyfive/tests/opaque_fixed.hdf5",
+  "cve_hdf5/cvefiles/cve-2016-4330.h5",
+  "cve_hdf5/cvefiles/cve-2016-4331.h5",
+  "cve_hdf5/cvefiles/cve-2016-4332-mtime-new.h5",
+  "cve_hdf5/cvefiles/cve-2016-4332-mtime.h5",
+  "cve_hdf5/cvefiles/cve-2016-4333.h5",
+  "cve_hdf5/cvefiles/cve-2017-17505.h5",
+  "cve_hdf5/cvefiles/cve-2017-17506.h5",
+  "cve_hdf5/cvefiles/cve-2017-17507.h5",
+  "cve_hdf5/cvefiles/cve-2017-17508.h5",
+  "cve_hdf5/cvefiles/cve-2017-17509.h5",
+  "cve_hdf5/cvefiles/cve-2018-11202.h5",
+  "cve_hdf5/cvefiles/cve-2018-11203.h5",
+  "cve_hdf5/cvefiles/cve-2018-11204.h5",
+  "cve_hdf5/cvefiles/cve-2018-11205.h5",
+  "cve_hdf5/cvefiles/cve-2018-11206-new.h5",
+  "cve_hdf5/cvefiles/cve-2018-11206-old.h5",
+  "cve_hdf5/cvefiles/cve-2018-11207.h5",
+  "cve_hdf5/cvefiles/cve-2018-13867.h5",
+  "cve_hdf5/cvefiles/cve-2018-13868.h5",
+  "cve_hdf5/cvefiles/cve-2018-13869.h5",
+  "cve_hdf5/cvefiles/cve-2018-13870.h5",
+  "cve_hdf5/cvefiles/cve-2018-13871.h5",
+  "cve_hdf5/cvefiles/cve-2018-13872.h5",
+  "cve_hdf5/cvefiles/cve-2018-13873.h5",
+  "cve_hdf5/cvefiles/cve-2018-13875.h5",
+  "cve_hdf5/cvefiles/cve-2018-14031.h5",
+  "cve_hdf5/cvefiles/cve-2018-14033.h5",
+  "cve_hdf5/cvefiles/cve-2018-14034.h5",
+  "cve_hdf5/cvefiles/cve-2018-14035.h5",
+  "cve_hdf5/cvefiles/cve-2018-14460.h5",
+  "cve_hdf5/cvefiles/cve-2018-15671.h5",
+  "cve_hdf5/cvefiles/cve-2018-15672.h5",
+  "cve_hdf5/cvefiles/cve-2018-16438.h5",
+  "cve_hdf5/cvefiles/cve-2018-17233.h5",
+  "cve_hdf5/cvefiles/cve-2018-17234.h5",
+  "cve_hdf5/cvefiles/cve-2018-17237.h5",
+  "cve_hdf5/cvefiles/cve-2018-17432.h5",
+  "cve_hdf5/cvefiles/cve-2018-17434.h5",
+  "cve_hdf5/cvefiles/cve-2018-17435.h5",
+  "cve_hdf5/cvefiles/cve-2018-17437.h5",
+  "cve_hdf5/cvefiles/cve-2019-8396.h5",
+  "cve_hdf5/cvefiles/cve-2019-9152.h5",
+  "cve_hdf5/cvefiles/cve-2020-10811.h5",
+  "cve_hdf5/cvefiles/cve-2020-18232.h5",
+  "cve_hdf5/cvefiles/cve-2021-36977.h5",
+  "cve_hdf5/cvefiles/cve-2021-37501.h5",
+  "cve_hdf5/cvefiles/cve-2021-45829.h5",
+  "cve_hdf5/cvefiles/cve-2021-45833.h5",
+  "cve_hdf5/cvefiles/cve-2024-29157.h5",
+  "cve_hdf5/cvefiles/cve-2024-29158.h5",
+  "cve_hdf5/cvefiles/cve-2024-29159.h5",
+  "cve_hdf5/cvefiles/cve-2024-29160.h5",
+  "cve_hdf5/cvefiles/cve-2024-29161.h5",
+  "cve_hdf5/cvefiles/cve-2024-29162.h5",
+  "cve_hdf5/cvefiles/cve-2024-29163.h5",
+  "cve_hdf5/cvefiles/cve-2024-29164.h5",
+  "cve_hdf5/cvefiles/cve-2024-29165.h5",
+  "cve_hdf5/cvefiles/cve-2024-29166.h5",
+  "cve_hdf5/cvefiles/cve-2024-32605.h5",
+  "cve_hdf5/cvefiles/cve-2024-32606.h5",
+  "cve_hdf5/cvefiles/cve-2024-32607-1.h5",
+  "cve_hdf5/cvefiles/cve-2024-32607-2.h5",
+  "cve_hdf5/cvefiles/cve-2024-32608.h5",
+  "cve_hdf5/cvefiles/cve-2024-32610.h5",
+  "cve_hdf5/cvefiles/cve-2024-32611.h5",
+  "cve_hdf5/cvefiles/cve-2024-32612.h5",
+  "cve_hdf5/cvefiles/cve-2024-32613.h5",
+  "cve_hdf5/cvefiles/cve-2024-32614.h5",
+  "cve_hdf5/cvefiles/cve-2024-32615.h5",
+  "cve_hdf5/cvefiles/cve-2024-32616.h5",
+  "cve_hdf5/cvefiles/cve-2024-32617.h5",
+  "cve_hdf5/cvefiles/cve-2024-32619.h5",
+  "cve_hdf5/cvefiles/cve-2024-32620.h5",
+  "cve_hdf5/cvefiles/cve-2024-32621.h5",
+  "cve_hdf5/cvefiles/cve-2024-32622.h5",
+  "cve_hdf5/cvefiles/cve-2024-32624.h5",
+  "cve_hdf5/cvefiles/cve-2024-33873.h5",
+  "cve_hdf5/cvefiles/cve-2024-33875.h5",
+  "cve_hdf5/cvefiles/cve-2024-33876.h5",
+  "cve_hdf5/cvefiles/cve-2024-33877.h5",
+  "cve_hdf5/cvefiles/cve-2025-2310.h5",
+  "cve_hdf5/cvefiles/cve-2025-2924.h5",
+  "cve_hdf5/cvefiles/cve-2025-2925.h5",
+  "cve_hdf5/cvefiles/cve-2025-6269-1.h5",
+  "cve_hdf5/cvefiles/cve-2025-6269-2.h5",
+  "cve_hdf5/cvefiles/cve-2025-6269-3.h5",
+  "cve_hdf5/cvefiles/cve-2025-6269-4.h5",
+  "cve_hdf5/cvefiles/cve-2025-6516.h5",
+  "cve_hdf5/cvefiles/cve-2025-6857.h5",
+  "cve_hdf5/cvefiles/cve-2025-7067.h5",
+  "cve_hdf5/cvefiles/cve-2026-26200.h5",
+  "cve_hdf5/cvefiles/cve-2026-34734.h5",
+  "cve_hdf5/cvefiles/cve-2026-92627.h5",
+  "cve_hdf5/cvefiles/unknown-1.h5",
+  "cve_hdf5/fuzzerfiles/gh-4431-poc-03.h5",
+  "cve_hdf5/fuzzerfiles/gh-4432-poc-05.h5",
+  "cve_hdf5/fuzzerfiles/gh-4433-poc-08.h5",
+  "cve_hdf5/fuzzerfiles/gh-4435-poc-10.h5",
+  "cve_hdf5/fuzzerfiles/gh_2649_flawed.h5",
+  "cve_hdf5/fuzzerfiles/gh_2649_plain_model.h5",
+  "h5py_data/compound-dtype-complex.h5",
+  "h5py_data/vlen_string_dset.h5",
+  "h5py_data/vlen_string_dset_utc.h5",
+  "h5py_data/vlen_string_s390x.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_bitgroom.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_blosc.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_bshuf.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_bzip2.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_granularbr.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_jpeg.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_lz4.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_lzf.h5",
+  "hdf5/HDF5Examples/C/H5FLT/tfiles/h5ex_d_zstd.h5",
+  "hdf5/HDF5Examples/C/H5G/16/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/C/H5G/16/h5ex_g_traverse.h5",
+  "hdf5/HDF5Examples/C/H5G/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/C/H5G/h5ex_g_traverse.h5",
+  "hdf5/HDF5Examples/C/H5G/h5ex_g_visit.h5",
+  "hdf5/HDF5Examples/FORTRAN/H5G/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/FORTRAN/H5G/h5ex_g_traverse.h5",
+  "hdf5/HDF5Examples/FORTRAN/H5G/h5ex_g_visit.h5",
+  "hdf5/HDF5Examples/JAVA/H5G/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/JAVA/H5G/h5ex_g_visit.h5",
+  "hdf5/HDF5Examples/JAVA/compat/H5G/110/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/JAVA/compat/H5G/110/h5ex_g_visit.h5",
+  "hdf5/HDF5Examples/JAVA/compat/H5G/h5ex_g_iterate.h5",
+  "hdf5/HDF5Examples/JAVA/compat/H5G/h5ex_g_visit.h5",
+  "hdf5/c++/test/th5s.h5",
+  "hdf5/hl/test/testfiles/test_ds_be.h5",
+  "hdf5/hl/test/testfiles/test_ds_be_new_ref-32bit.h5",
+  "hdf5/hl/test/testfiles/test_ds_be_new_ref.h5",
+  "hdf5/hl/test/testfiles/test_ds_le.h5",
+  "hdf5/hl/test/testfiles/test_ds_le_new_ref.h5",
+  "hdf5/hl/test/testfiles/test_ld.h5",
+  "hdf5/hl/test/testfiles/test_table_be.h5",
+  "hdf5/hl/test/testfiles/test_table_cray.h5",
+  "hdf5/hl/test/testfiles/test_table_le.h5",
+  "hdf5/test/testfiles/aggr.h5",
+  "hdf5/test/testfiles/bad_chunk_ndims.h5",
+  "hdf5/test/testfiles/bad_compound.h5",
+  "hdf5/test/testfiles/bad_offset.h5",
+  "hdf5/test/testfiles/be_data.h5",
+  "hdf5/test/testfiles/be_extlink1.h5",
+  "hdf5/test/testfiles/be_extlink2.h5",
+  "hdf5/test/testfiles/btree_idx_1_6.h5",
+  "hdf5/test/testfiles/btree_idx_1_8.h5",
+  "hdf5/test/testfiles/charsets.h5",
+  "hdf5/test/testfiles/corrupt_stab_msg.h5",
+  "hdf5/test/testfiles/deflate.h5",
+  "hdf5/test/testfiles/file_image_core_test.h5",
+  "hdf5/test/testfiles/filespace_1_6.h5",
+  "hdf5/test/testfiles/filespace_1_8.h5",
+  "hdf5/test/testfiles/fill18.h5",
+  "hdf5/test/testfiles/fill_old.h5",
+  "hdf5/test/testfiles/filter_error.h5",
+  "hdf5/test/testfiles/fsm_aggr_nopersist.h5",
+  "hdf5/test/testfiles/fsm_aggr_persist.h5",
+  "hdf5/test/testfiles/group_old.h5",
+  "hdf5/test/testfiles/h5fc_ext1_f.h5",
+  "hdf5/test/testfiles/h5fc_ext1_i.h5",
+  "hdf5/test/testfiles/h5fc_ext2_if.h5",
+  "hdf5/test/testfiles/h5fc_ext2_sf.h5",
+  "hdf5/test/testfiles/h5fc_ext3_isf.h5",
+  "hdf5/test/testfiles/h5fc_ext_none.h5",
+  "hdf5/test/testfiles/le_data.h5",
+  "hdf5/test/testfiles/le_extlink1.h5",
+  "hdf5/test/testfiles/le_extlink2.h5",
+  "hdf5/test/testfiles/memleak_H5O_dtype_decode_helper_H5Odtype.h5",
+  "hdf5/test/testfiles/mergemsg.h5",
+  "hdf5/test/testfiles/noencoder.h5",
+  "hdf5/test/testfiles/none.h5",
+  "hdf5/test/testfiles/paged_nopersist.h5",
+  "hdf5/test/testfiles/paged_persist.h5",
+  "hdf5/test/testfiles/specmetaread.h5",
+  "hdf5/test/testfiles/tarrold.h5",
+  "hdf5/test/testfiles/tbad_msg_count.h5",
+  "hdf5/test/testfiles/tbogus.h5",
+  "hdf5/test/testfiles/test_filters_be.h5",
+  "hdf5/test/testfiles/test_filters_le.h5",
+  "hdf5/test/testfiles/th5s.h5",
+  "hdf5/test/testfiles/tlayouto.h5",
+  "hdf5/test/testfiles/tmisc38a.h5",
+  "hdf5/test/testfiles/tmisc38b.h5",
+  "hdf5/test/testfiles/tmtimen.h5",
+  "hdf5/test/testfiles/tmtimeo.h5",
+  "hdf5/test/testfiles/tnullspace.h5",
+  "hdf5/test/testfiles/tsizeslheap.h5",
+  "hdf5/tools/test/testfiles/bigendian/tdset2.h5",
+  "hdf5/tools/test/testfiles/binfp64.h5",
+  "hdf5/tools/test/testfiles/binin16.h5",
+  "hdf5/tools/test/testfiles/binin32.h5",
+  "hdf5/tools/test/testfiles/binin8.h5",
+  "hdf5/tools/test/testfiles/binin8w.h5",
+  "hdf5/tools/test/testfiles/binuin16.h5",
+  "hdf5/tools/test/testfiles/binuin32.h5",
+  "hdf5/tools/test/testfiles/bounds_latest_latest.h5",
+  "hdf5/tools/test/testfiles/charsets.h5",
+  "hdf5/tools/test/testfiles/compounds_array_vlen1.h5",
+  "hdf5/tools/test/testfiles/compounds_array_vlen2.h5",
+  "hdf5/tools/test/testfiles/err_attr_dspace.h5",
+  "hdf5/tools/test/testfiles/file_space.h5",
+  "hdf5/tools/test/testfiles/filter_fail.h5",
+  "hdf5/tools/test/testfiles/h5clear_fsm_persist_equal.h5",
+  "hdf5/tools/test/testfiles/h5clear_fsm_persist_less.h5",
+  "hdf5/tools/test/testfiles/h5clear_fsm_persist_noclose.h5",
+  "hdf5/tools/test/testfiles/h5clear_fsm_persist_user_equal.h5",
+  "hdf5/tools/test/testfiles/h5clear_fsm_persist_user_less.h5",
+  "hdf5/tools/test/testfiles/h5clear_sec2_v0.h5",
+  "hdf5/tools/test/testfiles/h5clear_sec2_v2.h5",
+  "hdf5/tools/test/testfiles/h5copy_extlinks_src.h5",
+  "hdf5/tools/test/testfiles/h5copy_extlinks_trg.h5",
+  "hdf5/tools/test/testfiles/h5copy_ref.h5",
+  "hdf5/tools/test/testfiles/h5copytst.h5",
+  "hdf5/tools/test/testfiles/h5copytst_new.h5",
+  "hdf5/tools/test/testfiles/h5diff_attr1.h5",
+  "hdf5/tools/test/testfiles/h5diff_attr2.h5",
+  "hdf5/tools/test/testfiles/h5diff_attr3.h5",
+  "hdf5/tools/test/testfiles/h5diff_attr_v_level1.h5",
+  "hdf5/tools/test/testfiles/h5diff_attr_v_level2.h5",
+  "hdf5/tools/test/testfiles/h5diff_basic1.h5",
+  "hdf5/tools/test/testfiles/h5diff_basic2.h5",
+  "hdf5/tools/test/testfiles/h5diff_comp_vl_strs.h5",
+  "hdf5/tools/test/testfiles/h5diff_danglelinks1.h5",
+  "hdf5/tools/test/testfiles/h5diff_danglelinks2.h5",
+  "hdf5/tools/test/testfiles/h5diff_dset1.h5",
+  "hdf5/tools/test/testfiles/h5diff_dset2.h5",
+  "hdf5/tools/test/testfiles/h5diff_dset3.h5",
+  "hdf5/tools/test/testfiles/h5diff_dset_zero_dim_size1.h5",
+  "hdf5/tools/test/testfiles/h5diff_dset_zero_dim_size2.h5",
+  "hdf5/tools/test/testfiles/h5diff_dtypes.h5",
+  "hdf5/tools/test/testfiles/h5diff_empty.h5",
+  "hdf5/tools/test/testfiles/h5diff_enum_invalid_values.h5",
+  "hdf5/tools/test/testfiles/h5diff_eps1.h5",
+  "hdf5/tools/test/testfiles/h5diff_eps2.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude1-1.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude1-2.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude2-1.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude2-2.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude3-1.h5",
+  "hdf5/tools/test/testfiles/h5diff_exclude3-2.h5",
+  "hdf5/tools/test/testfiles/h5diff_ext2softlink_src.h5",
+  "hdf5/tools/test/testfiles/h5diff_ext2softlink_trg.h5",
+  "hdf5/tools/test/testfiles/h5diff_extlink_src.h5",
+  "hdf5/tools/test/testfiles/h5diff_extlink_trg.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse1.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse2.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse_ext1.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse_ext2-1.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse_ext2-2.h5",
+  "hdf5/tools/test/testfiles/h5diff_grp_recurse_ext2-3.h5",
+  "hdf5/tools/test/testfiles/h5diff_hyper1.h5",
+  "hdf5/tools/test/testfiles/h5diff_hyper2.h5",
+  "hdf5/tools/test/testfiles/h5diff_linked_softlink.h5",
+  "hdf5/tools/test/testfiles/h5diff_links.h5",
+  "hdf5/tools/test/testfiles/h5diff_onion_dset_1d.h5",
+  "hdf5/tools/test/testfiles/h5diff_onion_dset_ext.h5",
+  "hdf5/tools/test/testfiles/h5diff_onion_objs.h5",
+  "hdf5/tools/test/testfiles/h5diff_softlinks.h5",
+  "hdf5/tools/test/testfiles/h5diff_strings1.h5",
+  "hdf5/tools/test/testfiles/h5diff_strings2.h5",
+  "hdf5/tools/test/testfiles/h5fc_edge_v3.h5",
+  "hdf5/tools/test/testfiles/h5fc_err_level.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext1_f.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext1_i.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext1_s.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext2_if.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext2_is.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext2_sf.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext3_isf.h5",
+  "hdf5/tools/test/testfiles/h5fc_ext_none.h5",
+  "hdf5/tools/test/testfiles/h5fc_non_v3.h5",
+  "hdf5/tools/test/testfiles/h5repack_CVE-2018-14460.h5",
+  "hdf5/tools/test/testfiles/h5repack_CVE-2018-17432.h5",
+  "hdf5/tools/test/testfiles/h5repack_aggr.h5",
+  "hdf5/tools/test/testfiles/h5repack_attr.h5",
+  "hdf5/tools/test/testfiles/h5repack_attr_refs.h5",
+  "hdf5/tools/test/testfiles/h5repack_deflate.h5",
+  "hdf5/tools/test/testfiles/h5repack_early.h5",
+  "hdf5/tools/test/testfiles/h5repack_ext.h5",
+  "hdf5/tools/test/testfiles/h5repack_f32le.h5",
+  "hdf5/tools/test/testfiles/h5repack_f32le_ex.h5",
+  "hdf5/tools/test/testfiles/h5repack_fill.h5",
+  "hdf5/tools/test/testfiles/h5repack_filters.h5",
+  "hdf5/tools/test/testfiles/h5repack_fletcher.h5",
+  "hdf5/tools/test/testfiles/h5repack_fsm_aggr_nopersist.h5",
+  "hdf5/tools/test/testfiles/h5repack_fsm_aggr_persist.h5",
+  "hdf5/tools/test/testfiles/h5repack_hlink.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_1d.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_1d_ex.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_2d.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_2d_ex.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_3d.h5",
+  "hdf5/tools/test/testfiles/h5repack_int32le_3d_ex.h5",
+  "hdf5/tools/test/testfiles/h5repack_layout.UD.h5",
+  "hdf5/tools/test/testfiles/h5repack_layout.h5",
+  "hdf5/tools/test/testfiles/h5repack_layout2.h5",
+  "hdf5/tools/test/testfiles/h5repack_layout3.h5",
+  "hdf5/tools/test/testfiles/h5repack_layouto.h5",
+  "hdf5/tools/test/testfiles/h5repack_named_dtypes.h5",
+  "hdf5/tools/test/testfiles/h5repack_nbit.h5",
+  "hdf5/tools/test/testfiles/h5repack_nested_8bit_enum.h5",
+  "hdf5/tools/test/testfiles/h5repack_nested_8bit_enum_deflated.h5",
+  "hdf5/tools/test/testfiles/h5repack_none.h5",
+  "hdf5/tools/test/testfiles/h5repack_objs.h5",
+  "hdf5/tools/test/testfiles/h5repack_paged_nopersist.h5",
+  "hdf5/tools/test/testfiles/h5repack_paged_persist.h5",
+  "hdf5/tools/test/testfiles/h5repack_refs.h5",
+  "hdf5/tools/test/testfiles/h5repack_shuffle.h5",
+  "hdf5/tools/test/testfiles/h5repack_soffset.h5",
+  "hdf5/tools/test/testfiles/h5repack_szip.h5",
+  "hdf5/tools/test/testfiles/h5repack_uint8be.h5",
+  "hdf5/tools/test/testfiles/h5repack_uint8be_ex.h5",
+  "hdf5/tools/test/testfiles/h5stat_err_old_fill.h5",
+  "hdf5/tools/test/testfiles/h5stat_err_old_layout.h5",
+  "hdf5/tools/test/testfiles/h5stat_err_refcount.h5",
+  "hdf5/tools/test/testfiles/h5stat_filters.h5",
+  "hdf5/tools/test/testfiles/h5stat_idx.h5",
+  "hdf5/tools/test/testfiles/h5stat_newgrat.h5",
+  "hdf5/tools/test/testfiles/h5stat_threshold.h5",
+  "hdf5/tools/test/testfiles/h5stat_tsohm.h5",
+  "hdf5/tools/test/testfiles/mod_h5clear_mdc_image.h5",
+  "hdf5/tools/test/testfiles/non_comparables1.h5",
+  "hdf5/tools/test/testfiles/non_comparables2.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext1_f.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext1_i.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext1_s.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext2_if.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext2_is.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext2_sf.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext3_isf.h5",
+  "hdf5/tools/test/testfiles/old_h5fc_ext_none.h5",
+  "hdf5/tools/test/testfiles/packedbits.h5",
+  "hdf5/tools/test/testfiles/t128bit_float.h5",
+  "hdf5/tools/test/testfiles/tCVE-2021-37501_attr_decode.h5",
+  "hdf5/tools/test/testfiles/tCVE_2018_11206_fill_new.h5",
+  "hdf5/tools/test/testfiles/tCVE_2018_11206_fill_old.h5",
+  "hdf5/tools/test/testfiles/taindices.h5",
+  "hdf5/tools/test/testfiles/tarray1.h5",
+  "hdf5/tools/test/testfiles/tarray1_big.h5",
+  "hdf5/tools/test/testfiles/tarray2.h5",
+  "hdf5/tools/test/testfiles/tarray4.h5",
+  "hdf5/tools/test/testfiles/tarray5.h5",
+  "hdf5/tools/test/testfiles/tarray8.h5",
+  "hdf5/tools/test/testfiles/tattr.h5",
+  "hdf5/tools/test/testfiles/tattr2.h5",
+  "hdf5/tools/test/testfiles/tattr4_be.h5",
+  "hdf5/tools/test/testfiles/tattrintsize.h5",
+  "hdf5/tools/test/testfiles/tattrreg.h5",
+  "hdf5/tools/test/testfiles/tbfloat16.h5",
+  "hdf5/tools/test/testfiles/tbfloat16_be.h5",
+  "hdf5/tools/test/testfiles/tbigdims.h5",
+  "hdf5/tools/test/testfiles/tbinary.h5",
+  "hdf5/tools/test/testfiles/tbitnopaque.h5",
+  "hdf5/tools/test/testfiles/tchar.h5",
+  "hdf5/tools/test/testfiles/tcmpdattrintsize.h5",
+  "hdf5/tools/test/testfiles/tcmpdintarray.h5",
+  "hdf5/tools/test/testfiles/tcmpdints.h5",
+  "hdf5/tools/test/testfiles/tcmpdintsize.h5",
+  "hdf5/tools/test/testfiles/tcomplex.h5",
+  "hdf5/tools/test/testfiles/tcompound.h5",
+  "hdf5/tools/test/testfiles/tcompound_complex.h5",
+  "hdf5/tools/test/testfiles/tcompound_complex2.h5",
+  "hdf5/tools/test/testfiles/tdatareg.h5",
+  "hdf5/tools/test/testfiles/tdset.h5",
+  "hdf5/tools/test/testfiles/tdset2.h5",
+  "hdf5/tools/test/testfiles/tdset_idx.h5",
+  "hdf5/tools/test/testfiles/tempty.h5",
+  "hdf5/tools/test/testfiles/textlink.h5",
+  "hdf5/tools/test/testfiles/textlinkfar.h5",
+  "hdf5/tools/test/testfiles/textlinksrc.h5",
+  "hdf5/tools/test/testfiles/textlinktar.h5",
+  "hdf5/tools/test/testfiles/textpfe.h5",
+  "hdf5/tools/test/testfiles/tfcontents2.h5",
+  "hdf5/tools/test/testfiles/tfilters.h5",
+  "hdf5/tools/test/testfiles/tfloat16.h5",
+  "hdf5/tools/test/testfiles/tfloat16_be.h5",
+  "hdf5/tools/test/testfiles/tfloat4.h5",
+  "hdf5/tools/test/testfiles/tfloat6.h5",
+  "hdf5/tools/test/testfiles/tfloat8.h5",
+  "hdf5/tools/test/testfiles/tfloatsattrs.h5",
+  "hdf5/tools/test/testfiles/tfpformat.h5",
+  "hdf5/tools/test/testfiles/tfvalues.h5",
+  "hdf5/tools/test/testfiles/tgroup.h5",
+  "hdf5/tools/test/testfiles/tgrp_comments.h5",
+  "hdf5/tools/test/testfiles/tgrpnullspace.h5",
+  "hdf5/tools/test/testfiles/thlink.h5",
+  "hdf5/tools/test/testfiles/thyperslab.h5",
+  "hdf5/tools/test/testfiles/tintascii.h5",
+  "hdf5/tools/test/testfiles/tints4dims.h5",
+  "hdf5/tools/test/testfiles/tintsattrs.h5",
+  "hdf5/tools/test/testfiles/tintsnodata.h5",
+  "hdf5/tools/test/testfiles/tlarge_objname.h5",
+  "hdf5/tools/test/testfiles/tldouble.h5",
+  "hdf5/tools/test/testfiles/tldouble_scalar.h5",
+  "hdf5/tools/test/testfiles/tlonglinks.h5",
+  "hdf5/tools/test/testfiles/tloop.h5",
+  "hdf5/tools/test/testfiles/tnamed_dtype_attr.h5",
+  "hdf5/tools/test/testfiles/tnestedcmpddt.h5",
+  "hdf5/tools/test/testfiles/tnestedcomp.h5",
+  "hdf5/tools/test/testfiles/tno-subset.h5",
+  "hdf5/tools/test/testfiles/tnullspace.h5",
+  "hdf5/tools/test/testfiles/torderattr.h5",
+  "hdf5/tools/test/testfiles/tordergr.h5",
+  "hdf5/tools/test/testfiles/trefer_attr.h5",
+  "hdf5/tools/test/testfiles/trefer_compat.h5",
+  "hdf5/tools/test/testfiles/trefer_ext1.h5",
+  "hdf5/tools/test/testfiles/trefer_ext2.h5",
+  "hdf5/tools/test/testfiles/trefer_grp.h5",
+  "hdf5/tools/test/testfiles/trefer_obj.h5",
+  "hdf5/tools/test/testfiles/trefer_obj_del.h5",
+  "hdf5/tools/test/testfiles/trefer_param.h5",
+  "hdf5/tools/test/testfiles/trefer_reg.h5",
+  "hdf5/tools/test/testfiles/trefer_reg_1d.h5",
+  "hdf5/tools/test/testfiles/tsaf.h5",
+  "hdf5/tools/test/testfiles/tscalarattrintsize.h5",
+  "hdf5/tools/test/testfiles/tscalarintattrsize.h5",
+  "hdf5/tools/test/testfiles/tscalarintsize.h5",
+  "hdf5/tools/test/testfiles/tscalarstring.h5",
+  "hdf5/tools/test/testfiles/tslink.h5",
+  "hdf5/tools/test/testfiles/tsoftlinks.h5",
+  "hdf5/tools/test/testfiles/tst_onion_dset_1d.h5",
+  "hdf5/tools/test/testfiles/tst_onion_dset_ext.h5",
+  "hdf5/tools/test/testfiles/tst_onion_objs.h5",
+  "hdf5/tools/test/testfiles/tstr.h5",
+  "hdf5/tools/test/testfiles/tstr2.h5",
+  "hdf5/tools/test/testfiles/tstr3.h5",
+  "hdf5/tools/test/testfiles/tudfilter.h5",
+  "hdf5/tools/test/testfiles/tudfilter2.h5",
+  "hdf5/tools/test/testfiles/tvldtypes1.h5",
+  "hdf5/tools/test/testfiles/tvldtypes2.h5",
+  "hdf5/tools/test/testfiles/tvldtypes3.h5",
+  "hdf5/tools/test/testfiles/tvldtypes4.h5",
+  "hdf5/tools/test/testfiles/tvldtypes5.h5",
+  "hdf5/tools/test/testfiles/tvlenstr_array.h5",
+  "hdf5/tools/test/testfiles/tvlstr.h5",
+  "hdf5/tools/test/testfiles/tvms.h5",
+  "hdf5/tools/test/testfiles/txtfp32.h5",
+  "hdf5/tools/test/testfiles/txtfp64.h5",
+  "hdf5/tools/test/testfiles/txtin16.h5",
+  "hdf5/tools/test/testfiles/txtin32.h5",
+  "hdf5/tools/test/testfiles/txtin8.h5",
+  "hdf5/tools/test/testfiles/txtstr.h5",
+  "hdf5/tools/test/testfiles/txtuin16.h5",
+  "hdf5/tools/test/testfiles/txtuin32.h5",
+  "hdf5/tools/test/testfiles/vds/1_a.h5",
+  "hdf5/tools/test/testfiles/vds/1_b.h5",
+  "hdf5/tools/test/testfiles/vds/1_c.h5",
+  "hdf5/tools/test/testfiles/vds/1_d.h5",
+  "hdf5/tools/test/testfiles/vds/1_e.h5",
+  "hdf5/tools/test/testfiles/vds/1_f.h5",
+  "hdf5/tools/test/testfiles/vds/1_vds.h5",
+  "hdf5/tools/test/testfiles/vds/2_a.h5",
+  "hdf5/tools/test/testfiles/vds/2_b.h5",
+  "hdf5/tools/test/testfiles/vds/2_c.h5",
+  "hdf5/tools/test/testfiles/vds/2_d.h5",
+  "hdf5/tools/test/testfiles/vds/2_e.h5",
+  "hdf5/tools/test/testfiles/vds/2_vds.h5",
+  "hdf5/tools/test/testfiles/vds/3_1_vds.h5",
+  "hdf5/tools/test/testfiles/vds/3_2_vds.h5",
+  "hdf5/tools/test/testfiles/vds/4_0.h5",
+  "hdf5/tools/test/testfiles/vds/4_1.h5",
+  "hdf5/tools/test/testfiles/vds/4_2.h5",
+  "hdf5/tools/test/testfiles/vds/4_vds.h5",
+  "hdf5/tools/test/testfiles/vds/5_a.h5",
+  "hdf5/tools/test/testfiles/vds/5_b.h5",
+  "hdf5/tools/test/testfiles/vds/5_c.h5",
+  "hdf5/tools/test/testfiles/vds/5_vds.h5",
+  "hdf5/tools/test/testfiles/vds/a.h5",
+  "hdf5/tools/test/testfiles/vds/b.h5",
+  "hdf5/tools/test/testfiles/vds/c.h5",
+  "hdf5/tools/test/testfiles/vds/d.h5",
+  "hdf5/tools/test/testfiles/vds/f-0.h5",
+  "hdf5/tools/test/testfiles/vds/f-3.h5",
+  "hdf5/tools/test/testfiles/vds/vds-eiger.h5",
+  "hdf5/tools/test/testfiles/vds/vds-percival-unlim-maxmin.h5",
+  "hdf5/tools/test/testfiles/xml/tbitfields.h5",
+  "hdf5/tools/test/testfiles/xml/tcompound2.h5",
+  "hdf5/tools/test/testfiles/xml/tdset2.h5",
+  "hdf5/tools/test/testfiles/xml/tenum.h5",
+  "hdf5/tools/test/testfiles/xml/test35.nc",
+  "hdf5/tools/test/testfiles/xml/tloop2.h5",
+  "hdf5/tools/test/testfiles/xml/tname-amp.h5",
+  "hdf5/tools/test/testfiles/xml/tname-apos.h5",
+  "hdf5/tools/test/testfiles/xml/tname-gt.h5",
+  "hdf5/tools/test/testfiles/xml/tname-lt.h5",
+  "hdf5/tools/test/testfiles/xml/tname-quot.h5",
+  "hdf5/tools/test/testfiles/xml/tname-sp.h5",
+  "hdf5/tools/test/testfiles/xml/tnodata.h5",
+  "hdf5/tools/test/testfiles/xml/tobjref.h5",
+  "hdf5/tools/test/testfiles/xml/topaque.h5",
+  "hdf5/tools/test/testfiles/xml/tref-escapes-at.h5",
+  "hdf5/tools/test/testfiles/xml/tref-escapes.h5",
+  "hdf5/tools/test/testfiles/xml/tref.h5",
+  "hdf5/tools/test/testfiles/xml/tstring-at.h5",
+  "hdf5/tools/test/testfiles/xml/tstring.h5",
+  "hdf5/tools/test/testfiles/zerodim.h5",
+  "netcdf-c/h5_test/ref_tst_h_compounds.h5",
+  "netcdf-c/h5_test/ref_tst_h_compounds2.h5",
+  "netcdf-c/nc_test4/ref_hdf5_compat1.nc",
+  "netcdf-c/nc_test4/ref_hdf5_compat2.nc",
+  "netcdf-c/nc_test4/ref_hdf5_compat3.nc",
+  "netcdf-c/nc_test4/ref_szip.h5",
+  "netcdf-c/nc_test4/ref_tst_compounds.nc",
+  "netcdf-c/nc_test4/ref_tst_dims.nc",
+  "netcdf-c/nc_test4/ref_tst_interops4.nc",
+  "netcdf-c/nc_test4/ref_tst_xplatform2_1.nc",
+  "netcdf-c/nc_test4/ref_tst_xplatform2_2.nc",
+  "netcdf-c/nc_test4/tdset.h5",
+  "netcdf-c/ncdump/ref_nc_test_netcdf4_4_0.nc",
+  "netcdf-c/ncdump/ref_no_ncproperty.nc",
+  "netcdf-c/ncdump/ref_provenance_v1.nc",
+  "netcdf-c/ncdump/ref_test_corrupt_magic.nc",
+  "netcdf-c/ncdump/ref_tst_compounds2.nc",
+  "netcdf-c/ncdump/ref_tst_compounds3.nc",
+  "netcdf-c/ncdump/ref_tst_compounds4.nc",
+  "netcdf-c/ncdump/ref_tst_irish_rover.nc",
+  "netcdf4-python/examples/data/prmsl.2000.nc",
+  "netcdf4-python/examples/data/prmsl.2001.nc",
+  "netcdf4-python/examples/data/prmsl.2002.nc",
+  "netcdf4-python/examples/data/prmsl.2003.nc",
+  "netcdf4-python/examples/data/prmsl.2004.nc",
+  "netcdf4-python/examples/data/prmsl.2005.nc",
+  "netcdf4-python/examples/data/prmsl.2006.nc",
+  "netcdf4-python/examples/data/prmsl.2007.nc",
+  "netcdf4-python/examples/data/prmsl.2008.nc",
+  "netcdf4-python/examples/data/prmsl.2009.nc",
+  "netcdf4-python/examples/data/prmsl.2010.nc",
+  "netcdf4-python/examples/data/prmsl.2011.nc",
+  "netcdf4-python/examples/data/rtofs_glo_3dz_f006_6hrly_reg3.nc",
+  "netcdf4-python/test/20171025_2056.Cloud_Top_Height.nc",
+  "netcdf4-python/test/issue1152.nc",
+  "netcdf4-python/test/issue671.nc",
+  "netcdf4-python/test/issue672.nc",
+  "netcdf4-python/test/test_gold.nc",
+  "usnistgov_h5wasm/test/array.h5",
+  "usnistgov_h5wasm/test/compressed.h5",
+  "usnistgov_h5wasm/test/empty.h5",
+  "usnistgov_h5wasm/test/float16.h5",
+  "usnistgov_h5wasm/test/vlen.h5",
+  "xarray-data/ROMS_example.nc",
+  "xarray-data/basin_mask.nc",
+  "xarray-data/imerghh_730.hdf5",
+  "xarray-data/precipitation.nc4"
+ ]
+}

conformance/check.py

@@ -0,0 +1,88 @@
+#!/usr/bin/env python3
+"""check.py <results_dir> <baseline.json> [--update]
+
+The conformance gate. Fails (exit 1) when
+  * clawhdf5 panicked, hung, crashed or ran out of memory on any file, or
+  * the ok count fell below the baseline's, or
+  * a file the baseline lists as ok is no longer ok (even if another file
+    became ok and the total held).
+New ok files are reported so the baseline can be raised (--update rewrites it
+from the results).
+"""
+import json
+import os
+import sys
+
+FATAL = ("panic", "hang", "crash", "oom")
+
+
+def main():
+    args = [a for a in sys.argv[1:] if not a.startswith("--")]
+    update = "--update" in sys.argv
+    res_dir, base_path = args
+    res = json.load(open(os.path.join(res_dir, "results.json")))
+    rows = res["rows"]
+    counts = {}
+    per_corpus = {}
+    for r in rows:
+        counts[r["class"]] = counts.get(r["class"], 0) + 1
+        pc = per_corpus.setdefault(r["corpus"], {})
+        pc[r["class"]] = pc.get(r["class"], 0) + 1
+    ok_files = sorted(r["file"] for r in rows if r["class"] == "ok")
+
+    if update:
+        meta = {}
+        mp = os.path.join(res_dir, "report-meta.json")
+        if os.path.exists(mp):
+            meta = json.load(open(mp))
+        base = {
+            "comment": "conformance/run.sh fails if the ok count drops below `ok` or a file in `ok_files` stops being ok. "
+                       "Regenerate with `conformance/run.sh --update-baseline` after an intended change.",
+            "commit": meta.get("commit", ""),
+            "date": meta.get("date", ""),
+            "reference": meta.get("reference", ""),
+            "files": len(rows),
+            "ok": len(ok_files),
+            "counts": dict(sorted(counts.items())),
+            "per_corpus": {k: dict(sorted(v.items())) for k, v in sorted(per_corpus.items())},
+            "ok_files": ok_files,
+        }
+        with open(base_path, "w") as fh:
+            json.dump(base, fh, indent=1)
+            fh.write("\n")
+        print(f"baseline updated: {len(ok_files)} ok of {len(rows)} files -> {base_path}")
+        return 0
+
+    base = json.load(open(base_path))
+    failures = []
+    fatal = [r for r in rows if r["class"] in FATAL]
+    for r in fatal:
+        failures.append(f"{r['class']}: {r['file']}: {r['ours_detail'][:200]}")
+    if len(ok_files) < base["ok"]:
+        failures.append(f"ok count dropped: {len(ok_files)} < baseline {base['ok']}")
+    now_ok = set(ok_files)
+    by_file = {r["file"]: r for r in rows}
+    for f in base["ok_files"]:
+        if f not in now_ok:
+            r = by_file.get(f)
+            why = f"now {r['class']}: {(r['ours_detail'] or r['first_issue'])[:200]}" if r else "no longer in the corpus"
+            failures.append(f"regressed: {f}: {why}")
+    gained = sorted(now_ok - set(base["ok_files"]))
+
+    print(f"conformance: {len(ok_files)} ok of {len(rows)} files (baseline {base['ok']} of {base['files']}); "
+          + ", ".join(f"{k} {v}" for k, v in sorted(counts.items())))
+    if gained:
+        print(f"{len(gained)} file(s) newly ok — raise the baseline with `conformance/run.sh --update-baseline`:")
+        for f in gained:
+            print(f"  + {f}")
+    if failures:
+        print(f"CONFORMANCE GATE FAILED ({len(failures)}):")
+        for f in failures:
+            print(f"  - {f}")
+        return 1
+    print("conformance gate passed")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

conformance/compare.py

@@ -0,0 +1,289 @@
+#!/usr/bin/env python3
+"""compare.py <results_dir>: classify each file and group failures by root cause.
+
+Writes <results_dir>/results.csv, results.json and summary.md.
+File classes (first match wins):
+  hang, oom, crash, panic       ours: timeout / allocation failure / signal / any panic (caught or not)
+  h5py-cannot-read              libhdf5/h5py failed to open the file (or crashed/hung)
+  our-error                     we fail to open, list, or read something h5py reads
+  mismatch                      we read something with different shape/values, or a different object set
+  ok
+"""
+import collections
+import csv
+import json
+import os
+import re
+import sys
+
+R = sys.argv[1]
+RUNS = os.path.join(R, "runs")
+
+
+def load(d, name):
+    rc_p = os.path.join(d, name + ".rc")
+    if not os.path.exists(rc_p):
+        return None
+    rc = int(open(rc_p).read().strip() or -1)
+    err = open(os.path.join(d, name + ".err"), errors="replace").read()
+    js = None
+    try:
+        js = json.load(open(os.path.join(d, name + ".json")))
+    except Exception:  # noqa: BLE001
+        pass
+    return {"rc": rc, "err": err, "json": js}
+
+
+def proc_status(p):
+    """-> (status, detail)"""
+    if p is None:
+        return "missing", ""
+    rc, err = p["rc"], p["err"]
+    first_panic = next((ln for ln in err.splitlines() if ln.startswith("PANIC:") or "panicked at" in ln), "")
+    if rc == 0 and p["json"] is not None:
+        return "ok", ""
+    if rc == 137 or rc == 124:
+        return "hang", f"timeout ({os.environ.get('TMO', '20')} s)"
+    if "memory allocation of" in err or "MemoryError" in err or "std::bad_alloc" in err:
+        m = re.search(r"memory allocation of \d+ bytes failed", err)
+        return "oom", m.group(0) if m else "allocation failure"
+    if "overflowed its stack" in err:
+        return "crash", "stack overflow"
+    if rc == 101:
+        return "panic", first_panic or (err.strip().splitlines() or [""])[-1]
+    if rc in (134, 139, 136, 135, 132) or rc > 128:
+        sig = {134: "SIGABRT", 139: "SIGSEGV", 136: "SIGFPE", 135: "SIGBUS", 132: "SIGILL"}.get(rc, f"signal {rc - 128}")
+        tail = [ln for ln in err.strip().splitlines() if ln.strip()][-1:]
+        return "crash", f"{sig}: {tail[0][:200] if tail else ''}"
+    tail = [ln for ln in err.strip().splitlines() if ln.strip()][-1:]
+    return "crash", f"rc={rc}: {tail[0][:200] if tail else ''}"
+
+
+def norm(msg):
+    m = msg.split("\n")[0]
+    m = re.sub(r"0x[0-9a-fA-F]+", "X", m)
+    m = re.sub(r'"[^"]*"', '"…"', m)
+    m = re.sub(r"'[^']*'", "'…'", m)
+    m = re.sub(r"\d+", "N", m)
+    return m[:160]
+
+
+def panic_head(msg):
+    """First line + first clawhdf5 frame of a PANIC record."""
+    lines = msg.split("\n")
+    frame = next((ln.strip() for ln in lines[1:] if "clawhdf5_format" in ln), "")
+    return lines[0][:300], frame[:300]
+
+
+def eq_shape(a, b):
+    return a == b
+
+
+rows = []
+issues_by_file = {}
+root_causes = collections.defaultdict(lambda: {"files": set(), "count": 0, "examples": []})
+mismatch_causes = collections.defaultdict(lambda: {"files": set(), "count": 0, "examples": []})
+panics = []
+ref_only_errors = collections.Counter()
+incomparable = collections.Counter()
+
+
+def add(bucket, key, file, example):
+    b = bucket[key]
+    b["count"] += 1
+    if file not in b["files"] and len(b["examples"]) < 6:
+        b["examples"].append(example)
+    b["files"].add(file)
+
+
+files = [ln.strip() for ln in open(os.path.join(R, "files.txt")) if ln.strip()]
+for rel in files:
+    d = os.path.join(RUNS, rel.replace("/", "__"))
+    corpus = rel.split("/")[0]
+    ours, ref = load(d, "ours"), load(d, "ref")
+    h5dump = load(d, "h5dump")
+    os_, od = proc_status(ours)
+    rs, rd = proc_status(ref)
+    oj = ours["json"] if ours else None
+    rj = ref["json"] if ref else None
+    issues = []  # (kind, detail)
+    caught_panics = []
+
+    def scan_err(path, what, msg):
+        if msg.startswith("PANIC:"):
+            caught_panics.append((path, what, msg))
+
+    if oj:
+        for o in oj.get("objects", []):
+            for k in ("error", "attrs_error", "list_error"):
+                if k in o:
+                    scan_err(o["path"], k, o[k])
+            for an, av in (o.get("attrs") or {}).items():
+                if "error" in av:
+                    scan_err(o["path"], f"attr {an}", av["error"])
+        if oj.get("open_error", "").startswith("PANIC:"):
+            caught_panics.append(("<open>", "open", oj["open_error"]))
+
+    ref_open_fail = rs != "ok" or (rj is not None and "open_error" in rj)
+    ours_open_err = oj.get("open_error") if oj else None
+    n_obj = n_ok = 0
+    if os_ == "ok" and rj and not ref_open_fail and not ours_open_err:
+        ro = {x["path"]: x for x in rj.get("objects", [])}
+        oo = {x["path"]: x for x in oj.get("objects", [])}
+        our_list_errors = [x for x in oo.values() if "list_error" in x]
+        for p in sorted(set(ro) | set(oo)):
+            a, b = ro.get(p), oo.get(p)
+            n_obj += 1
+            if a is None:
+                issues.append(("mismatch", f"extra object {p} (kind={b.get('kind')})", "extra-object", b))
+                continue
+            if b is None:
+                if our_list_errors:
+                    continue  # accounted for by the list_error
+                issues.append(("mismatch", f"missing object {p} (kind={a.get('kind')})", "missing-object", a))
+                continue
+            ok = True
+            if a.get("kind") != b.get("kind") and "error" not in b and "error" not in a:
+                issues.append(("mismatch", f"{p}: kind {a.get('kind')} vs ours {b.get('kind')}", "kind", b))
+                ok = False
+            for k in ("error", "list_error", "attrs_error"):
+                if k in b and k not in a:
+                    issues.append(("our-error", f"{p}: {k}: {b[k]}", b[k], b))
+                    ok = False
+                elif k in a and k not in b and k == "error":
+                    ref_only_errors[norm(a[k])] += 1
+            if a.get("kind") == "dataset" and "error" not in a and "error" not in b:
+                if "skipped" in a or "skipped" in b:
+                    pass
+                elif a.get("converted"):
+                    incomparable[f"dataset {a['converted']}"] += 1
+                elif a.get("shape") != b.get("shape"):
+                    issues.append(("mismatch", f"{p}: shape {a.get('shape')} vs ours {b.get('shape')}", "shape", b))
+                    ok = False
+                elif a.get("hash") != b.get("hash"):
+                    issues.append(("mismatch", f"{p}: values differ (h5py {a.get('dtype')} vs ours {b.get('dtype')})", "values", b | {"ref_head": a.get("head"), "ref_dtype": a.get("dtype")}))
+                    ok = False
+            ra, oa = a.get("attrs") or {}, b.get("attrs") or {}
+            if "attrs_error" not in b and "attrs_error" not in a:
+                for an in sorted(set(ra) | set(oa)):
+                    x, y = ra.get(an), oa.get(an)
+                    if x is None:
+                        issues.append(("mismatch", f"{p}@{an}: extra attribute", "extra-attr", y or {}))
+                    elif y is None:
+                        issues.append(("mismatch", f"{p}@{an}: missing attribute", "missing-attr", x))
+                    elif "error" in y and "error" not in x:
+                        issues.append(("our-error", f"{p}@{an}: {y['error']}", y["error"], y))
+                    elif "error" in x:
+                        continue
+                    elif x.get("converted"):
+                        incomparable[f"attr {x['converted']}"] += 1
+                    elif x.get("shape") != y.get("shape"):
+                        issues.append(("mismatch", f"{p}@{an}: attr shape {x.get('shape')} vs ours {y.get('shape')}", "attr-shape", y | {"ref_dtype": x.get("dtype")}))
+                    elif x.get("hash") != y.get("hash"):
+                        issues.append(("mismatch", f"{p}@{an}: attr values differ (h5py {x.get('dtype')} vs ours {y.get('dtype')})", "attr-values", y | {"ref_head": x.get("head"), "ref_dtype": x.get("dtype")}))
+            if ok:
+                n_ok += 1
+
+    # classify
+    if os_ in ("hang", "oom", "crash", "panic"):
+        cls = os_
+    elif caught_panics:
+        cls = "panic"
+    elif ref_open_fail:
+        cls = "h5py-cannot-read"
+    elif ours_open_err:
+        cls = "our-error"
+        issues.append(("our-error", f"open: {ours_open_err}", ours_open_err, {}))
+    elif any(i[0] == "our-error" for i in issues):
+        cls = "our-error"
+    elif issues:
+        cls = "mismatch"
+    else:
+        cls = "ok"
+
+    if os_ in ("hang", "oom", "crash", "panic") or caught_panics:
+        panics.append({
+            "file": rel, "class": cls, "detail": od,
+            "stderr": (ours["err"] if ours else "")[:3000],
+            "caught": [(p, w, m[:2500]) for p, w, m in caught_panics[:3]],
+            "n_caught": len(caught_panics),
+        })
+    for kind, detail, key, rec in issues:
+        if kind == "our-error":
+            add(root_causes, norm(key), rel, detail[:300])
+        else:
+            if key in ("values", "attr-values", "shape", "attr-shape"):
+                mk = f"{key}: ours={rec.get('dtype')} h5py={rec.get('ref_dtype')} layout={rec.get('layout','-')} filters={rec.get('filters','-')}"
+            else:
+                mk = key
+            add(mismatch_causes, mk, rel, detail[:300] + (f" | ref_head={rec.get('ref_head')} our_head={rec.get('head')}" if rec.get("ref_head") else ""))
+    ref_detail = rd if rs != "ok" else ((rj or {}).get("open_error") or "")
+    h5d = ""
+    if h5dump:
+        rc = h5dump["rc"]
+        h5d = {0: "ok", 1: "error", 137: "hang", 124: "hang", 134: "SIGABRT", 139: "SIGSEGV", 136: "SIGFPE", 135: "SIGBUS"}.get(rc, f"rc={rc}")
+        if "memory allocation" in h5dump["err"] or "Cannot allocate" in h5dump["err"]:
+            h5d += "(oom)"
+    rows.append({
+        "file": rel, "corpus": corpus, "class": cls,
+        "ours": os_ if os_ != "ok" else ("open-error" if ours_open_err else ("panic" if caught_panics else "ok")),
+        "ours_detail": (od or ours_open_err or (caught_panics[0][2].split("\n")[0] if caught_panics else ""))[:300],
+        "ref": rs if rs != "ok" else ("open-error" if (rj or {}).get("open_error") else "ok"),
+        "ref_detail": ref_detail[:300],
+        "h5dump_1_14_6": h5d,
+        "h5dump_detail": ([ln for ln in h5dump["err"].splitlines() if ln.strip()][-1:] or [""])[0][:200] if h5dump else "",
+        "objects": n_obj, "objects_ok": n_ok,
+        "issues": len(issues), "first_issue": issues[0][1][:300] if issues else "",
+        "superblock": (oj or {}).get("superblock_version", ""),
+    })
+    # the first issues of each file, for report.py's known-cause matching
+    issues_by_file[rel] = [
+        {"kind": k, "key": key, "detail": det[:300], "ours_dtype": rec.get("dtype"), "ref_dtype": rec.get("ref_dtype")}
+        for k, det, key, rec in issues[:50]
+    ]
+
+with open(os.path.join(R, "results.csv"), "w", newline="") as fh:
+    w = csv.DictWriter(fh, fieldnames=list(rows[0].keys()))
+    w.writeheader()
+    w.writerows(rows)
+
+
+def ser(b):
+    return {k: {"files": len(v["files"]), "count": v["count"], "examples": v["examples"], "file_list": sorted(v["files"])} for k, v in sorted(b.items(), key=lambda kv: -len(kv[1]["files"]))}
+
+
+json.dump({"rows": rows, "issues": issues_by_file, "root_causes": ser(root_causes), "mismatch_causes": ser(mismatch_causes),
+           "panics": panics, "incomparable": incomparable.most_common(), "ref_only_errors": ref_only_errors.most_common()},
+          open(os.path.join(R, "results.json"), "w"), indent=1)
+
+classes = ["ok", "our-error", "mismatch", "h5py-cannot-read", "hang", "panic", "crash", "oom"]
+by_corpus = collections.defaultdict(collections.Counter)
+for r in rows:
+    by_corpus[r["corpus"]][r["class"]] += 1
+    by_corpus["ALL"][r["class"]] += 1
+lines = ["# Conformance sweep summary", "", "| corpus | files | " + " | ".join(classes) + " |", "|---" * (len(classes) + 2) + "|"]
+for c in sorted(by_corpus, key=lambda k: (k == "ALL", k)):
+    cnt = by_corpus[c]
+    lines.append(f"| {c} | {sum(cnt.values())} | " + " | ".join(str(cnt.get(k, 0)) for k in classes) + " |")
+lines += ["", "## Panics / hangs / crashes / OOM", ""]
+for p in panics:
+    lines.append(f"- **{p['file']}** [{p['class']}] {p['detail']}")
+    for path, what, m in p["caught"][:1]:
+        lines.append("  ```\n  " + f"{path} ({what}): " + m.replace("\n", "\n  ")[:1500] + "\n  ```")
+    if not p["caught"] and p["stderr"]:
+        lines.append("  ```\n  " + p["stderr"].strip()[:1500].replace("\n", "\n  ") + "\n  ```")
+lines += ["", "## Our-error root causes (files affected)", ""]
+for k, v in ser(root_causes).items():
+    lines.append(f"- [{v['files']} files, {v['count']} objs] `{k}`")
+    for ex in v["examples"][:3]:
+        lines.append(f"    - {ex}")
+lines += ["", "## Mismatch root causes", ""]
+for k, v in ser(mismatch_causes).items():
+    lines.append(f"- [{v['files']} files, {v['count']} objs] `{k}`")
+    for ex in v["examples"][:3]:
+        lines.append(f"    - {ex}")
+lines += ["", "## Objects h5py fails on but we read (top)", ""]
+for k, n in ref_only_errors.most_common(15):
+    lines.append(f"- {n} x `{k}`")
+open(os.path.join(R, "summary.md"), "w").write("\n".join(lines) + "\n")
+print("\n".join(lines[:4 + len(by_corpus)]))

conformance/corpus.txt

@@ -0,0 +1,19 @@
+# Conformance corpora, pinned by commit. fetch-corpus.sh reads this file.
+#
+# name  git-url  commit  root  [sparse-checkout patterns...]
+#
+# `root` is the directory inside the checkout that is swept ("." = all of it).
+# Patterns are git non-cone sparse-checkout patterns; none = whole repository.
+# Every file under <root> with an HDF5/netCDF-4 extension is probed; for
+# cve_hdf5 the extension-less files in cvefiles/ and fuzzerfiles/ are too.
+# Licences: each corpus keeps its upstream licence; nothing here is committed
+# to this repository — the files are downloaded into the gitignored cache.
+hdf5              https://github.com/HDFGroup/hdf5.git            a3cf1ea82cc7a66e50029a688121e1b105a7ce88  .  *.h5 *.he5 *.nc *.hdf5 *.h5f
+cve_hdf5          https://github.com/HDFGroup/cve_hdf5.git        3fd1f5ae3869e01b8ae02b41d7108de7ffb1a374  .
+netcdf-c          https://github.com/Unidata/netcdf-c.git         beb7b9585273c1548386231a59b809d906359033  .  /nc_test4/*.nc /ncdump/*.nc /nc_test4/*.h5 /ncdump/*.h5 /h5_test/*.h5 /hdf5_test/*.h5
+NCAS-CMS_pyfive   https://github.com/NCAS-CMS/pyfive.git          8cf07b8749133f41c5e30b8a4c604486f687fe74  .  *.h5 *.hdf5 *.hdf *.nc *.he5
+usnistgov_h5wasm  https://github.com/usnistgov/h5wasm.git         02f6336527d2812783fcedabfbf42127ec8d06d2  .  *.h5 *.hdf5 *.hdf *.nc *.he5
+netcdf4-python    https://github.com/Unidata/netcdf4-python.git   6e67576d39aef8091fb20bd767b4f1a52ddc1bec  .  *.nc *.h5
+xarray-data       https://github.com/pydata/xarray-data.git       a35297e9da2cc99c811014f0c8a4297345a5c28d  .  /basin_mask.nc /precipitation.nc4 /imerghh_730.hdf5 /eraint_uvz.nc /ROMS_example.nc /tiny.nc
+# h5py 3.16.0 (tag 3.16.0), its test data files.
+h5py_data         https://github.com/h5py/h5py.git                b2f0347c4200333acd89b43733f1caa0c115162f  h5py/tests/data_files  /h5py/tests/data_files/*

conformance/fetch-corpus.sh

@@ -0,0 +1,39 @@
+#!/usr/bin/env bash
+# fetch-corpus.sh [cache_dir]
+#
+# Download the corpora pinned in conformance/corpus.txt into the (gitignored)
+# cache: <cache>/src/<name> is a shallow, sparse, blob-filtered checkout of the
+# pinned commit and <cache>/corpus/<name> links to the swept root inside it.
+# A corpus already checked out at its pinned commit is left alone, so a second
+# run costs nothing and needs no network.
+set -euo pipefail
+HERE="$(cd "$(dirname "$0")" && pwd)"
+CACHE="${1:-${CONFORMANCE_CACHE:-$HERE/.cache}}"
+mkdir -p "$CACHE/src" "$CACHE/corpus"
+CACHE="$(cd "$CACHE" && pwd)"
+
+retry() { local i; for i in 1 2 3 4; do "$@" && return 0; sleep $((i * 5)); done; return 1; }
+
+grep -v '^[[:space:]]*\(#\|$\)' "$HERE/corpus.txt" | while read -r name url commit root patterns; do
+  src="$CACHE/src/$name"
+  if [ -d "$src/.git" ] && [ "$(git -C "$src" rev-parse HEAD 2>/dev/null)" = "$commit" ]; then
+    echo "cached   $name @ ${commit:0:12}"
+  else
+    echo "fetching $name @ ${commit:0:12} from $url"
+    rm -rf "$src"
+    git init -q "$src"
+    git -C "$src" remote add origin "$url"
+    git -C "$src" config advice.detachedHead false
+    if [ -n "$patterns" ]; then
+      git -C "$src" config core.sparseCheckout true
+      # no-cone patterns (globs); `set -f` keeps the shell from expanding them
+      (set -f; printf '%s\n' $patterns) > "$src/.git/info/sparse-checkout"
+    fi
+    retry git -C "$src" fetch -q --depth 1 --filter=blob:none origin "$commit"
+    retry git -C "$src" checkout -q FETCH_HEAD
+    got="$(git -C "$src" rev-parse HEAD)"
+    [ "$got" = "$commit" ] || { echo "error: $name checked out $got, expected $commit" >&2; exit 1; }
+  fi
+  ln -sfn "$src/$root" "$CACHE/corpus/$name"
+done
+echo "corpus ready in $CACHE/corpus"

conformance/list_files.py

@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+"""list_files.py <corpus_dir>: print the files the sweep probes, one per line,
+as <corpus>/<path> in byte order.
+
+  * every file named *.h5 *.hdf5 *.he5 *.nc *.nc4 *.hdf *.h5f in each corpus,
+    except netCDF classic / 64-bit-offset / CDF5 files (magic "CDF"): they are
+    not HDF5, so neither side can read them and they say nothing;
+  * plus, for cve_hdf5, every file in cvefiles/ and fuzzerfiles/ except
+    .md/.c sources — the reproducers are mostly extension-less, and they are
+    kept whatever their bytes look like (that is their point).
+"""
+import os
+import sys
+
+EXTS = (".h5", ".hdf5", ".he5", ".nc", ".nc4", ".hdf", ".h5f")
+
+
+def walk(top):
+    for dirpath, dirnames, filenames in os.walk(top):
+        dirnames[:] = [d for d in dirnames if d != ".git"]
+        for fn in filenames:
+            p = os.path.join(dirpath, fn)
+            if os.path.isfile(p) and not os.path.islink(p):
+                yield os.path.relpath(p, top)
+
+
+def main(root):
+    out = set()
+    for corpus in sorted(os.listdir(root)):
+        top = os.path.join(root, corpus)
+        if not os.path.isdir(top):
+            continue
+        for rel in walk(top):
+            path = os.path.join(top, rel)
+            if rel.lower().endswith(EXTS):
+                with open(path, "rb") as fh:
+                    if fh.read(3) == b"CDF":
+                        continue
+                out.add(f"{corpus}/{rel}")
+            elif corpus == "cve_hdf5" and rel.split(os.sep)[0] in ("cvefiles", "fuzzerfiles") \
+                    and not rel.endswith((".md", ".c")):
+                out.add(f"{corpus}/{rel}")
+    for f in sorted(out, key=lambda s: s.encode()):
+        print(f)
+
+
+if __name__ == "__main__":
+    main(sys.argv[1])

conformance/probe/Cargo.lock

@@ -0,0 +1,492 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 4
+
+[[package]]
+name = "adler2"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "320119579fcad9c21884f5c4861d16174d0e06250625266f50fe6898340abefa"
+
+[[package]]
+name = "better_io"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ef0a3155e943e341e557863e69a708999c94ede624e37865c8e2a91b94efa78f"
+
+[[package]]
+name = "block-buffer"
+version = "0.10.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3078c7629b62d3f0439517fa394996acacc5cbc91c5a20d8c658e77abd503a71"
+dependencies = [
+ "generic-array",
+]
+
+[[package]]
+name = "byteorder"
+version = "1.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fd0f2584146f6f2ef48085050886acf353beff7305ebd1ae69500e27c67f64b"
+
+[[package]]
+name = "bzip2"
+version = "0.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f3a53fac24f34a81bc9954b5d6cfce0c21e18ec6959f44f56e8e90e4bb7c346c"
+dependencies = [
+ "libbz2-rs-sys",
+]
+
+[[package]]
+name = "cc"
+version = "1.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f360145194ee8e21db5ee7f3fcd4fe52210864c75c985dae33218202c8bbe040"
+dependencies = [
+ "find-msvc-tools",
+ "jobserver",
+ "libc",
+ "shlex",
+]
+
+[[package]]
+name = "cfg-if"
+version = "1.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4e7648175b45a9a48536d676f68d918270699102aa8dab5496df06904c914600"
+
+[[package]]
+name = "clawhdf5-format"
+version = "2.7.0"
+dependencies = [
+ "byteorder",
+ "bzip2",
+ "flate2",
+ "libaec-sys",
+ "libc",
+ "lz4_flex",
+ "pco",
+ "portable-atomic",
+ "ruzstd",
+ "sha2",
+ "snap",
+ "zstd",
+]
+
+[[package]]
+name = "conformance-probe"
+version = "0.1.0"
+dependencies = [
+ "clawhdf5-format",
+ "serde_json",
+ "sha2",
+]
+
+[[package]]
+name = "cpufeatures"
+version = "0.2.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59ed5838eebb26a2bb2e58f6d5b5316989ae9d08bab10e0e6d103e656d1b0280"
+dependencies = [
+ "libc",
+]
+
+[[package]]
+name = "crc32fast"
+version = "1.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "01a7799fd6b852db0e61728dde9a204c423b44d689dbd432522543614b490e78"
+dependencies = [
+ "cfg-if",
+]
+
+[[package]]
+name = "crunchy"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "460fbee9c2c2f33933d720630a6a0bac33ba7053db5344fac858d4b8952d77d5"
+
+[[package]]
+name = "crypto-common"
+version = "0.1.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78c8292055d1c1df0cce5d180393dc8cce0abec0a7102adb6c7b1eef6016d60a"
+dependencies = [
+ "generic-array",
+ "typenum",
+]
+
+[[package]]
+name = "digest"
+version = "0.10.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
+dependencies = [
+ "block-buffer",
+ "crypto-common",
+]
+
+[[package]]
+name = "dtype_dispatch"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ab23e69df104e2fd85ee63a533a22d2132ef5975dc6b36f9f3e5a7305e4a8ed7"
+
+[[package]]
+name = "find-msvc-tools"
+version = "0.1.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aedcfb3409746eddb02b9e19ebda1c3394f759a152e48ee875a0844d1b955484"
+
+[[package]]
+name = "flate2"
+version = "1.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6e634e2e0ebac1ee034020da1ca582e17ffe4e0f5e985823721e168928136dcb"
+dependencies = [
+ "crc32fast",
+ "miniz_oxide",
+ "zlib-rs",
+]
+
+[[package]]
+name = "generic-array"
+version = "0.14.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85649ca51fd72272d7821adaf274ad91c288277713d9c18820d8499a7ff69e9a"
+dependencies = [
+ "typenum",
+ "version_check",
+]
+
+[[package]]
+name = "getrandom"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "300e883d756b2e4ec94e02791f39b04b522276138852cfc41d9fb7e904106099"
+dependencies = [
+ "cfg-if",
+ "libc",
+ "r-efi",
+]
+
+[[package]]
+name = "half"
+version = "2.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6ea2d84b969582b4b1864a92dc5d27cd2b77b622a8d79306834f1be5ba20d84b"
+dependencies = [
+ "cfg-if",
+ "crunchy",
+ "zerocopy",
+]
+
+[[package]]
+name = "itoa"
+version = "1.0.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682"
+
+[[package]]
+name = "jobserver"
+version = "0.1.35"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1c00acbd29eabad4a2392fa0e921c874934dbbf4194312ad20f04a0ed67a3cb3"
+dependencies = [
+ "getrandom",
+ "libc",
+]
+
+[[package]]
+name = "libaec-sys"
+version = "0.1.0"
+dependencies = [
+ "pkg-config",
+]
+
+[[package]]
+name = "libbz2-rs-sys"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34b357333733e8260735ba5894eb928c02ecc69c78715f01a8019e7fa7f2db4c"
+
+[[package]]
+name = "libc"
+version = "0.2.189"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3eaf3ede3fee6db1a4c2ee091bf8a8b4dccdc6d17f656fb07896ee72867612f2"
+
+[[package]]
+name = "lz4_flex"
+version = "0.11.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "373f5eceeeab7925e0c1098212f2fbc4d416adec9d35051a6ab251e824c1854a"
+dependencies = [
+ "twox-hash",
+]
+
+[[package]]
+name = "memchr"
+version = "2.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cf8baf1c55e62ffcace7a9f06f4bd9cd3f0c4beb022d3b367256b91b87513d98"
+
+[[package]]
+name = "miniz_oxide"
+version = "0.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b63fbc4a50860e98e7b2aa7804ded1db5cbc3aff9193adaff57a6931bf7c4b4c"
+dependencies = [
+ "adler2",
+ "simd-adler32",
+]
+
+[[package]]
+name = "pco"
+version = "1.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "386342cad4c6e97f081568e5d910ea7d871314c843aa8fc564f2a6b64cab9456"
+dependencies = [
+ "better_io",
+ "dtype_dispatch",
+ "half",
+ "rand_xoshiro",
+]
+
+[[package]]
+name = "pkg-config"
+version = "0.3.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6b464fbc74e149a392436b17d523f769e057cb6877f6a5c4618bc6f11800548"
+
+[[package]]
+name = "portable-atomic"
+version = "1.15.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05c8b63e8d9609db387f0324918f81d68fe27748f084ef092fb35954d0539a85"
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.107"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "985e7ec9bb745e6ce6535b544d84d6cd6f7ad8bd711c398938ae983b91a766d9"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.47"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fbf4db142a473a8d80c26bbf18454ed458bf8d26c8219c331daecfdbd079001"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "r-efi"
+version = "6.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8dcc9c7d52a811697d2151c701e0d08956f92b0e24136cf4cf27b57a6a0d9bf"
+
+[[package]]
+name = "rand_core"
+version = "0.6.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ec0be4795e2f6a28069bec0b5ff3e2ac9bafc99e6a9a7dc3547996c5c816922c"
+
+[[package]]
+name = "rand_xoshiro"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6f97cdb2a36ed4183de61b2f824cc45c9f1037f28afe0a322e9fff4c108b5aaa"
+dependencies = [
+ "rand_core",
+]
+
+[[package]]
+name = "ruzstd"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a252f5e20f038fe7b4ea53e073e65398d652c864cc162fc77c56c2f13717b888"
+dependencies = [
+ "twox-hash",
+]
+
+[[package]]
+name = "serde"
+version = "1.0.229"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4148590afebada386688f18773da617792bf2ef03ffc1e4cbd2b1d45b023e0ba"
+dependencies = [
+ "serde_core",
+]
+
+[[package]]
+name = "serde_core"
+version = "1.0.229"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "67dca2c9c51e58a4791a4b1ed58308b39c64224d349a935ab5039aa360942a48"
+dependencies = [
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_derive"
+version = "1.0.229"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e7a5d71263a5a7d47b41f6b3f06ba276f10cc18b0931f1799f710578e2309348"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 3.0.6",
+]
+
+[[package]]
+name = "serde_json"
+version = "1.0.151"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c841b55ecdae098c80dcae9cf767f6f8a0c2cdb3416bbef72181df4d0fe73f14"
+dependencies = [
+ "itoa",
+ "memchr",
+ "serde",
+ "serde_core",
+ "zmij",
+]
+
+[[package]]
+name = "sha2"
+version = "0.10.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a7507d819769d01a365ab707794a4084392c824f54a7a6a7862f8c3d0892b283"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "digest",
+]
+
+[[package]]
+name = "shlex"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8fadd59c855ef2080decdef8ff161eb6661b86933c9d82e5ba29dc602a55aba"
+
+[[package]]
+name = "simd-adler32"
+version = "0.3.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a219298ac11a56ea9a6d2120044824d6f01aeb034955e7af7bc16858527deea"
+
+[[package]]
+name = "snap"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "199905e6153d6405f9728fe44daace35f8f837bbf830bb6e85fbd5828709a886"
+
+[[package]]
+name = "syn"
+version = "2.0.119"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "872831b642d1a07999a962a351ed35b955ea2cfc8f3862091e2a240a84f17297"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "syn"
+version = "3.0.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8593e8e72159ed2257d083c7a454a85cbf854f37a0966d8d483aff8c8a3ebcee"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "twox-hash"
+version = "2.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5283634e518fe9e82c7b20520bb4bc209009fd16c82077c802f8111ecbb0117a"
+
+[[package]]
+name = "typenum"
+version = "1.20.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6f5e870be6c3b371b77fe0ee0bafb859fa4964b4404c27de1d380043c4dda20"
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.26"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d245f478577f809a851594d02313b640fb437e0bb33866753cff937863096954"
+
+[[package]]
+name = "version_check"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
+
+[[package]]
+name = "zerocopy"
+version = "0.8.59"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6df92bf3d9227be3d53173901ddbffac2babc27ae50f397776ffd6dc33f800cb"
+dependencies = [
+ "zerocopy-derive",
+]
+
+[[package]]
+name = "zerocopy-derive"
+version = "0.8.59"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ac4f328cf2f05d084e496c3e9c3f33ed0a183656a16e1fcec4d464d8373aec82"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.119",
+]
+
+[[package]]
+name = "zlib-rs"
+version = "0.6.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b268e58e7c693d7c271f93ffc4ba3b380412554231c85bf61ca7af91042a4112"
+
+[[package]]
+name = "zmij"
+version = "1.0.23"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "29666d0abbfad1e3dc4dcf6144730dd3a3ab225bbbdac83319345b1b44ccfc1b"
+
+[[package]]
+name = "zstd"
+version = "0.13.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e91ee311a569c327171651566e07972200e76fcfe2242a4fa446149a3881c08a"
+dependencies = [
+ "zstd-safe",
+]
+
+[[package]]
+name = "zstd-safe"
+version = "7.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "64d80649ab6db9d9f6f9c80a40becd948eda4714a0a5ac8c4d157a32231c7882"
+dependencies = [
+ "zstd-sys",
+]
+
+[[package]]
+name = "zstd-sys"
+version = "2.1.0+zstd.1.5.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0ef0a8027ec3ee71300ab3bcbcd0393f434aa72b91ca6d635a39941deae8eea0"
+dependencies = [
+ "cc",
+ "pkg-config",
+]

conformance/probe/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "conformance-probe"
+version = "0.1.0"
+edition = "2024"
+rust-version = "1.92"
+publish = false
+description = "Walks an HDF5 file with clawhdf5-format and prints a canonical JSON description (see conformance/README.md)"
+
+# Deliberately outside the main workspace: `cargo test --workspace` never
+# builds it, and it links the optional C codecs (zstd, libaec) that the core
+# crates' default build must not.
+[workspace]
+
+[dependencies]
+clawhdf5-format = { path = "../../crates/clawhdf5-format", features = ["lz4", "zstd", "szip", "pcodec", "plugin-filters"] }
+serde_json = "1"
+sha2 = "0.10"
+
+[profile.release]
+# Keep panics catchable (the probe records them per object) and turn integer
+# overflow into a reported panic instead of silent wraparound.
+debug = 1
+overflow-checks = true
+debug-assertions = true
+panic = "unwind"

conformance/probe/src/main.rs

@@ -0,0 +1,885 @@
+//! Conformance probe: walks an HDF5 file with clawhdf5-format (the same calls
+//! the `clawhdf5` facade makes) and prints a canonical JSON description:
+//! every hard-linked object (sorted-name DFS, deduplicated by header address),
+//! and for each dataset / attribute its shape plus the SHA-256 of its values
+//! in a canonical encoding shared with `ref.py`.
+//!
+//! Canonical value encoding (per element, concatenated, row-major):
+//!   int / float / bitfield / enum / time : element bytes, little-endian
+//!   non-IEEE-layout float (e.g. N-Bit)   : the IEEE float of the same size it converts to
+//!   int with bit offset / short precision: the full-width integer it converts to
+//!   opaque                               : raw bytes
+//!   compound                             : members in declaration order (padding dropped)
+//!   array                                : base elements row-major
+//!   string (fixed or VL)                 : b'S' + u32le len + bytes (cut at first NUL, trailing spaces stripped)
+//!   VL sequence                          : b'V' + u32le count + base elements
+//!   reference                            : b'R' (payload not compared)
+//!
+//! Every object is processed inside catch_unwind; a caught panic is recorded
+//! with its message, location and the clawhdf5 frames of its backtrace.
+
+use std::cell::RefCell;
+use std::collections::HashSet;
+use std::panic::{self, AssertUnwindSafe};
+
+use clawhdf5_format::attribute::extract_attributes_full;
+use clawhdf5_format::data_layout::DataLayout;
+use clawhdf5_format::data_read;
+use clawhdf5_format::dataspace::{Dataspace, DataspaceType};
+use clawhdf5_format::datatype::{Datatype, DatatypeByteOrder};
+use clawhdf5_format::filter_pipeline::FilterPipeline;
+use clawhdf5_format::group_v1::{self, GroupEntry};
+use clawhdf5_format::group_v2;
+use clawhdf5_format::message_type::MessageType;
+use clawhdf5_format::object_header::ObjectHeader;
+use clawhdf5_format::signature;
+use clawhdf5_format::superblock::Superblock;
+use clawhdf5_format::symbol_table::SymbolTableMessage;
+use clawhdf5_format::vl_data::{VlResolver, check_element_size};
+use serde_json::{Map, Value, json};
+use sha2::{Digest, Sha256};
+
+const MAX_BYTES: u64 = 200 * 1024 * 1024;
+const MAX_OBJECTS: usize = 200_000;
+
+thread_local! {
+    static LAST_PANIC: RefCell<Option<String>> = const { RefCell::new(None) };
+}
+
+fn install_hook() {
+    panic::set_hook(Box::new(|info| {
+        let msg = if let Some(s) = info.payload().downcast_ref::<&str>() {
+            s.to_string()
+        } else if let Some(s) = info.payload().downcast_ref::<String>() {
+            s.clone()
+        } else {
+            "<non-string panic>".into()
+        };
+        let loc = info
+            .location()
+            .map(|l| format!("{}:{}", l.file(), l.line()))
+            .unwrap_or_default();
+        let bt = std::backtrace::Backtrace::force_capture().to_string();
+        // keep only frames from clawhdf5 code
+        let mut frames = Vec::new();
+        let lines: Vec<&str> = bt.lines().collect();
+        for (i, l) in lines.iter().enumerate() {
+            let t = l.trim();
+            if t.contains("clawhdf5_format::") || t.contains("conformance_probe::") {
+                let at = lines
+                    .get(i + 1)
+                    .map(|n| n.trim())
+                    .filter(|n| n.starts_with("at "))
+                    .map(|n| {
+                        let n = n.trim_start_matches("at ");
+                        match n.find("/crates/") {
+                            Some(p) => n[p + 1..].to_string(),
+                            None => n.to_string(),
+                        }
+                    })
+                    .unwrap_or_default();
+                let name = t.split_once(": ").map(|x| x.1).unwrap_or(t);
+                frames.push(format!("{name} ({at})"));
+                if frames.len() >= 12 {
+                    break;
+                }
+            }
+        }
+        let full = format!("PANIC: {msg} @ {loc}\n  {}", frames.join("\n  "));
+        eprintln!("{full}");
+        LAST_PANIC.with(|p| *p.borrow_mut() = Some(full));
+    }));
+}
+
+/// Run `f`, turning a panic into Err("PANIC: ...").
+fn guarded<T>(f: impl FnOnce() -> Result<T, String>) -> Result<T, String> {
+    match panic::catch_unwind(AssertUnwindSafe(f)) {
+        Ok(r) => r,
+        Err(_) => Err(LAST_PANIC
+            .with(|p| p.borrow_mut().take())
+            .unwrap_or_else(|| "PANIC: <unknown>".into())),
+    }
+}
+
+fn e<E: std::fmt::Debug>(x: E) -> String {
+    format!("{x:?}")
+}
+
+struct Ctx<'a> {
+    data: &'a [u8],
+    os: u8,
+    ls: u8,
+    base_dir: std::path::PathBuf,
+    /// Resolves variable-length elements as the library does (null
+    /// elements, strings cut at a NUL, heap objects of the wrong size
+    /// refused), caching each heap collection.
+    vl: RefCell<VlResolver<'a>>,
+}
+
+impl<'a> Ctx<'a> {
+    fn header(&self, addr: u64) -> Result<ObjectHeader, String> {
+        ObjectHeader::parse(self.data, addr as usize, self.os, self.ls).map_err(e)
+    }
+
+    fn payload(&self, h: &ObjectHeader, t: MessageType) -> Result<Option<Vec<u8>>, String> {
+        match h.messages.iter().find(|m| m.msg_type == t) {
+            None => Ok(None),
+            Some(m) => {
+                clawhdf5_format::shared_message::message_data(self.data, m, self.os, self.ls)
+                    .map(|c| Some(c.into_owned()))
+                    .map_err(e)
+            }
+        }
+    }
+
+    fn canon(&self, dt: &Datatype, b: &[u8], out: &mut Vec<u8>) -> Result<(), String> {
+        let size = dt.type_size() as usize;
+        if b.len() < size {
+            return Err(format!(
+                "canon: element slice {} < type size {size}",
+                b.len()
+            ));
+        }
+        match dt {
+            Datatype::FloatingPoint { .. } if !ieee_layout(dt) => {
+                canon_custom_float(dt, &b[..size], out)?
+            }
+            Datatype::FixedPoint { .. } if partial_int(dt) => {
+                canon_partial_int(dt, &b[..size], out)?
+            }
+            Datatype::FixedPoint { byte_order, .. }
+            | Datatype::BitField { byte_order, .. }
+            | Datatype::FloatingPoint { byte_order, .. } => match byte_order {
+                DatatypeByteOrder::LittleEndian => out.extend_from_slice(&b[..size]),
+                DatatypeByteOrder::BigEndian => out.extend(b[..size].iter().rev()),
+                DatatypeByteOrder::Vax => return Err("canon: VAX byte order".into()),
+            },
+            Datatype::Time { .. } | Datatype::Opaque { .. } => out.extend_from_slice(&b[..size]),
+            Datatype::String { .. } => canon_str(&b[..size], out),
+            Datatype::Compound { members, .. } => {
+                for m in members {
+                    let off = m.byte_offset as usize;
+                    let ms = m.datatype.type_size() as usize;
+                    if off.checked_add(ms).is_none_or(|end| end > size) {
+                        return Err(format!("canon: member {} out of bounds", m.name));
+                    }
+                    self.canon(&m.datatype, &b[off..off + ms], out)?;
+                }
+            }
+            Datatype::Reference { .. } => out.push(b'R'),
+            Datatype::Enumeration { base_type, .. } => self.canon(base_type, b, out)?,
+            Datatype::Array {
+                base_type,
+                dimensions,
+            } => {
+                let n: usize = dimensions.iter().map(|d| *d as usize).product();
+                let bs = base_type.type_size() as usize;
+                for i in 0..n {
+                    self.canon(base_type, &b[i * bs..], out)?;
+                }
+            }
+            Datatype::VariableLength {
+                size: vl_size,
+                is_string,
+                base_type,
+                ..
+            } => {
+                check_element_size(*vl_size, self.os).map_err(e)?;
+                let el = &b[..size];
+                if *is_string {
+                    let s = self.vl.borrow_mut().string_bytes(el).map_err(e)?;
+                    canon_str(&s[0], out);
+                } else {
+                    let bs = base_type.type_size() as usize;
+                    // The borrow ends here: the base type may itself be
+                    // variable-length.
+                    let seq = self.vl.borrow_mut().sequences(el, bs).map_err(e)?;
+                    let seq = &seq[0];
+                    let len = seq.len() / bs;
+                    out.push(b'V');
+                    out.extend_from_slice(&(len as u32).to_le_bytes());
+                    for i in 0..len {
+                        self.canon(base_type, &seq[i * bs..], out)?;
+                    }
+                }
+            }
+        }
+        Ok(())
+    }
+
+    /// Returns (shape json, n_elements)
+    fn shape(ds: &Dataspace) -> (Value, u64) {
+        match ds.space_type {
+            DataspaceType::Null => (Value::String("null".into()), 0),
+            DataspaceType::Scalar => (json!([]), 1),
+            DataspaceType::Simple => {
+                let n = ds.dimensions.iter().fold(1u64, |a, d| a.saturating_mul(*d));
+                (json!(ds.dimensions), n)
+            }
+        }
+    }
+
+    fn hash_values(
+        &self,
+        dt: &Datatype,
+        raw: &[u8],
+        n: u64,
+        rec: &mut Map<String, Value>,
+    ) -> Result<(), String> {
+        let size = dt.type_size() as usize;
+        let need = (n as usize).checked_mul(size).ok_or("n*size overflow")?;
+        if raw.len() != need {
+            return Err(format!(
+                "raw length {} != n_elements {n} * type_size {size}",
+                raw.len()
+            ));
+        }
+        let mut canon = Vec::with_capacity(need);
+        for i in 0..n as usize {
+            self.canon(dt, &raw[i * size..(i + 1) * size], &mut canon)?;
+        }
+        let h = Sha256::digest(&canon);
+        rec.insert("hash".into(), Value::String(hex(&h)));
+        rec.insert(
+            "head".into(),
+            Value::String(hex(&canon[..canon.len().min(48)])),
+        );
+        Ok(())
+    }
+
+    /// VDS source files resolve next to the virtual file; like the library,
+    /// refuse absolute paths and `..`.
+    fn vds_resolver(
+        &self,
+    ) -> impl Fn(&str) -> Result<Option<Vec<u8>>, clawhdf5_format::error::FormatError> + use<> {
+        let base = self.base_dir.clone();
+        move |name: &str| {
+            use clawhdf5_format::error::FormatError;
+            let p = std::path::Path::new(name);
+            if p.is_absolute()
+                || p.components()
+                    .any(|c| matches!(c, std::path::Component::ParentDir))
+            {
+                return Err(FormatError::ChunkedReadError(format!("refused {name}")));
+            }
+            match std::fs::read(base.join(p)) {
+                Ok(b) => Ok(Some(b)),
+                Err(err) if err.kind() == std::io::ErrorKind::NotFound => Ok(None),
+                Err(err) => Err(FormatError::ChunkedReadError(err.to_string())),
+            }
+        }
+    }
+
+    fn read_named_datatype(&self, h: &ObjectHeader) -> Result<(), String> {
+        let dtb = self
+            .payload(h, MessageType::Datatype)?
+            .ok_or("MissingMessage(Datatype)")?;
+        Datatype::parse_in_header(&dtb, h.version).map_err(e)?;
+        Ok(())
+    }
+
+    fn read_dataset(&self, h: &ObjectHeader, rec: &mut Map<String, Value>) -> Result<(), String> {
+        let dtb = self
+            .payload(h, MessageType::Datatype)?
+            .ok_or("MissingMessage(Datatype)")?;
+        let (dt, _) = Datatype::parse_in_header(&dtb, h.version).map_err(e)?;
+        rec.insert("dtype".into(), Value::String(dtype_str(&dt)));
+        let dsb = self
+            .payload(h, MessageType::Dataspace)?
+            .ok_or("MissingMessage(Dataspace)")?;
+        let mut ds = Dataspace::parse(&dsb, self.ls).map_err(e)?;
+        // A virtual dataset's extent can come from its sources (unlimited /
+        // printf mappings), as h5py reports it, rather than the stored one.
+        if let Some(lm) = h
+            .messages
+            .iter()
+            .find(|m| m.msg_type == MessageType::DataLayout)
+            && let Ok(dl @ DataLayout::Virtual { .. }) =
+                DataLayout::parse(&lm.data, self.os, self.ls)
+        {
+            let resolver = self.vds_resolver();
+            ds.dimensions = clawhdf5_format::vds::virtual_dataset_extent(
+                self.data,
+                &dl,
+                &ds,
+                self.os,
+                self.ls,
+                Some(&resolver),
+            )
+            .map_err(e)?;
+        }
+        let (shape, n) = Self::shape(&ds);
+        rec.insert("shape".into(), shape);
+        if n.saturating_mul(dt.type_size() as u64) > MAX_BYTES {
+            rec.insert("skipped".into(), Value::String("too large".into()));
+            return Ok(());
+        }
+        let lm = h
+            .messages
+            .iter()
+            .find(|m| m.msg_type == MessageType::DataLayout)
+            .ok_or("MissingMessage(DataLayout)")?;
+        let dl = DataLayout::parse(&lm.data, self.os, self.ls).map_err(e)?;
+        rec.insert(
+            "layout".into(),
+            Value::String(
+                match &dl {
+                    DataLayout::Compact { .. } => "compact",
+                    DataLayout::Contiguous { .. } => "contiguous",
+                    DataLayout::Chunked { .. } => "chunked",
+                    DataLayout::Virtual { .. } => "virtual",
+                }
+                .into(),
+            ),
+        );
+        let pipeline = match self.payload(h, MessageType::FilterPipeline)? {
+            Some(p) => Some(FilterPipeline::parse(&p).map_err(e)?),
+            None => None,
+        };
+        if let Some(p) = &pipeline {
+            rec.insert(
+                "filters".into(),
+                json!(p.filters.iter().map(|f| f.filter_id).collect::<Vec<_>>()),
+            );
+        }
+        let raw = if matches!(dl, DataLayout::Virtual { .. }) {
+            let resolver = self.vds_resolver();
+            let fill = clawhdf5_format::fill_value::dataset_fill_value_in(
+                self.data,
+                &h.messages,
+                self.os,
+                self.ls,
+            )
+            .map_err(e)?;
+            clawhdf5_format::vds::read_virtual_dataset(
+                self.data,
+                &dl,
+                &ds,
+                &dt,
+                fill.as_deref(),
+                self.os,
+                self.ls,
+                Some(&resolver),
+            )
+            .map_err(e)?
+            .data
+        } else {
+            let cache = clawhdf5_format::chunk_cache::ChunkCache::new();
+            clawhdf5_format::fill_value::read_full_with_fill::<clawhdf5_format::error::FormatError>(
+                &h.messages,
+                self.data,
+                &dl,
+                &ds,
+                dt.type_size() as usize,
+                self.os,
+                self.ls,
+                || {
+                    data_read::read_raw_data_cached(
+                        self.data,
+                        &dl,
+                        &ds,
+                        &dt,
+                        pipeline.as_ref(),
+                        self.os,
+                        self.ls,
+                        &cache,
+                    )
+                },
+            )
+            .map_err(e)?
+        };
+        self.hash_values(&dt, &raw, n, rec)
+    }
+
+    fn attrs(&self, h: &ObjectHeader) -> Result<Map<String, Value>, String> {
+        let msgs = extract_attributes_full(self.data, h, self.os, self.ls).map_err(e)?;
+        let mut out = Map::new();
+        for a in &msgs {
+            let r = guarded(|| {
+                let mut rec = Map::new();
+                rec.insert("dtype".into(), Value::String(dtype_str(&a.datatype)));
+                let (shape, n) = Self::shape(&a.dataspace);
+                rec.insert("shape".into(), shape);
+                self.hash_values(&a.datatype, &a.raw_data, n, &mut rec)?;
+                Ok(rec)
+            });
+            let v = match r {
+                Ok(rec) => Value::Object(rec),
+                Err(msg) => json!({ "error": msg }),
+            };
+            out.insert(a.name.clone(), v);
+        }
+        Ok(out)
+    }
+
+    fn entries(&self, h: &ObjectHeader) -> Result<Vec<GroupEntry>, String> {
+        let v1 = h
+            .messages
+            .iter()
+            .find(|m| m.msg_type == MessageType::SymbolTable);
+        if let Some(m) = v1 {
+            let stm = SymbolTableMessage::parse(&m.data, self.os).map_err(e)?;
+            group_v1::resolve_v1_group_entries(self.data, &stm, self.os, self.ls).map_err(e)
+        } else if h
+            .messages
+            .iter()
+            .any(|m| m.msg_type == MessageType::LinkInfo || m.msg_type == MessageType::Link)
+        {
+            group_v2::resolve_v2_group_entries(self.data, h, self.os, self.ls).map_err(e)
+        } else {
+            Ok(Vec::new())
+        }
+    }
+}
+
+/// Element bytes as an unsigned integer (at most 16 bytes), honouring byte order.
+fn element_bits(b: &[u8], byte_order: &DatatypeByteOrder) -> Result<u128, String> {
+    if b.len() > 16 {
+        return Err(format!("canon: {}-byte numeric element", b.len()));
+    }
+    let mut v = 0u128;
+    match byte_order {
+        DatatypeByteOrder::LittleEndian => {
+            for (i, x) in b.iter().enumerate() {
+                v |= u128::from(*x) << (8 * i);
+            }
+        }
+        DatatypeByteOrder::BigEndian => {
+            for x in b {
+                v = (v << 8) | u128::from(*x);
+            }
+        }
+        DatatypeByteOrder::Vax => return Err("canon: VAX byte order".into()),
+    }
+    Ok(v)
+}
+
+fn field(v: u128, pos: u32, len: u32) -> u128 {
+    if len == 0 || pos >= 128 {
+        return 0;
+    }
+    let v = v >> pos;
+    if len >= 128 {
+        v
+    } else {
+        v & ((1u128 << len) - 1)
+    }
+}
+
+/// True when a float's bit fields are exactly IEEE 754 binary16/32/64 for its
+/// size. h5py hands back such a type's bytes untouched; any other layout (an
+/// N-Bit `H5Tset_precision` float, say) is *converted* by libhdf5 into the
+/// numpy float of the same size, so comparing raw bytes would be meaningless.
+fn ieee_layout(dt: &Datatype) -> bool {
+    let Datatype::FloatingPoint {
+        size,
+        bit_offset,
+        bit_precision,
+        exponent_location,
+        exponent_size,
+        mantissa_location,
+        mantissa_size,
+        exponent_bias,
+        ..
+    } = dt
+    else {
+        return true;
+    };
+    let std = match size {
+        2 => (16, 10, 5, 10, 15),
+        4 => (32, 23, 8, 23, 127),
+        8 => (64, 52, 11, 52, 1023),
+        _ => return true, // no same-size numpy float to convert to: compare raw
+    };
+    *bit_offset == 0
+        && (
+            *bit_precision,
+            *exponent_location,
+            *exponent_size,
+            *mantissa_size,
+            *exponent_bias,
+        ) == (std.0, std.1, std.2, std.3, std.4)
+        && *mantissa_location == 0
+}
+
+/// Canonicalise a non-IEEE-layout float the way libhdf5's float->float
+/// conversion presents it to h5py: as the IEEE float of the same size.
+/// Assumes the implied-leading-one normalisation and the sign bit at the top
+/// of the precision (what `H5Tset_precision` produces; the parser does not
+/// keep either field).
+fn canon_custom_float(dt: &Datatype, b: &[u8], out: &mut Vec<u8>) -> Result<(), String> {
+    let Datatype::FloatingPoint {
+        size,
+        byte_order,
+        bit_offset,
+        bit_precision,
+        exponent_location,
+        exponent_size,
+        mantissa_location,
+        mantissa_size,
+        exponent_bias,
+    } = dt
+    else {
+        unreachable!()
+    };
+    let (esize, msize) = (u32::from(*exponent_size), u32::from(*mantissa_size));
+    if esize == 0 || esize > 30 || msize > 64 {
+        return Err(format!("canon: unsupported float layout e{esize} m{msize}"));
+    }
+    let v = element_bits(b, byte_order)?;
+    let sign_pos = (u32::from(*bit_offset) + u32::from(*bit_precision)).saturating_sub(1);
+    let neg = field(v, sign_pos, 1) == 1;
+    let e = field(v, u32::from(*exponent_location), esize) as i64;
+    let m = field(v, u32::from(*mantissa_location), msize);
+    let emax = (1i64 << esize) - 1;
+    let bias = i64::from(*exponent_bias);
+    let mag = if e == emax {
+        if m == 0 { f64::INFINITY } else { f64::NAN }
+    } else if e == 0 {
+        (m as f64) * 2f64.powi((1 - bias - msize as i64) as i32)
+    } else {
+        ((1u128 << msize) as f64 + m as f64) * 2f64.powi((e - bias - msize as i64) as i32)
+    };
+    let x = if neg { -mag } else { mag };
+    match size {
+        2 => out
+            .extend_from_slice(&clawhdf5_format::float16::f32_to_f16_bits(x as f32).to_le_bytes()),
+        4 => out.extend_from_slice(&(x as f32).to_le_bytes()),
+        8 => out.extend_from_slice(&x.to_le_bytes()),
+        _ => unreachable!("ieee_layout keeps other sizes raw"),
+    }
+    Ok(())
+}
+
+/// Integers stored with a bit offset or reduced precision (N-Bit): libhdf5
+/// converts them to the full-width integer of the same size, shifting the
+/// value down and sign-extending from the top precision bit.
+fn canon_partial_int(dt: &Datatype, b: &[u8], out: &mut Vec<u8>) -> Result<(), String> {
+    let Datatype::FixedPoint {
+        size,
+        byte_order,
+        signed,
+        bit_offset,
+        bit_precision,
+    } = dt
+    else {
+        unreachable!()
+    };
+    let prec = u32::from(*bit_precision);
+    let v = element_bits(b, byte_order)?;
+    let mut x = field(v, u32::from(*bit_offset), prec);
+    if *signed && prec > 0 && prec < 128 && field(x, prec - 1, 1) == 1 {
+        x |= !0u128 << prec;
+    }
+    out.extend_from_slice(&x.to_le_bytes()[..*size as usize]);
+    Ok(())
+}
+
+fn partial_int(dt: &Datatype) -> bool {
+    matches!(dt, Datatype::FixedPoint { size, bit_offset, bit_precision, .. }
+        if *bit_offset != 0 || u32::from(*bit_precision) != size * 8)
+}
+
+fn canon_str(b: &[u8], out: &mut Vec<u8>) {
+    let cut = b.iter().position(|&c| c == 0).unwrap_or(b.len());
+    let mut s = &b[..cut];
+    while let [rest @ .., b' '] = s {
+        s = rest;
+    }
+    out.push(b'S');
+    out.extend_from_slice(&(s.len() as u32).to_le_bytes());
+    out.extend_from_slice(s);
+}
+
+fn hex(b: &[u8]) -> String {
+    b.iter().map(|x| format!("{x:02x}")).collect()
+}
+
+fn dtype_str(dt: &Datatype) -> String {
+    match dt {
+        Datatype::FixedPoint {
+            size,
+            signed,
+            byte_order,
+            ..
+        } => {
+            format!(
+                "{}{}{}",
+                bo(byte_order),
+                if *signed { "i" } else { "u" },
+                size
+            )
+        }
+        Datatype::FloatingPoint {
+            size, byte_order, ..
+        } => format!("{}f{}", bo(byte_order), size),
+        Datatype::BitField {
+            size, byte_order, ..
+        } => format!("{}b{}", bo(byte_order), size),
+        Datatype::Time { size, .. } => format!("time{size}"),
+        Datatype::String { size, .. } => format!("S{size}"),
+        Datatype::Opaque { size, .. } => format!("V{size}"),
+        Datatype::Compound { size, members } => format!(
+            "{{{}}}{size}",
+            members
+                .iter()
+                .map(|m| format!("{}:{}", m.name, dtype_str(&m.datatype)))
+                .collect::<Vec<_>>()
+                .join(",")
+        ),
+        Datatype::Reference { ref_type, .. } => format!("ref({ref_type:?})"),
+        Datatype::Enumeration { base_type, .. } => format!("enum({})", dtype_str(base_type)),
+        Datatype::VariableLength {
+            is_string: true, ..
+        } => "vlstr".into(),
+        Datatype::VariableLength { base_type, .. } => format!("vlen({})", dtype_str(base_type)),
+        Datatype::Array {
+            base_type,
+            dimensions,
+        } => format!("({}){dimensions:?}", dtype_str(base_type)),
+    }
+}
+
+fn bo(b: &DatatypeByteOrder) -> &'static str {
+    match b {
+        DatatypeByteOrder::LittleEndian => "<",
+        DatatypeByteOrder::BigEndian => ">",
+        DatatypeByteOrder::Vax => "vax",
+    }
+}
+
+fn is_group(h: &ObjectHeader) -> bool {
+    h.messages.iter().any(|m| {
+        matches!(
+            m.msg_type,
+            MessageType::LinkInfo | MessageType::Link | MessageType::SymbolTable
+        )
+    })
+}
+
+fn main() {
+    install_hook();
+    let path = std::env::args().nth(1).expect("usage: probe <file>");
+    let mut top = Map::new();
+    top.insert("file".into(), Value::String(path.clone()));
+    let data = match std::fs::read(&path) {
+        Ok(d) => d,
+        Err(err) => {
+            top.insert("open_error".into(), Value::String(format!("Io({err})")));
+            println!("{}", Value::Object(top));
+            return;
+        }
+    };
+    // Every address is relative to the superblock: look at the file from
+    // there on (past any user block), as libhdf5 does.
+    let hdf5: &[u8] = match signature::find_signature(&data) {
+        Ok(off) => &data[off..],
+        Err(_) => &data,
+    };
+    let sb = guarded(|| Superblock::parse(hdf5, 0).map_err(e));
+    let sb = match sb {
+        Ok(sb) => sb,
+        Err(msg) => {
+            top.insert("open_error".into(), Value::String(msg));
+            println!("{}", Value::Object(top));
+            return;
+        }
+    };
+    // libhdf5 refuses a truncated file and reads nothing past the recorded
+    // end of file.
+    let base = (data.len() - hdf5.len()) as u64;
+    let hdf5 = match sb.data_end(base, data.len() as u64) {
+        Ok(end) => &hdf5[..end as usize],
+        Err(err) => {
+            top.insert("open_error".into(), Value::String(e(err)));
+            println!("{}", Value::Object(top));
+            return;
+        }
+    };
+    top.insert("superblock_version".into(), json!(sb.version));
+    let ctx = Ctx {
+        data: hdf5,
+        os: sb.offset_size,
+        ls: sb.length_size,
+        base_dir: std::path::Path::new(&path)
+            .parent()
+            .map(|p| p.to_path_buf())
+            .unwrap_or_default(),
+        vl: RefCell::new(VlResolver::new(hdf5, sb.offset_size, sb.length_size)),
+    };
+    let mut objects: Vec<Value> = Vec::new();
+    let mut visited = HashSet::new();
+    let mut soft_v1 = 0u64;
+    // explicit DFS stack: (address, path)
+    let mut stack: Vec<(u64, String)> = vec![(sb.root_group_address, "/".to_string())];
+    while let Some((addr, p)) = stack.pop() {
+        if objects.len() >= MAX_OBJECTS {
+            top.insert("truncated".into(), json!(true));
+            break;
+        }
+        if !visited.insert(addr) {
+            continue;
+        }
+        let mut rec = Map::new();
+        rec.insert("path".into(), Value::String(p.clone()));
+        let r = guarded(|| {
+            let h = ctx.header(addr)?;
+            Ok(h)
+        });
+        let h = match r {
+            Ok(h) => h,
+            Err(msg) => {
+                rec.insert("kind".into(), Value::String("unknown".into()));
+                rec.insert("error".into(), Value::String(msg));
+                objects.push(Value::Object(rec));
+                continue;
+            }
+        };
+        let is_ds = h
+            .messages
+            .iter()
+            .any(|m| m.msg_type == MessageType::DataLayout);
+        let kind = if is_ds {
+            "dataset"
+        } else if is_group(&h) || addr == sb.root_group_address {
+            "group"
+        } else if h
+            .messages
+            .iter()
+            .any(|m| m.msg_type == MessageType::Datatype)
+        {
+            "datatype"
+        } else {
+            "unknown"
+        };
+        rec.insert("kind".into(), Value::String(kind.into()));
+        if kind == "dataset"
+            && let Err(msg) = guarded(|| ctx.read_dataset(&h, &mut rec))
+        {
+            rec.insert("error".into(), Value::String(msg));
+        }
+        // Opening a committed datatype decodes it (h5py's `f[name]` fails on
+        // one libhdf5 cannot decode), so decode it here too.
+        if kind == "datatype"
+            && let Err(msg) = guarded(|| ctx.read_named_datatype(&h))
+        {
+            rec.insert("error".into(), Value::String(msg));
+        }
+        if kind != "datatype" {
+            match guarded(|| ctx.attrs(&h)) {
+                Ok(m) => {
+                    rec.insert("attrs".into(), Value::Object(m));
+                }
+                Err(msg) => {
+                    rec.insert("attrs_error".into(), Value::String(msg));
+                }
+            }
+        }
+        if kind == "group" {
+            match guarded(|| ctx.entries(&h)) {
+                Ok(mut ents) => {
+                    ents.retain(|en| {
+                        if en.cache_type == 2 {
+                            soft_v1 += 1;
+                            false
+                        } else {
+                            true
+                        }
+                    });
+                    ents.sort_by(|a, b| a.name.cmp(&b.name));
+                    let base = if p == "/" { String::new() } else { p.clone() };
+                    for en in ents.into_iter().rev() {
+                        stack.push((en.object_header_address, format!("{base}/{}", en.name)));
+                    }
+                }
+                Err(msg) => {
+                    rec.insert("list_error".into(), Value::String(msg));
+                }
+            }
+        }
+        objects.push(Value::Object(rec));
+    }
+    if soft_v1 > 0 {
+        top.insert("v1_soft_link_entries".into(), json!(soft_v1));
+    }
+    top.insert("objects".into(), Value::Array(objects));
+    println!("{}", Value::Object(top));
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// The N-Bit float of libhdf5's `test/testfiles/le_data.h5`
+    /// (`Nbit_float_data_le`): offset 7, precision 20, sign bit 26, exponent
+    /// 20+6 (bias 31), mantissa 7+13.
+    fn nbit_f32(byte_order: DatatypeByteOrder) -> Datatype {
+        Datatype::FloatingPoint {
+            size: 4,
+            byte_order,
+            bit_offset: 7,
+            bit_precision: 20,
+            exponent_location: 20,
+            exponent_size: 6,
+            mantissa_location: 7,
+            mantissa_size: 13,
+            exponent_bias: 31,
+        }
+    }
+
+    fn canon_one(dt: &Datatype, bytes: &[u8]) -> Vec<u8> {
+        let mut out = Vec::new();
+        canon_custom_float(dt, bytes, &mut out).unwrap();
+        out
+    }
+
+    #[test]
+    fn nbit_float_canonicalises_to_the_value_libhdf5_returns() {
+        let le = nbit_f32(DatatypeByteOrder::LittleEndian);
+        let be = nbit_f32(DatatypeByteOrder::BigEndian);
+        assert!(!ieee_layout(&le));
+        // 1.0: exponent = bias, mantissa 0
+        let one: u32 = 31 << 20;
+        assert_eq!(canon_one(&le, &one.to_le_bytes()), 1.0f32.to_le_bytes());
+        assert_eq!(canon_one(&be, &one.to_be_bytes()), 1.0f32.to_le_bytes());
+        // -2.1999512 (h5py's reading of the file's -2.2): sign, e = 32, m = 819
+        let v: u32 = (1 << 26) | (32 << 20) | (819 << 7);
+        assert_eq!(
+            canon_one(&le, &v.to_le_bytes()),
+            (-2.199_951_2f32).to_le_bytes()
+        );
+        assert_eq!(canon_one(&le, &[0; 4]), 0.0f32.to_le_bytes());
+    }
+
+    #[test]
+    fn ieee_floats_keep_their_raw_bytes() {
+        let f32le = Datatype::FloatingPoint {
+            size: 4,
+            byte_order: DatatypeByteOrder::LittleEndian,
+            bit_offset: 0,
+            bit_precision: 32,
+            exponent_location: 23,
+            exponent_size: 8,
+            mantissa_location: 0,
+            mantissa_size: 23,
+            exponent_bias: 127,
+        };
+        assert!(ieee_layout(&f32le));
+    }
+
+    #[test]
+    fn partial_precision_int_is_shifted_and_sign_extended() {
+        let dt = Datatype::FixedPoint {
+            size: 4,
+            byte_order: DatatypeByteOrder::BigEndian,
+            signed: true,
+            bit_offset: 4,
+            bit_precision: 17,
+        };
+        assert!(partial_int(&dt));
+        let stored = (((-5i32) as u32) & 0x1_FFFF) << 4;
+        let mut out = Vec::new();
+        canon_partial_int(&dt, &stored.to_be_bytes(), &mut out).unwrap();
+        assert_eq!(out, (-5i32).to_le_bytes());
+    }
+}

conformance/ref.py

@@ -0,0 +1,259 @@
+#!/usr/bin/env python3
+"""Reference probe: same JSON as the Rust `conformance-probe`, produced with h5py.
+
+Walk: iterative DFS from '/', children in sorted (UTF-8 byte) name order, hard
+links only, each object once (first path wins, deduplicated by object identity).
+Canonical value encoding: see harness/src/main.rs.
+"""
+import hashlib
+import json
+import os
+import struct
+import sys
+
+import numpy as np
+import h5py
+
+try:
+    import hdf5plugin  # noqa: F401  registers blosc/lz4/zstd/bzip2/... filters
+except Exception:  # pragma: no cover
+    pass
+
+MAX_BYTES = 200 * 1024 * 1024
+MAX_OBJECTS = 200_000
+
+
+def canon_str(b, out):
+    if isinstance(b, str):
+        b = b.encode("utf-8", "surrogateescape")
+    b = bytes(b)
+    cut = b.find(b"\x00")
+    if cut >= 0:
+        b = b[:cut]
+    b = b.rstrip(b" ")
+    out += b"S" + struct.pack("<I", len(b)) + b
+
+
+def simple(dt):
+    if dt.fields:
+        return all(simple(dt.fields[n][0]) for n in dt.names)
+    if dt.subdtype:
+        return simple(dt.subdtype[0])
+    return dt.kind in "iufcbV"
+
+
+def packed(dt):
+    if dt.fields:
+        return np.dtype([(n, packed(dt.fields[n][0])) for n in dt.names])
+    if dt.subdtype:
+        base, shape = dt.subdtype
+        return np.dtype((packed(base), shape))
+    if dt.kind in "iufcb":
+        return dt.newbyteorder("<")
+    return dt
+
+
+def canon_el(dt, val, out):
+    if dt.fields:
+        for n in dt.names:
+            canon_el(dt.fields[n][0], val[n], out)
+        return
+    if dt.subdtype:
+        base, _ = dt.subdtype
+        for x in np.asarray(val).reshape(-1):
+            canon_el(base, x, out)
+        return
+    k = dt.kind
+    if k in "iufcb":
+        out += np.asarray(val, dtype=dt).astype(dt.newbyteorder("<")).tobytes()
+    elif k == "V":
+        out += np.asarray(val, dtype=dt).tobytes()
+    elif k == "S":
+        canon_str(val, out)
+    elif k == "O":
+        if h5py.check_string_dtype(dt) is not None:
+            canon_str(val if val is not None else b"", out)
+        elif h5py.check_ref_dtype(dt) is not None:
+            out += b"R"
+        else:
+            base = h5py.check_vlen_dtype(dt)
+            if base is None:
+                raise TypeError(f"unhandled object dtype {dt!r}")
+            arr = np.asarray(val if val is not None else [], dtype=base).reshape(-1)
+            out += b"V" + struct.pack("<I", arr.shape[0])
+            if simple(base):
+                out += arr.astype(packed(base)).tobytes()
+            else:
+                for x in arr:
+                    canon_el(base, x, out)
+    elif k == "U":
+        canon_str(str(val), out)
+    else:
+        raise TypeError(f"unhandled dtype kind {k} ({dt!r})")
+
+
+def has_obj(dt):
+    if dt.fields:
+        return any(has_obj(dt.fields[n][0]) for n in dt.names)
+    if dt.subdtype:
+        return has_obj(dt.subdtype[0])
+    return dt.kind == "O"
+
+
+def note_conversion(tid, dt, rec):
+    """h5py converts some file types (FP8, bfloat16, x87 long double, ...) to a
+    different-sized numpy type; then value bytes are not comparable."""
+    try:
+        if not has_obj(dt) and tid.get_size() != dt.itemsize:
+            rec["converted"] = f"file type size {tid.get_size()} -> numpy {dt} ({dt.itemsize})"
+    except Exception:  # noqa: BLE001
+        pass
+
+
+def hash_values(arr, dt, rec):
+    if dt.subdtype is not None:
+        # h5py expands an HDF5 array element type into trailing array dims
+        dt = dt.subdtype[0]
+    arr = np.asarray(arr, dtype=dt)
+    if simple(dt):
+        c = np.ascontiguousarray(arr).astype(packed(dt)).tobytes()
+    else:
+        out = bytearray()
+        for x in arr.reshape(-1):
+            canon_el(dt, x, out)
+        c = bytes(out)
+    rec["hash"] = hashlib.sha256(c).hexdigest()
+    rec["head"] = c[:48].hex()
+
+
+def err(e):
+    s = f"{type(e).__name__}: {e}"
+    return s.splitlines()[0][:400] if s else type(e).__name__
+
+
+def shape_of(s):
+    return "null" if s is None else list(s)
+
+
+def n_bytes(shape, tid):
+    n = 1
+    for d in shape or ():
+        n *= d
+    return n * tid.get_size()
+
+
+def read_attrs(obj):
+    out = {}
+    names = sorted(obj.attrs.keys(), key=lambda s: s.encode("utf-8", "surrogateescape"))
+    for name in names:
+        rec = {}
+        try:
+            aid = obj.attrs.get_id(name)
+            rec["dtype"] = str(aid.dtype)
+            rec["shape"] = shape_of(aid.shape)
+            note_conversion(aid.get_type(), aid.dtype, rec)
+            if aid.shape is None:
+                hash_values(np.empty((0,), dtype=aid.dtype), aid.dtype, rec)
+            else:
+                val = obj.attrs[name]
+                hash_values(val, aid.dtype, rec)
+        except Exception as e:  # noqa: BLE001
+            rec = {"error": err(e)}
+        out[name] = rec
+    return out
+
+
+def main(path):
+    top = {"file": path}
+    try:
+        f = h5py.File(path, "r")
+    except Exception as e:  # noqa: BLE001
+        top["open_error"] = err(e)
+        print(json.dumps(top))
+        return
+    objects = []
+    seen = set()
+    stack = [("/", None)]
+    while stack:
+        p, obj = stack.pop()
+        if len(objects) >= MAX_OBJECTS:
+            top["truncated"] = True
+            break
+        rec = {"path": p}
+        try:
+            if obj is None:
+                obj = f[p]
+            key = hash(obj.id)  # h5py ObjectID hash = (fileno, object address/token)
+        except Exception as e:  # noqa: BLE001
+            rec["kind"] = "unknown"
+            rec["error"] = err(e)
+            objects.append(rec)
+            continue
+        if key in seen:
+            continue
+        seen.add(key)
+        if isinstance(obj, h5py.Dataset):
+            kind = "dataset"
+        elif isinstance(obj, h5py.Group):
+            kind = "group"
+        elif isinstance(obj, h5py.Datatype):
+            kind = "datatype"
+        else:
+            kind = "unknown"
+        rec["kind"] = kind
+        if kind == "dataset":
+            try:
+                dt = obj.dtype
+                rec["dtype"] = str(dt)
+                rec["shape"] = shape_of(obj.shape)
+                note_conversion(obj.id.get_type(), dt, rec)
+                if obj.shape is None:
+                    hash_values(np.empty((0,), dtype=dt), dt, rec)
+                elif n_bytes(obj.shape, obj.id.get_type()) > MAX_BYTES:
+                    rec["skipped"] = "too large"
+                else:
+                    arr = np.empty(obj.shape, dtype=dt)
+                    if arr.size:
+                        try:
+                            obj.read_direct(arr)
+                        except Exception:  # noqa: BLE001
+                            arr = obj[()]
+                    hash_values(arr, dt, rec)
+            except Exception as e:  # noqa: BLE001
+                rec["error"] = err(e)
+        if kind != "datatype":
+            try:
+                rec["attrs"] = read_attrs(obj)
+            except Exception as e:  # noqa: BLE001
+                rec["attrs_error"] = err(e)
+        if kind == "group":
+            try:
+                names = sorted(obj.keys(), key=lambda s: s.encode("utf-8", "surrogateescape"))
+                base = "" if p == "/" else p
+                kids = []
+                for n in names:
+                    try:
+                        link = obj.get(n, getlink=True)
+                    except Exception:  # noqa: BLE001
+                        link = None
+                    if link is not None and not isinstance(link, h5py.HardLink):
+                        continue
+                    kids.append(f"{base}/{n}")
+                for k in reversed(kids):
+                    stack.append((k, None))
+            except Exception as e:  # noqa: BLE001
+                rec["list_error"] = err(e)
+        objects.append(rec)
+    top["objects"] = objects
+    print(json.dumps(top), flush=True)
+    # Exit without tearing down the h5py objects: freeing them for some files
+    # that hold references (hdf5's h5repack_attr_refs.h5, cve-2024-32623.h5)
+    # makes libhdf5 2.0 abort with "free(): chunks in smallbin corrupted"
+    # about half the time. That happens after the reading is done, so it says
+    # nothing about what h5py read, but it flipped those files between ok and
+    # h5py-cannot-read from one run to the next.
+    os._exit(0)
+
+
+if __name__ == "__main__":
+    main(sys.argv[1])

conformance/report.py

@@ -0,0 +1,335 @@
+#!/usr/bin/env python3
+"""report.py <results_dir> <CONFORMANCE.md> <corpus_dir>
+
+Render the sweep's results (compare.py's results.json plus the raw per-side
+runs) as CONFORMANCE.md, and write <results_dir>/report-meta.json (commit,
+date, versions) for check.py --update.
+"""
+import collections
+import datetime
+import json
+import os
+import platform
+
+import subprocess
+import sys
+
+import h5py
+import numpy
+
+try:
+    import hdf5plugin
+    HDF5PLUGIN = hdf5plugin.version
+except Exception:  # noqa: BLE001
+    HDF5PLUGIN = "not installed"
+
+R, OUT_MD, CORPUS = sys.argv[1], sys.argv[2], sys.argv[3]
+HERE = os.path.dirname(os.path.abspath(__file__))
+ROOT = os.path.dirname(HERE)
+CLASSES = ["ok", "our-error", "mismatch", "h5py-cannot-read", "panic", "hang", "crash", "oom"]
+
+
+def sh(*cmd, cwd=ROOT):
+    try:
+        return subprocess.run(cmd, cwd=cwd, capture_output=True, text=True, timeout=30).stdout.strip()
+    except Exception:  # noqa: BLE001
+        return ""
+
+
+def cpu_model():
+    try:
+        for ln in open("/proc/cpuinfo"):
+            if ln.startswith(("model name", "Model")):
+                return ln.split(":", 1)[1].strip()
+    except OSError:
+        pass
+    return platform.processor() or "unknown"
+
+
+def mem_gib():
+    try:
+        for ln in open("/proc/meminfo"):
+            if ln.startswith("MemTotal:"):
+                return f"{int(ln.split()[1]) / 1048576:.0f} GiB"
+    except OSError:
+        pass
+    return "?"
+
+
+res = json.load(open(os.path.join(R, "results.json")))
+meta_run = json.load(open(os.path.join(R, "meta.json"))) if os.path.exists(os.path.join(R, "meta.json")) else {}
+rows = res["rows"]
+issues = res.get("issues", {})
+
+# safe.directory: a checkout owned by another user (a container) is still ours to read
+commit = sh("git", "-c", "safe.directory=*", "rev-parse", "HEAD") or os.environ.get("GITHUB_SHA", "unknown")
+lib_dirty = sh("git", "-c", "safe.directory=*", "status", "--porcelain", "--", "crates", "Cargo.toml")
+h5dump_v = sh("h5dump", "--version").replace("h5dump: ", "")
+meta = {
+    "date": datetime.datetime.now(datetime.timezone.utc).strftime("%Y-%m-%d %H:%M UTC"),
+    "commit": commit + (" (library sources modified)" if lib_dirty else ""),
+    "reference": f"h5py {h5py.__version__} / HDF5 {h5py.version.hdf5_version}",
+}
+json.dump(meta, open(os.path.join(R, "report-meta.json"), "w"), indent=1)
+
+pins = []
+for ln in open(os.path.join(HERE, "corpus.txt")):
+    if ln.strip() and not ln.lstrip().startswith("#"):
+        name, url, rev, root, *_ = ln.split()
+        pins.append((name, url, rev, root))
+
+by_corpus = collections.defaultdict(collections.Counter)
+for r in rows:
+    by_corpus[r["corpus"]][r["class"]] += 1
+total = collections.Counter(r["class"] for r in rows)
+
+
+def ex_list(files, n=3):
+    s = ", ".join(f"`{f}`" for f in files[:n])
+    return s + (f" (+{len(files) - n} more)" if len(files) > n else "")
+
+
+# --- known causes that are not clawhdf5 bugs --------------------------------
+def is_h5py_be_vlen(i):
+    """h5py returns the elements of a VL sequence of a big-endian base type
+    with their file (big-endian) bytes but a native-endian dtype."""
+    return (i["kind"] == "mismatch" and i["key"] in ("values", "attr-values")
+            and (i.get("ref_dtype") == "object") and (i.get("ours_dtype") or "").startswith("vlen(")
+            and ">" in (i.get("ours_dtype") or ""))
+
+
+known = collections.defaultdict(list)
+for r in rows:
+    if r["class"] != "mismatch":
+        continue
+    iss = issues.get(r["file"], [])
+    if iss and all(is_h5py_be_vlen(i) for i in iss):
+        known["h5py-be-vlen"].append(r["file"])
+
+
+# --- the CVE corpus: clawhdf5 vs h5dump vs h5py ------------------------------
+def side(run, name):
+    p = os.path.join(R, "runs", run, name)
+    if not os.path.exists(p + ".rc"):
+        return None
+    rc = int(open(p + ".rc").read().strip() or -1)
+    err = open(p + ".err", errors="replace").read()
+    try:
+        j = json.load(open(p + ".json"))
+    except Exception:  # noqa: BLE001
+        j = None
+    return rc, err, j
+
+
+def outcome(s, rust=False):
+    """-> (bucket, text). bucket in read / error / panic / crash / hang / oom."""
+    if s is None:
+        return "missing", "not run"
+    rc, err, j = s
+    if rc in (137, 124):
+        return "hang", "hang (killed at timeout)"
+    if "memory allocation of" in err or "MemoryError" in err or "bad_alloc" in err or "Cannot allocate" in err:
+        return "oom", "out of memory"
+    if rust and (rc == 101 or "PANIC:" in err):
+        return "panic", "panic"
+    if "overflowed its stack" in err:
+        return "crash", "stack overflow"
+    if rc == 139:
+        return "crash", "SIGSEGV"
+    if rc == 134:
+        return "crash", "SIGABRT" + (" (heap corruption)" if ("corrupted" in err or "free()" in err) else "")
+    if rc > 128:
+        return "crash", f"signal {rc - 128}"
+    if j is None:
+        return ("error", "error exit") if rc in (0, 1) else ("crash", f"exit {rc}")
+    if "open_error" in j:
+        return "error", "open error"
+    objs = j.get("objects", [])
+    ne = sum(1 for o in objs for k in ("error", "attrs_error", "list_error") if k in o)
+    ne += sum(1 for o in objs for a in (o.get("attrs") or {}).values() if "error" in a)
+    return "read", f"read {len(objs)} obj" + (f", {ne} errors" if ne else "")
+
+
+def h5dump_outcome(s):
+    if s is None:
+        return "missing", "not run"
+    rc, err, _ = s
+    if rc in (137, 124):
+        return "hang", "hang (killed at timeout)"
+    if "memory allocation" in err or "Cannot allocate" in err:
+        return "oom", "out of memory"
+    if rc == 139:
+        return "crash", "SIGSEGV"
+    if rc == 134:
+        return "crash", "SIGABRT" + (" (heap corruption)" if ("corrupted" in err or "free()" in err) else "")
+    if rc > 128:
+        return "crash", f"signal {rc - 128}"
+    return ("read", "ok") if rc == 0 else ("error", "error exit")
+
+
+cve_rows = []
+buckets = {"clawhdf5": collections.Counter(), "h5dump": collections.Counter(), "h5py": collections.Counter()}
+ours_panic = {r["file"] for r in rows if r["class"] == "panic"}
+for r in rows:
+    if r["corpus"] != "cve_hdf5":
+        continue
+    run = r["file"].replace("/", "__")
+    o = outcome(side(run, "ours"), rust=True)
+    if o[0] == "read" and r["file"] in ours_panic:
+        o = ("panic", "caught panic")
+    p = outcome(side(run, "ref"))
+    d = h5dump_outcome(side(run, "h5dump"))
+    buckets["clawhdf5"][o[0]] += 1
+    buckets["h5py"][p[0]] += 1
+    buckets["h5dump"][d[0]] += 1
+    cve_rows.append((r["file"].split("/", 1)[1], d[1], p[1], o[1], r["class"]))
+
+# --- render -----------------------------------------------------------------
+L = []
+w = L.append
+w("# clawhdf5 conformance report")
+w("")
+w("Every HDF5 file of eight public corpora (pinned by commit) is read twice — by")
+w("clawhdf5 (`conformance/probe`, the same `clawhdf5-format` calls the facade")
+w("makes) and by h5py/libhdf5 (`conformance/ref.py`) — and the two readings are")
+w("compared object by object: the set of hard-linked objects, each dataset's and")
+w("attribute's shape, and a SHA-256 of its values in a canonical encoding. The")
+w("CVE corpus is also run through `h5dump`. Each side runs under a timeout and an")
+w("address-space limit, so a hang, crash or runaway allocation is recorded, not")
+w("fatal. This file is generated by `conformance/run.sh`; do not edit it by hand.")
+w("")
+w("## Run")
+w("")
+w("| | |")
+w("|---|---|")
+w(f"| date | {meta['date']} |")
+w(f"| clawhdf5 commit | `{meta['commit']}` |")
+w(f"| machine | `{platform.node()}`: {cpu_model()}, {os.cpu_count()} CPUs, {mem_gib()}, {platform.system()} {platform.release()} {platform.machine()} |")
+w(f"| command | `{os.environ.get('CONFORMANCE_CMD', 'conformance/run.sh')}` |")
+w(f"| rustc | {sh('rustc', '-V')} |")
+w(f"| reference | h5py {h5py.__version__}, HDF5 {h5py.version.hdf5_version}, numpy {numpy.__version__}, hdf5plugin {HDF5PLUGIN}, Python {platform.python_version()} |")
+w(f"| h5dump | {h5dump_v} (CVE corpus only) |")
+if meta_run:
+    w(f"| limits | {meta_run.get('timeout_s')} s timeout (SIGKILL), {int(meta_run.get('mem_kb', 0)) // 1024} MiB address space, per process; {meta_run.get('jobs')} files in parallel |")
+    w(f"| runtime | {meta_run.get('probe_seconds')} s probing + comparing ({meta_run.get('build_seconds')} s fetch/build before it) |")
+w("")
+w("## Results")
+w("")
+w("A file's class is the first that applies:")
+w("")
+w("- **panic / hang / crash / oom** — clawhdf5 panicked (caught per object or not), hit the timeout, died on a signal, or failed an allocation. The CI gate fails on any of these.")
+w("- **h5py-cannot-read** — libhdf5 could not open the file (or itself crashed or hung). Nothing to compare against; most are the deliberately malformed CVE reproducers.")
+w("- **our-error** — clawhdf5 returned an error for something h5py reads.")
+w("- **mismatch** — both read it, but the shapes, values, object set or attribute set differ.")
+w("- **ok** — every object h5py reads, clawhdf5 reads identically.")
+w("")
+w("| corpus | files | " + " | ".join(CLASSES) + " |")
+w("|---" * (len(CLASSES) + 2) + "|")
+for c in sorted(by_corpus):
+    cnt = by_corpus[c]
+    w(f"| {c} | {sum(cnt.values())} | " + " | ".join(str(cnt.get(k, 0)) for k in CLASSES) + " |")
+w(f"| **all** | **{len(rows)}** | " + " | ".join(f"**{total.get(k, 0)}**" for k in CLASSES) + " |")
+w("")
+n_known = sum(len(v) for v in known.values())
+if n_known:
+    w(f"{n_known} of the {total.get('mismatch', 0)} mismatches are a known h5py bug, not ours (see *Known not-our-bug*).")
+    w("")
+w("Corpora (fetched by `conformance/fetch-corpus.sh` into the gitignored `conformance/.cache/`):")
+w("")
+w("| corpus | source | commit |")
+w("|---|---|---|")
+for name, url, rev, root in pins:
+    w(f"| {name} | {url.removesuffix('.git')}" + ("" if root == "." else f" (`{root}`)") + f" | `{rev[:12]}` |")
+w("")
+
+w("## Panics, hangs, crashes, out-of-memory")
+w("")
+if not res["panics"]:
+    w("None.")
+else:
+    for p in res["panics"]:
+        w(f"- `{p['file']}` [{p['class']}] {p['detail']}")
+w("")
+
+w("## Our-error root causes")
+w("")
+w("Grouped by normalised error message. *files* counts files whose class this cause affects.")
+w("")
+w("| files | objects | error | examples |")
+w("|---:|---:|---|---|")
+for k, v in res["root_causes"].items():
+    w(f"| {v['files']} | {v['count']} | `{k.replace('|', '/')}` | {ex_list(v['file_list'])} |")
+w("")
+w("## Mismatch root causes")
+w("")
+w("| files | objects | cause | examples |")
+w("|---:|---:|---|---|")
+for k, v in res["mismatch_causes"].items():
+    w(f"| {v['files']} | {v['count']} | `{k.replace('|', '/')}` | {ex_list(v['file_list'])} |")
+w("")
+
+w("## CVE corpus: clawhdf5 vs h5dump vs h5py")
+w("")
+w(f"The {len(cve_rows)} files of [HDFGroup/cve_hdf5](https://github.com/HDFGroup/cve_hdf5) — reproducers for")
+w("published libhdf5 CVEs and fuzzer finds. *read* = produced output (possibly with per-object")
+w("errors), *error* = refused cleanly. h5dump exits non-zero on any error anywhere in a file, so")
+w("its read/error split is not comparable with the other two rows; the panic, crash, hang and oom")
+w("columns are.")
+w("")
+w("| tool | read | error | panic | crash | hang | oom |")
+w("|---|---:|---:|---:|---:|---:|---:|")
+for tool, label in (("clawhdf5", "clawhdf5"), ("h5dump", f"h5dump {h5dump_v.split()[-1] if h5dump_v else ''}"),
+                    ("h5py", f"h5py {h5py.__version__} / HDF5 {h5py.version.hdf5_version}")):
+    b = buckets[tool]
+    w(f"| {label} | " + " | ".join(str(b.get(k, 0)) for k in ("read", "error", "panic", "crash", "hang", "oom")) + " |")
+w("")
+w("<details><summary>Per-file outcomes</summary>")
+w("")
+w("| file | h5dump | h5py | clawhdf5 | class |")
+w("|---|---|---|---|---|")
+for f, d, p, o, cls in cve_rows:
+    w(f"| {f} | {d} | {p} | {o} | {cls} |")
+w("")
+w("</details>")
+w("")
+
+w("## Known not-our-bug")
+w("")
+w("- **h5py big-endian variable-length sequences.** h5py returns the elements of a VL sequence")
+w("  whose base type is big-endian with the file's big-endian bytes but a native (little-endian)")
+w("  numpy dtype, so the values it reports are byte-swapped garbage; `h5dump` prints the values")
+w("  clawhdf5 reads. Reproducer: `h5py.vlen_dtype(np.dtype('>f4'))` dataset holding `[1.0, 2.0]`")
+w("  reads back in h5py as `[4.6e-41, 9.0e-44]`. Affected here: "
+  + (ex_list(sorted(known["h5py-be-vlen"]), 10) if known["h5py-be-vlen"] else "none") + ".")
+w("- **Non-IEEE floats and partial-precision integers (N-Bit).** libhdf5 converts a float whose")
+w("  bit layout is not IEEE (e.g. `H5Tset_precision` for the N-Bit filter) or an integer with a")
+w("  bit offset / reduced precision into the plain numpy type of the same size. The probe")
+w("  compares such values as converted numbers, not raw file bytes (before 2026-09-25 it compared")
+w("  raw bytes, which reported every N-Bit float dataset as a mismatch).")
+if res["incomparable"]:
+    w("- **Types h5py widens.** Where h5py reads a type into a numpy type of a different size")
+    w("  (FP8 -> float16, bfloat16 -> float32, x87 long double -> float128) the values are not")
+    w("  compared (shape and presence still are): "
+      + ", ".join(f"{k} ({n}x)" for k, n in res["incomparable"]) + ".")
+w("- **References** are compared by presence only (`R`), not by target.")
+w("")
+if res.get("ref_only_errors"):
+    w("## Objects h5py fails on but clawhdf5 reads")
+    w("")
+    for k, n in res["ref_only_errors"][:15]:
+        w(f"- {n} x `{k}`")
+    w("")
+w("## Reproduce")
+w("")
+w("```sh")
+w("# needs: Rust, python3 with h5py numpy hdf5plugin (conformance/requirements.txt), h5dump (hdf5-tools), git")
+w("CLAWHDF5_PYTHON=/path/to/venv/bin/python conformance/run.sh")
+w("```")
+w("")
+w("The corpus (about 450 MB of sparse checkouts) is cached in `conformance/.cache/`; results for")
+w("every file, both sides' raw JSON and stderr, are in `conformance/.cache/results/`.")
+w("`conformance/baseline.json` holds the ok files the nightly CI job (`.gitea/workflows/conformance.yml`)")
+w("must keep; `conformance/run.sh --update-baseline` rewrites it.")
+
+with open(OUT_MD, "w") as fh:
+    fh.write("\n".join(L) + "\n")

conformance/requirements.txt

@@ -0,0 +1,6 @@
+# The reference side of the conformance sweep. Pinned so the nightly job and a
+# local run compare against the same libhdf5 (h5py wheels bundle it).
+h5py==3.16.0
+numpy==2.5.3
+hdf5plugin==7.1.0
+netCDF4==1.7.4

conformance/run.sh

@@ -0,0 +1,88 @@
+#!/usr/bin/env bash
+# conformance/run.sh — the clawhdf5 conformance sweep, end to end.
+#
+#   fetch the pinned corpora (cached) -> build the probe -> probe every file
+#   with clawhdf5 and with h5py (and h5dump for the CVE corpus), each under a
+#   timeout and a memory limit -> compare -> write CONFORMANCE.md -> check the
+#   result against conformance/baseline.json.
+#
+# Usage: conformance/run.sh [--no-fetch] [--no-report] [--update-baseline]
+#
+# Environment:
+#   CLAWHDF5_PYTHON    python with h5py, numpy, hdf5plugin (default: repo .venv, then python3)
+#   CONFORMANCE_CACHE  corpus / build / results cache (default: conformance/.cache)
+#   CONFORMANCE_OUT    results directory (default: $CONFORMANCE_CACHE/results)
+#   CONFORMANCE_REPORT report path (default: CONFORMANCE.md at the repo root)
+#   JOBS               parallel files (default: nproc)
+#   CONFORMANCE_PROBE  use this prebuilt probe binary instead of building one
+#   TMO / MEM_KB       per-process timeout in seconds (20) / address-space limit in KiB (4 GiB)
+#
+# Exit status: 0 = gate passed; 1 = a panic/hang/crash/oom in clawhdf5, or the
+# ok count fell below the baseline, or a baseline-ok file regressed; 2 = setup error.
+set -euo pipefail
+HERE="$(cd "$(dirname "$0")" && pwd)"
+ROOT="$(cd "$HERE/.." && pwd)"
+FETCH=1 REPORT=1 UPDATE=0
+for a in "$@"; do
+  case "$a" in
+    --no-fetch) FETCH=0 ;;
+    --no-report) REPORT=0 ;;
+    --update-baseline) UPDATE=1 ;;
+    -h|--help) sed -n '2,23p' "$0"; exit 0 ;;
+    *) echo "unknown argument: $a" >&2; exit 2 ;;
+  esac
+done
+
+export PATH="$HOME/.cargo/bin:$PATH"
+CACHE="${CONFORMANCE_CACHE:-$HERE/.cache}"
+mkdir -p "$CACHE"; CACHE="$(cd "$CACHE" && pwd)"
+OUT="${CONFORMANCE_OUT:-$CACHE/results}"
+REPORT_PATH="${CONFORMANCE_REPORT:-$ROOT/CONFORMANCE.md}"
+JOBS="${JOBS:-$(nproc 2>/dev/null || echo 4)}"
+if [ -n "${CLAWHDF5_PYTHON:-}" ]; then PY="$CLAWHDF5_PYTHON"
+elif [ -x "$ROOT/.venv/bin/python" ]; then PY="$ROOT/.venv/bin/python"
+else PY="$(command -v python3)"; fi
+export PY TMO="${TMO:-20}" MEM_KB="${MEM_KB:-4194304}"
+command -v h5dump >/dev/null || { echo "error: h5dump not found (install hdf5-tools)" >&2; exit 2; }
+"$PY" -c 'import h5py, numpy, hdf5plugin' || { echo "error: $PY lacks h5py/numpy/hdf5plugin" >&2; exit 2; }
+
+t0=$(date +%s)
+[ "$FETCH" = 1 ] && bash "$HERE/fetch-corpus.sh" "$CACHE"
+C="$CACHE/corpus"
+[ -d "$C" ] || { echo "error: no corpus in $C (run without --no-fetch)" >&2; exit 2; }
+
+if [ -n "${CONFORMANCE_PROBE:-}" ]; then
+  export PROBE="$CONFORMANCE_PROBE"   # a prebuilt probe, e.g. an older one for a before/after
+else
+  echo "== building the probe"
+  CARGO_TARGET_DIR="${CARGO_TARGET_DIR:-$CACHE/target}" \
+    cargo build -q --release --manifest-path "$HERE/probe/Cargo.toml"
+  export PROBE="${CARGO_TARGET_DIR:-$CACHE/target}/release/conformance-probe"
+fi
+t1=$(date +%s)
+
+rm -rf "$OUT"; mkdir -p "$OUT"
+"$PY" "$HERE/list_files.py" "$C" > "$OUT/files.txt"
+echo "== probing $(wc -l <"$OUT/files.txt") files, $JOBS at a time (timeout ${TMO}s, limit $((MEM_KB / 1024)) MiB)"
+export C OUT HERE
+# The shell's "Segmentation fault (core dumped)" notices go to probe.log; the
+# signals themselves are recorded in each side's .rc.
+xargs -a "$OUT/files.txt" -d '\n' -P "$JOBS" -I{} bash -c '
+  f="$1"; d="$OUT/runs/${f//\//__}"
+  case "$f" in cve_hdf5/*) export WITH_H5DUMP=1 ;; esac
+  "$HERE/run_one.sh" "$C/$f" "$d"' _ {} 2>"$OUT/probe.log"
+echo "== comparing"
+"$PY" "$HERE/compare.py" "$OUT" >/dev/null
+t2=$(date +%s)
+cat > "$OUT/meta.json" <<EOF
+{"build_seconds": $((t1 - t0)), "probe_seconds": $((t2 - t1)), "jobs": $JOBS, "timeout_s": $TMO, "mem_kb": $MEM_KB}
+EOF
+export CONFORMANCE_CMD="${CONFORMANCE_CMD:-conformance/run.sh${*:+ $*}}"
+if [ "$REPORT" = 1 ]; then
+  "$PY" "$HERE/report.py" "$OUT" "$REPORT_PATH" "$C"
+  echo "== wrote $REPORT_PATH"
+fi
+if [ "$UPDATE" = 1 ]; then
+  "$PY" "$HERE/check.py" "$OUT" "$HERE/baseline.json" --update
+fi
+"$PY" "$HERE/check.py" "$OUT" "$HERE/baseline.json"

conformance/run_one.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+# run_one.sh <file> <outdir>
+#
+# Probe one file with clawhdf5 (PROBE) and with h5py (PY ref.py), and with
+# h5dump too when WITH_H5DUMP is set. Each side runs under a timeout (TMO
+# seconds, SIGKILL) and an address-space limit (MEM_KB), with core dumps off.
+# Writes <outdir>/<side>.{json,err,rc}; rc 137 = killed by the timeout.
+set -u
+f="$1"; out="$2"; mkdir -p "$out"
+HERE="$(cd "$(dirname "$0")" && pwd)"
+: "${PROBE:?PROBE must name the conformance-probe binary}"
+: "${PY:?PY must name a python with h5py}"
+TMO="${TMO:-20}"
+MEM_KB="${MEM_KB:-4194304}"
+run() { # name cmd...
+  local name=$1; shift
+  ( ulimit -v "$MEM_KB"; ulimit -c 0; RUST_BACKTRACE=1 exec timeout -s KILL "$TMO" "$@" ) \
+      >"$out/$name.json" 2>"$out/$name.err"
+  echo $? >"$out/$name.rc"
+}
+run ours "$PROBE" "$f"
+run ref "$PY" "$HERE/ref.py" "$f"
+if [ -n "${WITH_H5DUMP:-}" ]; then
+  run h5dump h5dump "$f"
+  : >"$out/h5dump.json" # h5dump's text dump is not compared, only its exit status
+fi
+exit 0

crates/clawhdf5-agent/Cargo.toml

@@ -19,6 +19,10 @@ clawhdf5-ann = { path = "../clawhdf5-ann", version = "2.7.0", optional = true }
 clawhdf5-gpu = { path = "../clawhdf5-gpu", version = "2.7.0", optional = true, default-features = false }
 serde = { workspace = true }
 byteorder = "1"
+# Signed checkpoints (MemoryConfig-independent; see `signing`). Pure Rust.
+ed25519-dalek = { version = "2", features = ["rand_core"] }
+sha2 = "0.10"
+rand_core = { version = "0.6", features = ["getrandom"] }
 half = { workspace = true, optional = true }
 rayon = { version = "1", optional = true }
 matrixmultiply = { version = "0.3", optional = true }

crates/clawhdf5-agent/src/consolidation.rs

@@ -203,7 +203,7 @@ impl ImportanceScorer {
     /// Novelty score: 1.0 − max cosine similarity against all existing records.
     /// Returns 1.0 when there are no existing memories.
     ///
-    /// Same result as [`Self::cosine_similarity`] against each record, but the
+    /// Same result as the reference cosine similarity against each record, but the
     /// new embedding's norm is computed once rather than per record, each
     /// record costs one fused pass (dot product and its norm together) rather
     /// than three, and a large working set is scored in parallel. Every insert

crates/clawhdf5-agent/src/lib.rs

@@ -1,4 +1,4 @@
-//! ZeroClaw agent memory HDF5 backend.
+//! Agent memory stored in a single HDF5 file.
 //!
 //! Provides persistent memory storage for AI agents using HDF5 files.
 //! All data is cached in-memory for fast access and flushed to disk
@@ -36,6 +36,7 @@ pub mod reranker;
 pub mod schema;
 pub mod search;
 pub mod session;
+pub mod signing;
 pub mod storage;
 mod store_lock;
 pub mod temporal;
@@ -78,6 +79,7 @@ pub use session::{SessionCache, SessionEntry};
 // --- Error type ---
 
 #[derive(Debug)]
+#[non_exhaustive]
 pub enum MemoryError {
     Io(std::io::Error),
     Hdf5(String),
@@ -88,6 +90,11 @@ pub enum MemoryError {
     /// A record the store cannot hold as given, e.g. an embedding value
     /// outside the half-precision range of a `float16` store.
     InvalidEntry(String),
+    /// The store's checkpoints are signed and no signing key is set, so a
+    /// checkpoint would leave it unsigned. Set the key with
+    /// [`HDF5Memory::set_signing_key`], or drop the signature on purpose with
+    /// [`HDF5Memory::remove_signature`].
+    SigningKeyRequired(String),
 }
 
 impl std::fmt::Display for MemoryError {
@@ -99,6 +106,7 @@ impl std::fmt::Display for MemoryError {
             MemoryError::NotFound(e) => write!(f, "not found: {e}"),
             MemoryError::Locked(e) => write!(f, "store is locked: {e}"),
             MemoryError::InvalidEntry(e) => write!(f, "invalid entry: {e}"),
+            MemoryError::SigningKeyRequired(e) => write!(f, "signing key required: {e}"),
         }
     }
 }
@@ -317,6 +325,12 @@ pub struct HDF5Memory {
     activations_dirty: bool,
     /// Opened with [`HDF5Memory::open_read_only`]: nothing may reach the disk.
     read_only: bool,
+    /// Key that signs every checkpoint; never persisted. See
+    /// [`HDF5Memory::set_signing_key`].
+    signing_key: Option<signing::SigningKey>,
+    /// Checkpoints of this store are signed: the file on disk is, or a key
+    /// has been set. A checkpoint without a key is then refused.
+    signed: bool,
     /// A WAL that `open()` could not read and moved aside; see
     /// [`HDF5Memory::quarantined_wal`].
     quarantined_wal: Option<PathBuf>,
@@ -372,6 +386,8 @@ impl HDF5Memory {
             bm25_filter: bm25::TokenFilter::default(),
             activations_dirty: false,
             read_only: false,
+            signing_key: None,
+            signed: false,
             quarantined_wal: None,
             _lock: Some(lock),
         })
@@ -550,6 +566,8 @@ impl HDF5Memory {
             bm25_filter: bm25::TokenFilter::default(),
             activations_dirty: false,
             read_only,
+            signing_key: None,
+            signed: checkpoint.signed,
             quarantined_wal,
             _lock: lock,
         })
@@ -710,6 +728,39 @@ impl HDF5Memory {
         }
     }
 
+    /// Sign every checkpoint from now on with `key` (Ed25519). The key is
+    /// never written anywhere; set it again after every `open`. Once a store
+    /// is signed, a checkpoint without the key is refused
+    /// ([`MemoryError::SigningKeyRequired`]) rather than silently leaving it
+    /// unsigned. Setting a different key re-signs the store under that key
+    /// from the next checkpoint; a verifier trusting the old key will then
+    /// reject it, which is the point. Call [`AgentMemory::flush_wal`] to sign
+    /// right away.
+    pub fn set_signing_key(&mut self, key: signing::SigningKey) {
+        self.signing_key = Some(key);
+        self.signed = true;
+    }
+
+    /// Stop signing: the next checkpoint writes the store unsigned. The
+    /// deliberate way out of [`MemoryError::SigningKeyRequired`].
+    pub fn remove_signature(&mut self) {
+        self.signing_key = None;
+        self.signed = false;
+    }
+
+    /// Checkpoints of this store are signed (on disk, or from the next
+    /// checkpoint because a key has been set).
+    pub fn is_signed(&self) -> bool {
+        self.signed
+    }
+
+    /// Check the checkpoint at `path` against the public key the caller
+    /// trusts; see [`signing::verify_store`]. Reads the file only: it works
+    /// on a store another process has open.
+    pub fn verify(path: &Path, trusted: &signing::VerifyingKey) -> Result<signing::VerifyReport> {
+        signing::verify_store(path, trusted)
+    }
+
     /// Flush current state to disk and truncate the WAL.
     ///
     /// Every code path that persists the full cache to the .h5 file must
@@ -725,10 +776,28 @@ impl HDF5Memory {
         // Record which WAL prefix this checkpoint contains, so a crash before
         // the truncate below can't replay those entries a second time.
         let wal_applied = self.wal.as_ref().map(|w| w.mark());
+        let signature = match &self.signing_key {
+            Some(key) => Some(signing::sign(
+                key,
+                &self.config,
+                &self.cache,
+                &self.sessions,
+                &self.knowledge,
+                wal_applied,
+            )),
+            None if self.signed => {
+                return Err(MemoryError::SigningKeyRequired(format!(
+                    "{} is signed; set its signing key before a checkpoint \
+                     (saves so far are held in the WAL or in memory)",
+                    self.config.path.display()
+                )));
+            }
+            None => None,
+        };
         // Written before the .h5 so a crash in between leaves a sidecar whose
         // generation matches no checkpoint (ignored), never the reverse.
         let ann_generation = self.persist_vector_index();
-        storage::write_to_disk_with_meta(
+        storage::write_to_disk_signed(
             &self.config.path,
             &self.config,
             &self.cache,
@@ -737,7 +806,9 @@ impl HDF5Memory {
             &schema::CheckpointMeta {
                 wal_applied,
                 ann_generation,
+                signed: signature.is_some(),
             },
+            signature.as_ref(),
         )?;
         if let Some(ref mut w) = self.wal {
             w.truncate()?;

crates/clawhdf5-agent/src/openclaw.rs

@@ -1,10 +1,12 @@
-//! OpenClaw Integration Layer.
+//! A Markdown-oriented memory backend over [`crate::HDF5Memory`].
 //!
-//! Bridge between OpenClaw agent gateway (Markdown + sqlite-vec) and the
-//! clawhdf5 HDF5-backed memory backend.  Provides:
+//! Named for OpenClaw, whose workspace memory is Markdown, but **not an
+//! OpenClaw plugin**: nothing here registers with OpenClaw, and the
+//! integration it was written for never worked (see `docs/openclaw.md`).
+//! Provides:
 //!
-//! - [`MemoryBackend`] — the trait OpenClaw implements against.
-//! - [`ClawhdfBackend`] — concrete HDF5-backed implementation.
+//! - [`MemoryBackend`] — search / read back / write / ingest / export.
+//! - [`ClawhdfBackend`] — the HDF5-backed implementation.
 //! - [`MarkdownParser`] — splits Markdown into [`MarkdownSection`] records.
 //! - [`MarkdownExporter`] — renders sections back to Markdown text.
 
@@ -61,7 +63,8 @@ pub struct BackendStats {
 // MemoryBackend trait
 // ─────────────────────────────────────────────────────────────────────────────
 
-/// Interface that OpenClaw uses to interact with a memory backend.
+/// A Markdown-oriented memory backend: search, read back by path, write,
+/// ingest and export.
 ///
 /// Implementors provide persistent storage, full-text + vector search,
 /// Markdown ingestion / export, and statistics.
@@ -318,7 +321,7 @@ impl MarkdownExporter {
 ///
 /// # Path mapping
 ///
-/// OpenClaw addresses memories by file path (e.g. `"memory/user.md"`).
+/// Memories are addressed by file path (e.g. `"memory/user.md"`).
 /// Internally every [`MemoryEntry`] stores the originating path as its
 /// `source_channel`.  Section sub-paths are stored as
 /// `"<path>::<heading>"`.
@@ -421,7 +424,7 @@ impl ClawhdfBackend {
 
     // ── Compaction & Consolidation hooks (7.6) ────────────────────────────
 
-    /// Run a compaction cycle — called by OpenClaw during session compaction.
+    /// Run a compaction cycle (decay, compaction, WAL flush).
     ///
     /// Sequence:
     /// 1. `tick_session()` — apply Hebbian decay to all activation weights.

crates/clawhdf5-agent/src/provenance.rs

@@ -2,7 +2,7 @@
 //!
 //! Records the origin, authorship, and a content hash of every memory chunk
 //! so the system can detect *accidental* corruption and trace data lineage.
-//! The hash is unkeyed (see [`fnv1a_64`]) — this is not a tamper-evidence or
+//! The hash is unkeyed (FNV-1a) — this is not a tamper-evidence or
 //! authenticity guarantee.
 
 use std::collections::HashMap;

crates/clawhdf5-agent/src/schema.rs

@@ -15,6 +15,9 @@ use crate::session::SessionCache;
 use crate::wal::WalMark;
 
 pub const SCHEMA_VERSION: &str = "1.0";
+/// Writer-version tag stored in `/meta` as `edgehdf5_version`. Kept for file
+/// compatibility; despite the name it has nothing to do with ZeroClaw, which
+/// does not use clawhdf5.
 pub const ZEROCLAW_VERSION: &str = "0.8.0";
 
 /// `/meta` attributes holding the [`WalMark`] of the WAL prefix already folded
@@ -23,6 +26,7 @@ pub const ZEROCLAW_VERSION: &str = "0.8.0";
 const WAL_APPLIED_LEN_ATTR: &str = "wal_applied_len";
 const WAL_APPLIED_CRC_ATTR: &str = "wal_applied_crc";
 const ANN_GENERATION_ATTR: &str = "ann_generation";
+const SIG_VERSION_ATTR: &str = "sig_version";
 
 /// Build a complete HDF5 file from the in-memory state.
 pub fn build_hdf5_file(
@@ -46,7 +50,7 @@ pub fn build_hdf5_file_with_mark(
 ) -> Result<Vec<u8>, MemoryError> {
     let meta = CheckpointMeta {
         wal_applied,
-        ann_generation: None,
+        ..CheckpointMeta::default()
     };
     build_hdf5_file_with_meta(config, cache, sessions, knowledge, &meta)
 }
@@ -61,6 +65,10 @@ pub struct CheckpointMeta {
     /// one left over from another checkpoint can never be attached to records
     /// it wasn't built from.
     pub ann_generation: Option<u64>,
+    /// The checkpoint carries an Ed25519 signature (see [`crate::signing`]).
+    /// Read-only: whether a checkpoint is *written* signed is decided by the
+    /// signature passed to [`build_hdf5_file_signed`].
+    pub signed: bool,
 }
 
 /// [`build_hdf5_file`] with checkpoint bookkeeping.
@@ -70,6 +78,19 @@ pub fn build_hdf5_file_with_meta(
     sessions: &SessionCache,
     knowledge: &KnowledgeCache,
     checkpoint: &CheckpointMeta,
+) -> Result<Vec<u8>, MemoryError> {
+    build_hdf5_file_signed(config, cache, sessions, knowledge, checkpoint, None)
+}
+
+/// [`build_hdf5_file_with_meta`], plus a signed manifest of the contents
+/// (see [`crate::signing`]).
+pub fn build_hdf5_file_signed(
+    config: &MemoryConfig,
+    cache: &MemoryCache,
+    sessions: &SessionCache,
+    knowledge: &KnowledgeCache,
+    checkpoint: &CheckpointMeta,
+    signature: Option<&crate::signing::StoredSignature>,
 ) -> Result<Vec<u8>, MemoryError> {
     let wal_applied = checkpoint.wal_applied;
     let mut builder = clawhdf5::FileBuilder::new();
@@ -130,11 +151,42 @@ pub fn build_hdf5_file_with_meta(
         // round trip through every reader.
         meta.set_attr(ANN_GENERATION_ATTR, AttrValue::I64(generation as i64));
     }
+    if let Some(sig) = signature {
+        use crate::signing::to_hex;
+        let m = &sig.manifest;
+        meta.set_attr(
+            SIG_VERSION_ATTR,
+            AttrValue::I64(crate::signing::MANIFEST_VERSION),
+        );
+        meta.set_attr("sig_algorithm", AttrValue::String("ed25519".into()));
+        meta.set_attr("sig_public_key", AttrValue::String(to_hex(&sig.public_key)));
+        meta.set_attr("sig_signature", AttrValue::String(to_hex(&sig.signature)));
+        meta.set_attr("sig_record_count", AttrValue::I64(m.record_count as i64));
+        meta.set_attr(
+            "sig_records_root",
+            AttrValue::String(to_hex(&m.records_root)),
+        );
+        meta.set_attr("sig_settings", AttrValue::String(to_hex(&m.settings)));
+        meta.set_attr("sig_sessions", AttrValue::String(to_hex(&m.sessions)));
+        meta.set_attr("sig_graph", AttrValue::String(to_hex(&m.graph)));
+    }
     // Need at least one dataset in the group for it to be a proper group
     meta.create_dataset("_marker").with_u8_data(&[1]).compact();
     let finished_meta = meta.finish();
     builder.add_group(finished_meta);
 
+    // /integrity: the signed per-record hashes, so verification can say
+    // which records changed.
+    if let Some(sig) = signature {
+        let mut group = builder.create_group("integrity");
+        let flat: Vec<u8> = sig.record_hashes.iter().flatten().copied().collect();
+        group
+            .create_dataset("record_hashes")
+            .with_u8_data(&flat)
+            .with_shape(&[sig.record_hashes.len() as u64, 32]);
+        builder.add_group(group.finish());
+    }
+
     // /memory group
     build_memory_group(&mut builder, config, cache)?;
 
@@ -425,10 +477,34 @@ fn write_string_dataset(
     }
 }
 
+/// `/meta`'s attributes, failing if any of them cannot be read.
+///
+/// `Group::attrs` leaves out an attribute it cannot decode. For the store's
+/// settings that would silently fall back to defaults (e.g. `float16`, the
+/// WAL mark), so an unreadable attribute is an error here, as it was before
+/// `attrs` became tolerant.
+fn meta_attrs(
+    file: &clawhdf5::File,
+) -> Result<std::collections::HashMap<String, AttrValue>, MemoryError> {
+    let meta = file
+        .group("meta")
+        .map_err(|e| MemoryError::Schema(format!("missing /meta group: {e}")))?;
+    let (attrs, errors) = meta
+        .attrs_with_errors()
+        .map_err(|e| MemoryError::Schema(format!("cannot read /meta attrs: {e}")))?;
+    if let Some(e) = errors.first() {
+        return Err(MemoryError::Schema(format!(
+            "cannot read /meta attrs: {} unreadable, first: {e}",
+            errors.len()
+        )));
+    }
+    Ok(attrs)
+}
+
 /// Validate an HDF5 file has the correct schema and load all data.
 /// Read the checkpoint's [`WalMark`] from `/meta`, if it has one.
 pub fn read_wal_mark(file: &clawhdf5::File) -> Option<WalMark> {
-    let attrs = file.group("meta").ok()?.attrs().ok()?;
+    let attrs = meta_attrs(file).ok()?;
     let len = match attrs.get(WAL_APPLIED_LEN_ATTR)? {
         AttrValue::I64(v) => u64::try_from(*v).ok()?,
         _ => return None,
@@ -440,19 +516,75 @@ pub fn read_wal_mark(file: &clawhdf5::File) -> Option<WalMark> {
     Some(WalMark { len, crc })
 }
 
+/// Read a checkpoint's signature, if it has one. A signature whose
+/// attributes are present but malformed is an error, not "unsigned".
+pub fn read_signature(
+    file: &clawhdf5::File,
+) -> Result<Option<crate::signing::StoredSignature>, MemoryError> {
+    use crate::signing::{Manifest, StoredSignature, from_hex};
+    let attrs = meta_attrs(file)?;
+    let version = match attrs.get(SIG_VERSION_ATTR) {
+        None => return Ok(None),
+        Some(AttrValue::I64(v)) => *v,
+        Some(_) => return Err(MemoryError::Schema("malformed sig_version".into())),
+    };
+    if version != crate::signing::MANIFEST_VERSION {
+        return Err(MemoryError::Schema(format!(
+            "unsupported signature version {version}"
+        )));
+    }
+    fn hex<const N: usize>(
+        attrs: &std::collections::HashMap<String, AttrValue>,
+        name: &str,
+    ) -> Result<[u8; N], MemoryError> {
+        match attrs.get(name) {
+            Some(AttrValue::String(s)) => from_hex::<N>(s),
+            _ => None,
+        }
+        .ok_or_else(|| MemoryError::Schema(format!("malformed or missing {name}")))
+    }
+    let record_count = match attrs.get("sig_record_count") {
+        Some(AttrValue::I64(v)) if *v >= 0 => *v as u64,
+        _ => return Err(MemoryError::Schema("malformed sig_record_count".into())),
+    };
+    let group = file
+        .group("integrity")
+        .map_err(|e| MemoryError::Schema(format!("signed checkpoint without /integrity: {e}")))?;
+    let flat = read_u8_dataset(&group, "record_hashes")?;
+    if flat.len() % 32 != 0 {
+        return Err(MemoryError::Schema(
+            "/integrity/record_hashes is not a whole number of hashes".into(),
+        ));
+    }
+    let record_hashes = flat.as_chunks::<32>().0.to_vec();
+    Ok(Some(StoredSignature {
+        manifest: Manifest {
+            record_count,
+            records_root: hex::<32>(&attrs, "sig_records_root")?,
+            settings: hex::<32>(&attrs, "sig_settings")?,
+            sessions: hex::<32>(&attrs, "sig_sessions")?,
+            graph: hex::<32>(&attrs, "sig_graph")?,
+        },
+        record_hashes,
+        public_key: hex::<32>(&attrs, "sig_public_key")?,
+        signature: hex::<64>(&attrs, "sig_signature")?,
+    }))
+}
+
 /// Read the checkpoint bookkeeping from `/meta`.
 pub fn read_checkpoint_meta(file: &clawhdf5::File) -> CheckpointMeta {
-    let ann_generation = file
-        .group("meta")
+    let ann_generation =
+        meta_attrs(file)
             .ok()
-        .and_then(|g| g.attrs().ok())
             .and_then(|attrs| match attrs.get(ANN_GENERATION_ATTR) {
                 Some(AttrValue::I64(v)) => Some(*v as u64),
                 _ => None,
             });
+    let signed = meta_attrs(file).is_ok_and(|attrs| attrs.contains_key(SIG_VERSION_ATTR));
     CheckpointMeta {
         wal_applied: read_wal_mark(file),
         ann_generation,
+        signed,
     }
 }
 
@@ -460,12 +592,7 @@ pub fn validate_and_load(
     file: &clawhdf5::File,
 ) -> Result<(MemoryConfig, MemoryCache, SessionCache, KnowledgeCache), MemoryError> {
     // Read /meta group attributes
-    let meta = file
-        .group("meta")
-        .map_err(|e| MemoryError::Schema(format!("missing /meta group: {e}")))?;
-    let attrs = meta
-        .attrs()
-        .map_err(|e| MemoryError::Schema(format!("cannot read /meta attrs: {e}")))?;
+    let attrs = meta_attrs(file)?;
 
     let schema_version = match attrs.get("schema_version") {
         Some(AttrValue::String(s)) => s.clone(),

crates/clawhdf5-agent/src/signing.rs

@@ -0,0 +1,419 @@
+//! Ed25519-signed checkpoints.
+//!
+//! When a signing key is set ([`crate::HDF5Memory::set_signing_key`]), every
+//! checkpoint writes a signed manifest of the store: a SHA-256 per memory
+//! record rolled into a Merkle root, plus hashes of the store's settings, its
+//! sessions and its knowledge graph. [`verify_store`] recomputes all of it from
+//! the file and checks the signature against a public key the caller trusts,
+//! so any change to the checkpointed file — a record's text or embedding, a
+//! setting, a session, a graph edge, made through this crate or any other HDF5
+//! tool — is detected, and the per-record hashes say which records changed.
+//!
+//! What it does not cover: saves still only in the WAL (made since the last
+//! checkpoint). [`VerifyReport::wal_entries_unsigned`] counts them.
+//!
+//! The hashes cover exactly what the file persists, in the form the loader
+//! returns it, so a store verifies after any number of reopen/checkpoint
+//! cycles. Derived data (L2 norms, the vector index) is not covered; it is
+//! recomputed from covered data.
+
+use ed25519_dalek::{Signature, Signer, Verifier};
+pub use ed25519_dalek::{SigningKey, VerifyingKey};
+use sha2::{Digest, Sha256};
+
+use crate::MemoryConfig;
+use crate::cache::MemoryCache;
+use crate::knowledge::KnowledgeCache;
+use crate::session::SessionCache;
+use crate::wal::WalMark;
+
+/// Version of the manifest encoding; part of what is signed.
+pub const MANIFEST_VERSION: i64 = 1;
+
+type Hash = [u8; 32];
+
+/// The hashes a signature covers.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct Manifest {
+    pub record_count: u64,
+    /// Merkle root over the per-record hashes.
+    pub records_root: Hash,
+    /// Settings persisted in `/meta`, plus the checkpoint's WAL mark.
+    pub settings: Hash,
+    pub sessions: Hash,
+    pub graph: Hash,
+}
+
+impl Manifest {
+    /// The exact bytes that are signed.
+    pub fn signed_bytes(&self) -> Vec<u8> {
+        let mut m = Vec::with_capacity(160);
+        m.extend_from_slice(b"clawhdf5-agent signed checkpoint\0");
+        m.extend_from_slice(&MANIFEST_VERSION.to_le_bytes());
+        m.extend_from_slice(&self.record_count.to_le_bytes());
+        m.extend_from_slice(&self.records_root);
+        m.extend_from_slice(&self.settings);
+        m.extend_from_slice(&self.sessions);
+        m.extend_from_slice(&self.graph);
+        m
+    }
+}
+
+/// A signature as stored in a checkpoint.
+#[derive(Debug, Clone)]
+pub struct StoredSignature {
+    pub manifest: Manifest,
+    pub record_hashes: Vec<Hash>,
+    pub public_key: [u8; 32],
+    pub signature: [u8; 64],
+}
+
+/// Build the manifest (and per-record hashes) for the state about to be
+/// checkpointed, and sign it.
+pub fn sign(
+    key: &SigningKey,
+    config: &MemoryConfig,
+    cache: &MemoryCache,
+    sessions: &SessionCache,
+    knowledge: &KnowledgeCache,
+    wal_applied: Option<WalMark>,
+) -> StoredSignature {
+    let (manifest, record_hashes) = manifest(config, cache, sessions, knowledge, wal_applied);
+    let signature = key.sign(&manifest.signed_bytes()).to_bytes();
+    StoredSignature {
+        manifest,
+        record_hashes,
+        public_key: key.verifying_key().to_bytes(),
+        signature,
+    }
+}
+
+/// Compute the manifest of a store's state.
+pub fn manifest(
+    config: &MemoryConfig,
+    cache: &MemoryCache,
+    sessions: &SessionCache,
+    knowledge: &KnowledgeCache,
+    wal_applied: Option<WalMark>,
+) -> (Manifest, Vec<Hash>) {
+    let record_hashes: Vec<Hash> = (0..cache.len()).map(|i| record_hash(cache, i)).collect();
+    let manifest = Manifest {
+        record_count: cache.len() as u64,
+        records_root: merkle_root(&record_hashes),
+        settings: settings_hash(config, wal_applied),
+        sessions: sessions_hash(sessions),
+        graph: graph_hash(knowledge),
+    };
+    (manifest, record_hashes)
+}
+
+// ---------------------------------------------------------------------------
+// Canonical encoding
+// ---------------------------------------------------------------------------
+
+/// A SHA-256 over length-prefixed fields, so no two different field lists
+/// hash the same bytes.
+struct Fields(Sha256);
+
+impl Fields {
+    fn new(domain: &str) -> Self {
+        let mut h = Sha256::new();
+        h.update((domain.len() as u64).to_le_bytes());
+        h.update(domain.as_bytes());
+        Self(h)
+    }
+    fn bytes(&mut self, b: &[u8]) -> &mut Self {
+        self.0.update((b.len() as u64).to_le_bytes());
+        self.0.update(b);
+        self
+    }
+    /// Strings as the loader returns them: stored null-padded, so a trailing
+    /// NUL cannot survive a round trip and must not be part of the hash.
+    fn str(&mut self, s: &str) -> &mut Self {
+        self.bytes(s.trim_end_matches('\0').as_bytes())
+    }
+    fn u64(&mut self, v: u64) -> &mut Self {
+        self.0.update(v.to_le_bytes());
+        self
+    }
+    fn f64(&mut self, v: f64) -> &mut Self {
+        self.0.update(v.to_bits().to_le_bytes());
+        self
+    }
+    fn f32(&mut self, v: f32) -> &mut Self {
+        self.0.update(v.to_bits().to_le_bytes());
+        self
+    }
+    fn finish(self) -> Hash {
+        self.0.finalize().into()
+    }
+}
+
+/// Everything persisted about record `i`, including its position. The
+/// embedding is hashed as the cache holds it — for a `float16` store that is
+/// the half-rounded value the file holds.
+fn record_hash(cache: &MemoryCache, i: usize) -> Hash {
+    let mut f = Fields::new("clawhdf5-agent/record");
+    f.u64(i as u64).str(&cache.chunks[i]);
+    let emb: Vec<u8> = cache.embeddings[i]
+        .iter()
+        .flat_map(|v| v.to_bits().to_le_bytes())
+        .collect();
+    f.bytes(&emb)
+        .str(&cache.source_channels[i])
+        .f64(cache.timestamps[i])
+        .str(&cache.session_ids[i])
+        .str(&cache.tags[i])
+        .u64(u64::from(cache.tombstones[i]))
+        .f32(cache.activation_weights[i]);
+    f.finish()
+}
+
+/// Binary Merkle tree: leaves are the record hashes; a parent hashes its two
+/// children with a node prefix; an odd node is carried up unchanged.
+fn merkle_root(leaves: &[Hash]) -> Hash {
+    if leaves.is_empty() {
+        return Fields::new("clawhdf5-agent/merkle-empty").finish();
+    }
+    let mut level: Vec<Hash> = leaves.to_vec();
+    while level.len() > 1 {
+        level = level
+            .chunks(2)
+            .map(|pair| match pair {
+                [l, r] => {
+                    let mut h = Sha256::new();
+                    h.update([1u8]);
+                    h.update(l);
+                    h.update(r);
+                    h.finalize().into()
+                }
+                [only] => *only,
+                _ => unreachable!(),
+            })
+            .collect();
+    }
+    level[0]
+}
+
+fn settings_hash(c: &MemoryConfig, wal_applied: Option<WalMark>) -> Hash {
+    let mut f = Fields::new("clawhdf5-agent/settings");
+    f.str(crate::schema::SCHEMA_VERSION)
+        .str(&c.created_at)
+        .str(&c.agent_id)
+        .str(&c.embedder)
+        .u64(c.embedding_dim as u64)
+        .u64(c.chunk_size as u64)
+        .u64(c.overlap as u64)
+        .u64(u64::from(c.float16))
+        .u64(u64::from(c.compression))
+        .u64(u64::from(c.compression_level))
+        .f32(c.compact_threshold)
+        .f32(c.hebbian_boost)
+        .f32(c.decay_factor)
+        .u64(u64::from(c.wal_enabled))
+        .u64(c.wal_max_entries as u64)
+        .u64(u64::from(c.quantized_index))
+        .u64(c.hnsw_m as u64)
+        .u64(c.hnsw_ef_construction as u64)
+        .u64(c.hnsw_ef_search as u64);
+    // An empty mark is not written to the file, so it must hash as none.
+    match wal_applied.filter(|m| m.len > 0) {
+        Some(m) => f.u64(1).u64(m.len).u64(u64::from(m.crc)),
+        None => f.u64(0),
+    };
+    f.finish()
+}
+
+fn sessions_hash(s: &SessionCache) -> Hash {
+    let mut f = Fields::new("clawhdf5-agent/sessions");
+    f.u64(s.entries.len() as u64);
+    for (i, e) in s.entries.iter().enumerate() {
+        f.str(&e.id)
+            .u64(e.start_idx)
+            .u64(e.end_idx)
+            .str(&e.channel)
+            .f64(e.ts)
+            .str(s.summaries.get(i).map(String::as_str).unwrap_or(""));
+    }
+    f.finish()
+}
+
+fn graph_hash(k: &KnowledgeCache) -> Hash {
+    let mut f = Fields::new("clawhdf5-agent/graph");
+    f.u64(k.entities.len() as u64);
+    for e in &k.entities {
+        f.u64(e.id)
+            .str(&e.name)
+            .str(&e.entity_type)
+            .u64(e.embedding_idx as u64);
+    }
+    f.u64(k.relations.len() as u64);
+    for r in &k.relations {
+        f.u64(r.src)
+            .u64(r.tgt)
+            .str(&r.relation)
+            .f32(r.weight)
+            .f64(r.ts);
+    }
+    f.u64(k.alias_strings.len() as u64);
+    for (s, id) in k.alias_strings.iter().zip(&k.alias_entity_ids) {
+        f.str(s).u64(*id as u64);
+    }
+    f.finish()
+}
+
+// ---------------------------------------------------------------------------
+// Verification
+// ---------------------------------------------------------------------------
+
+/// The outcome of [`verify_store`].
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct VerifyReport {
+    /// The checkpoint carries a signature.
+    pub signed: bool,
+    /// The signature was made by the key the caller trusts.
+    pub key_matches: bool,
+    /// The signature over the stored manifest is valid.
+    pub signature_valid: bool,
+    /// The file's current contents match the signed manifest.
+    pub records_match: bool,
+    pub settings_match: bool,
+    pub sessions_match: bool,
+    pub graph_match: bool,
+    /// Records whose contents differ from what was signed (by position),
+    /// when the stored per-record hashes are themselves authentic.
+    pub changed_records: Vec<usize>,
+    /// Records in the file versus in the signed manifest.
+    pub record_count: u64,
+    pub signed_record_count: u64,
+    /// The public key the checkpoint claims to be signed by.
+    pub public_key: Option<[u8; 32]>,
+    /// Saves in the WAL after the checkpoint: not covered by the signature.
+    pub wal_entries_unsigned: usize,
+}
+
+impl VerifyReport {
+    /// Signed by the trusted key, signature valid, and every part of the
+    /// file unchanged since it was signed.
+    pub fn is_valid(&self) -> bool {
+        self.signed
+            && self.key_matches
+            && self.signature_valid
+            && self.records_match
+            && self.settings_match
+            && self.sessions_match
+            && self.graph_match
+    }
+}
+
+/// Check a store file against the public key the caller trusts.
+///
+/// Reads the checkpoint (not the WAL), recomputes every hash from its
+/// contents and checks the signature. Never writes.
+pub fn verify_store(
+    path: &std::path::Path,
+    trusted: &VerifyingKey,
+) -> Result<VerifyReport, crate::MemoryError> {
+    let file = clawhdf5::File::open(path)
+        .map_err(|e| crate::MemoryError::Hdf5(format!("cannot open {}: {e}", path.display())))?;
+    let (config, cache, sessions, knowledge) = crate::schema::validate_and_load(&file)?;
+    let checkpoint = crate::schema::read_checkpoint_meta(&file);
+    let stored = crate::schema::read_signature(&file)?;
+    let wal_entries_unsigned = count_wal_entries_after(path, checkpoint.wal_applied);
+
+    let (current, current_hashes) = manifest(
+        &config,
+        &cache,
+        &sessions,
+        &knowledge,
+        checkpoint.wal_applied,
+    );
+
+    let Some(stored) = stored else {
+        return Ok(VerifyReport {
+            signed: false,
+            key_matches: false,
+            signature_valid: false,
+            records_match: false,
+            settings_match: false,
+            sessions_match: false,
+            graph_match: false,
+            changed_records: Vec::new(),
+            record_count: current.record_count,
+            signed_record_count: 0,
+            public_key: None,
+            wal_entries_unsigned,
+        });
+    };
+
+    let key_matches = stored.public_key == trusted.to_bytes();
+    let signature_valid = trusted
+        .verify(
+            &stored.manifest.signed_bytes(),
+            &Signature::from_bytes(&stored.signature),
+        )
+        .is_ok();
+    // The stored per-record hashes can localise a change only if they are
+    // the ones that were signed.
+    let hashes_authentic = signature_valid
+        && stored.record_hashes.len() as u64 == stored.manifest.record_count
+        && merkle_root(&stored.record_hashes) == stored.manifest.records_root;
+    let changed_records = if hashes_authentic {
+        let n = current_hashes.len().max(stored.record_hashes.len());
+        (0..n)
+            .filter(|&i| current_hashes.get(i) != stored.record_hashes.get(i))
+            .collect()
+    } else {
+        Vec::new()
+    };
+
+    Ok(VerifyReport {
+        signed: true,
+        key_matches,
+        signature_valid,
+        records_match: signature_valid
+            && current.record_count == stored.manifest.record_count
+            && current.records_root == stored.manifest.records_root,
+        settings_match: signature_valid && current.settings == stored.manifest.settings,
+        sessions_match: signature_valid && current.sessions == stored.manifest.sessions,
+        graph_match: signature_valid && current.graph == stored.manifest.graph,
+        changed_records,
+        record_count: current.record_count,
+        signed_record_count: stored.manifest.record_count,
+        public_key: Some(stored.public_key),
+        wal_entries_unsigned,
+    })
+}
+
+fn count_wal_entries_after(store: &std::path::Path, mark: Option<WalMark>) -> usize {
+    let wal = store.with_extension("h5.wal");
+    if !wal.exists() {
+        return 0;
+    }
+    crate::wal::WalFile::read_entries_for_migration(&wal, mark)
+        .map(|e| e.len())
+        .unwrap_or(0)
+}
+
+/// A new random signing key from the operating system's RNG.
+pub fn generate_key() -> SigningKey {
+    SigningKey::generate(&mut rand_core::OsRng)
+}
+
+/// Hex encoding for keys and signatures in attributes and the CLI.
+pub fn to_hex(bytes: &[u8]) -> String {
+    bytes.iter().map(|b| format!("{b:02x}")).collect()
+}
+
+/// Parse hex into exactly `N` bytes.
+pub fn from_hex<const N: usize>(s: &str) -> Option<[u8; N]> {
+    let s = s.trim();
+    if s.len() != 2 * N {
+        return None;
+    }
+    let mut out = [0u8; N];
+    for (i, byte) in out.iter_mut().enumerate() {
+        *byte = u8::from_str_radix(&s[2 * i..2 * i + 2], 16).ok()?;
+    }
+    Some(out)
+}

crates/clawhdf5-agent/src/storage.rs

@@ -36,7 +36,7 @@ pub fn write_to_disk_with_mark(
 ) -> Result<(), MemoryError> {
     let meta = schema::CheckpointMeta {
         wal_applied,
-        ann_generation: None,
+        ..schema::CheckpointMeta::default()
     };
     write_to_disk_with_meta(path, config, cache, sessions, knowledge, &meta)
 }
@@ -50,7 +50,21 @@ pub fn write_to_disk_with_meta(
     knowledge: &KnowledgeCache,
     checkpoint: &schema::CheckpointMeta,
 ) -> Result<(), MemoryError> {
-    let bytes = schema::build_hdf5_file_with_meta(config, cache, sessions, knowledge, checkpoint)?;
+    write_to_disk_signed(path, config, cache, sessions, knowledge, checkpoint, None)
+}
+
+/// [`write_to_disk_with_meta`] with a signed manifest of the contents.
+pub fn write_to_disk_signed(
+    path: &Path,
+    config: &MemoryConfig,
+    cache: &MemoryCache,
+    sessions: &SessionCache,
+    knowledge: &KnowledgeCache,
+    checkpoint: &schema::CheckpointMeta,
+    signature: Option<&crate::signing::StoredSignature>,
+) -> Result<(), MemoryError> {
+    let bytes =
+        schema::build_hdf5_file_signed(config, cache, sessions, knowledge, checkpoint, signature)?;
 
     if bytes.is_empty() {
         return Err(MemoryError::Hdf5("build_hdf5_file produced 0 bytes".into()));

crates/clawhdf5-agent/tests/float16_store.rs

@@ -258,3 +258,43 @@ fn an_existing_f32_store_stays_f32() {
     assert_eq!(&values[..before.1.len()], before.1.as_slice());
     assert_eq!(&values[before.1.len()..], odd.as_slice());
 }
+
+/// `Group::attrs` leaves out an attribute it cannot decode. A store whose
+/// `float16` setting is unreadable must not open as `float16 = false` (or with
+/// any other default in place of a setting it has): it is an error.
+#[test]
+fn unreadable_meta_attribute_fails_open_instead_of_defaulting() {
+    let dir = TempDir::new().unwrap();
+    let path = dir.path().join("store.h5");
+    {
+        let mut m = HDF5Memory::create(config(&dir, "store.h5", true)).unwrap();
+        m.save(entry(1)).unwrap();
+        m.flush_wal().unwrap();
+    }
+    assert!(HDF5Memory::open_read_only(&path).is_ok());
+
+    // Give the `float16` attribute message an unknown version (the name is
+    // at +8 in a version-1 message and +9 in a version-3 one).
+    let mut bytes = std::fs::read(&path).unwrap();
+    let name = b"float16\0";
+    let mut hit = false;
+    let positions: Vec<usize> = (9..bytes.len() - name.len())
+        .filter(|&p| &bytes[p..p + name.len()] == name)
+        .collect();
+    for pos in positions {
+        for (back, version) in [(8, 1u8), (9, 3u8)] {
+            if bytes[pos - back] == version {
+                bytes[pos - back] = 0x7f;
+                hit = true;
+            }
+        }
+    }
+    assert!(hit, "float16 attribute message not found");
+    std::fs::write(&path, &bytes).unwrap();
+
+    match HDF5Memory::open_read_only(&path) {
+        Err(MemoryError::Schema(msg)) => assert!(msg.contains("/meta"), "{msg}"),
+        Err(e) => panic!("unexpected error: {e}"),
+        Ok(_) => panic!("store opened with an unreadable float16 setting"),
+    }
+}

crates/clawhdf5-agent/tests/h5py_interop.rs

@@ -92,3 +92,64 @@ print(len(names))
         assert!(n >= 10, "only {n} datasets");
     }
 }
+
+#[test]
+fn an_edit_made_with_h5py_breaks_the_signature_and_names_the_record() {
+    if !h5py_available() {
+        assert!(
+            std::env::var("CLAWHDF5_REQUIRE_INTEROP").as_deref() != Ok("1"),
+            "CLAWHDF5_REQUIRE_INTEROP=1 but python3 with h5py is not available"
+        );
+        eprintln!("SKIP: python3 with h5py not available");
+        return;
+    }
+    use clawhdf5_agent::signing::SigningKey;
+    let dir = tempfile::tempdir().unwrap();
+    let path = dir.path().join("signed.h5");
+    let key = SigningKey::from_bytes(&[42; 32]);
+    let mut m = HDF5Memory::create(MemoryConfig::new(path.clone(), "agent", 8)).unwrap();
+    m.set_signing_key(key.clone());
+    m.save_batch(
+        (0..10)
+            .map(|i| MemoryEntry {
+                chunk: format!("memory {i}"),
+                embedding: (0..8).map(|j| ((i * 8 + j) as f32).cos()).collect(),
+                source_channel: "test".into(),
+                timestamp: i as f64,
+                session_id: "s".into(),
+                tags: String::new(),
+            })
+            .collect(),
+    )
+    .unwrap();
+    drop(m);
+    assert!(
+        HDF5Memory::verify(&path, &key.verifying_key())
+            .unwrap()
+            .is_valid()
+    );
+
+    // Someone edits one timestamp in place with h5py.
+    let script = format!(
+        r#"
+import h5py
+with h5py.File("{}", "r+") as f:
+    ts = f["memory/timestamps"]
+    ts[3] = 12345.0
+"#,
+        path.display()
+    );
+    let out = Command::new(python())
+        .args(["-c", &script])
+        .output()
+        .unwrap();
+    assert!(
+        out.status.success(),
+        "{}",
+        String::from_utf8_lossy(&out.stderr)
+    );
+
+    let r = HDF5Memory::verify(&path, &key.verifying_key()).unwrap();
+    assert!(r.signature_valid && !r.is_valid(), "{r:?}");
+    assert_eq!(r.changed_records, vec![3]);
+}

crates/clawhdf5-agent/tests/signed_store.rs

@@ -0,0 +1,330 @@
+//! Ed25519-signed checkpoints: `HDF5Memory::set_signing_key` and
+//! `HDF5Memory::verify`.
+
+use std::path::Path;
+
+use clawhdf5_agent::signing::{SigningKey, VerifyReport, VerifyingKey};
+use clawhdf5_agent::storage;
+use clawhdf5_agent::{AgentMemory, HDF5Memory, MemoryConfig, MemoryEntry, MemoryError, schema};
+use tempfile::TempDir;
+
+const DIM: usize = 16;
+
+fn key(seed: u8) -> SigningKey {
+    SigningKey::from_bytes(&[seed; 32])
+}
+
+fn entry(i: usize, chunk: &str) -> MemoryEntry {
+    MemoryEntry {
+        chunk: chunk.to_string(),
+        embedding: (0..DIM)
+            .map(|j| ((i * DIM + j) as f32 * 0.37).sin())
+            .collect(),
+        source_channel: "chat".into(),
+        timestamp: 1_700_000_000.0 + i as f64,
+        session_id: format!("s{}", i % 3),
+        tags: format!("t{i}"),
+    }
+}
+
+/// Awkward strings on purpose: they must hash the same after a round trip.
+const TEXTS: [&str; 6] = [
+    "plain text",
+    "ünïcödé — 日本語 🙂",
+    "",
+    "trailing spaces   ",
+    "tab\tand\nnewline",
+    "x",
+];
+
+fn signed_store(dir: &TempDir, float16: bool, k: &SigningKey) -> std::path::PathBuf {
+    let mut cfg = MemoryConfig::new(dir.path().join("s.h5"), "agent", DIM);
+    cfg.float16 = float16;
+    let path = cfg.path.clone();
+    let mut m = HDF5Memory::create(cfg).unwrap();
+    m.set_signing_key(k.clone());
+    let entries = (0..30).map(|i| entry(i, TEXTS[i % TEXTS.len()])).collect();
+    m.save_batch(entries).unwrap();
+    // Some graph and a deleted record, so every part of the manifest is used.
+    let a = m.knowledge_mut().add_entity("Alice", "person", 0);
+    let b = m.knowledge_mut().add_entity("Acme", "org", -1);
+    m.knowledge_mut().add_relation(a, b, "works_at", 0.75);
+    m.sessions_mut()
+        .add_at("s0", 0, 9, "chat", "first session", 1_700_000_000.0);
+    m.delete(4).unwrap();
+    m.flush_wal().unwrap();
+    path
+}
+
+fn verify(path: &Path, k: &SigningKey) -> VerifyReport {
+    HDF5Memory::verify(path, &k.verifying_key()).unwrap()
+}
+
+#[test]
+fn a_signed_store_verifies_through_reopen_and_checkpoint_cycles() {
+    for float16 in [true, false] {
+        let dir = TempDir::new().unwrap();
+        let k = key(7);
+        let path = signed_store(&dir, float16, &k);
+        let r = verify(&path, &k);
+        assert!(r.is_valid(), "float16={float16}: {r:?}");
+        assert_eq!(r.public_key, Some(k.verifying_key().to_bytes()));
+        assert_eq!(r.record_count, 30);
+        assert!(r.changed_records.is_empty());
+
+        // Reopen, change nothing, checkpoint again (with the key): still valid.
+        for _ in 0..3 {
+            let mut m = HDF5Memory::open(&path).unwrap();
+            assert!(m.is_signed());
+            m.set_signing_key(k.clone());
+            m.flush_wal().unwrap();
+            drop(m);
+            assert!(verify(&path, &k).is_valid());
+        }
+        // And after real changes, re-signed.
+        let mut m = HDF5Memory::open(&path).unwrap();
+        m.set_signing_key(k.clone());
+        m.save(entry(99, "added later")).unwrap();
+        m.hybrid_search(&entry(1, "").embedding, "text", 0.4, 0.6, 5);
+        m.flush_wal().unwrap();
+        drop(m);
+        let r = verify(&path, &k);
+        assert!(r.is_valid(), "{r:?}");
+        assert_eq!(r.record_count, 31);
+    }
+}
+
+#[test]
+fn a_signed_store_refuses_to_checkpoint_without_its_key() {
+    let dir = TempDir::new().unwrap();
+    let k = key(1);
+    let path = signed_store(&dir, true, &k);
+
+    let mut m = HDF5Memory::open(&path).unwrap();
+    m.save(entry(50, "pending")).unwrap();
+    match m.flush_wal() {
+        Err(MemoryError::SigningKeyRequired(msg)) => assert!(msg.contains("signed"), "{msg}"),
+        other => panic!("expected SigningKeyRequired, got {other:?}"),
+    }
+    // The file is untouched and still valid; the save is still in the WAL.
+    let r = verify(&path, &k);
+    assert!(r.is_valid());
+    assert_eq!(r.wal_entries_unsigned, 1);
+
+    // Supplying the key lets the checkpoint through, signed.
+    m.set_signing_key(k.clone());
+    m.flush_wal().unwrap();
+    drop(m);
+    let r = verify(&path, &k);
+    assert!(r.is_valid());
+    assert_eq!((r.record_count, r.wal_entries_unsigned), (31, 0));
+
+    // Removing the signature on purpose writes it unsigned.
+    let mut m = HDF5Memory::open(&path).unwrap();
+    m.remove_signature();
+    m.flush_wal().unwrap();
+    drop(m);
+    let r = verify(&path, &k);
+    assert!(!r.signed && !r.is_valid());
+    assert!(!HDF5Memory::open(&path).unwrap().is_signed());
+}
+
+#[test]
+fn the_wrong_key_does_not_verify_and_a_new_key_re_signs() {
+    let dir = TempDir::new().unwrap();
+    let (a, b) = (key(1), key(2));
+    let path = signed_store(&dir, true, &a);
+    let r = verify(&path, &b);
+    assert!(r.signed && !r.key_matches && !r.signature_valid && !r.is_valid());
+
+    let mut m = HDF5Memory::open(&path).unwrap();
+    m.set_signing_key(b.clone());
+    m.flush_wal().unwrap();
+    drop(m);
+    assert!(verify(&path, &b).is_valid());
+    assert!(!verify(&path, &a).is_valid());
+}
+
+/// Rewrite the store with changed contents but the *old* signature — what
+/// someone with write access to the file, but not the key, can do.
+fn tamper(path: &Path, change: impl FnOnce(&mut Tampered)) {
+    let file = clawhdf5::File::open(path).unwrap();
+    let (config, cache, sessions, knowledge) = schema::validate_and_load(&file).unwrap();
+    let checkpoint = schema::read_checkpoint_meta(&file);
+    let signature = schema::read_signature(&file).unwrap().unwrap();
+    drop(file);
+    let mut t = Tampered {
+        config,
+        cache,
+        sessions,
+        knowledge,
+    };
+    change(&mut t);
+    storage::write_to_disk_signed(
+        path,
+        &t.config,
+        &t.cache,
+        &t.sessions,
+        &t.knowledge,
+        &checkpoint,
+        Some(&signature),
+    )
+    .unwrap();
+}
+
+struct Tampered {
+    config: MemoryConfig,
+    cache: clawhdf5_agent::cache::MemoryCache,
+    sessions: clawhdf5_agent::SessionCache,
+    knowledge: clawhdf5_agent::knowledge::KnowledgeCache,
+}
+
+#[test]
+fn every_kind_of_edit_is_detected_and_located() {
+    let k = key(3);
+    type Edit = Box<dyn FnOnce(&mut Tampered)>;
+    type Case = (&'static str, Edit, fn(&VerifyReport) -> bool);
+    let cases: Vec<Case> = vec![
+        (
+            "record text",
+            Box::new(|t: &mut Tampered| t.cache.chunks[7] = "rewritten".into()),
+            |r| !r.records_match && r.changed_records == vec![7],
+        ),
+        (
+            "one embedding value",
+            Box::new(|t: &mut Tampered| {
+                let mut e = t.cache.embeddings[12].to_vec();
+                e[3] = 0.5;
+                t.cache.embeddings.set(12, &e);
+            }),
+            |r| r.changed_records == vec![12],
+        ),
+        (
+            "undelete",
+            Box::new(|t: &mut Tampered| t.cache.tombstones[4] = 0),
+            |r| r.changed_records == vec![4],
+        ),
+        (
+            "timestamp",
+            Box::new(|t: &mut Tampered| t.cache.timestamps[20] += 1.0),
+            |r| r.changed_records == vec![20],
+        ),
+        (
+            "record appended",
+            Box::new(|t: &mut Tampered| {
+                t.cache.push(
+                    "new".into(),
+                    vec![0.1; DIM],
+                    "x".into(),
+                    1.0,
+                    "s".into(),
+                    "".into(),
+                );
+            }),
+            |r| !r.records_match && r.changed_records == vec![30] && r.record_count == 31,
+        ),
+        (
+            "setting",
+            Box::new(|t: &mut Tampered| t.config.agent_id = "someone-else".into()),
+            |r| !r.settings_match && r.records_match,
+        ),
+        (
+            "session summary",
+            Box::new(|t: &mut Tampered| t.sessions.summaries[0] = "edited".into()),
+            |r| !r.sessions_match && r.records_match,
+        ),
+        (
+            "graph edge",
+            Box::new(|t: &mut Tampered| t.knowledge.relations[0].weight = 1.0),
+            |r| !r.graph_match && r.records_match,
+        ),
+    ];
+    for (name, edit, check) in cases {
+        let dir = TempDir::new().unwrap();
+        let path = signed_store(&dir, true, &k);
+        tamper(&path, edit);
+        let r = verify(&path, &k);
+        assert!(
+            r.signed && r.key_matches && r.signature_valid,
+            "{name}: {r:?}"
+        );
+        assert!(!r.is_valid(), "{name}: edit not detected: {r:?}");
+        assert!(check(&r), "{name}: {r:?}");
+    }
+}
+
+#[test]
+fn a_forged_manifest_fails_the_signature() {
+    // Recomputing the hashes for tampered contents does not help without the
+    // key: the signature no longer matches the manifest.
+    let dir = TempDir::new().unwrap();
+    let k = key(5);
+    let path = signed_store(&dir, true, &k);
+    let file = clawhdf5::File::open(&path).unwrap();
+    let (config, mut cache, sessions, knowledge) = schema::validate_and_load(&file).unwrap();
+    let checkpoint = schema::read_checkpoint_meta(&file);
+    let mut sig = schema::read_signature(&file).unwrap().unwrap();
+    drop(file);
+    cache.chunks[0] = "forged".into();
+    // Re-sign with an attacker key, then splice the victim's public key back.
+    let forged = clawhdf5_agent::signing::sign(
+        &key(66),
+        &config,
+        &cache,
+        &sessions,
+        &knowledge,
+        checkpoint.wal_applied,
+    );
+    sig.manifest = forged.manifest;
+    sig.record_hashes = forged.record_hashes;
+    storage::write_to_disk_signed(
+        &path,
+        &config,
+        &cache,
+        &sessions,
+        &knowledge,
+        &checkpoint,
+        Some(&sig),
+    )
+    .unwrap();
+    let r = verify(&path, &k);
+    assert!(
+        r.key_matches && !r.signature_valid && !r.is_valid(),
+        "{r:?}"
+    );
+}
+
+#[test]
+fn an_unsigned_store_reports_unsigned() {
+    let dir = TempDir::new().unwrap();
+    let mut m = HDF5Memory::create(MemoryConfig::new(dir.path().join("u.h5"), "a", DIM)).unwrap();
+    m.save_batch(vec![entry(0, "hello")]).unwrap();
+    drop(m);
+    let r = HDF5Memory::verify(&dir.path().join("u.h5"), &VerifyingKey::from(&key(1))).unwrap();
+    assert!(!r.signed && !r.is_valid());
+    assert_eq!(r.record_count, 1);
+}
+
+#[test]
+fn nul_bytes_in_text_still_verify() {
+    // Strings are stored null-padded; the hash must follow what a reopened
+    // store actually holds, or an untouched store would fail to verify.
+    let dir = TempDir::new().unwrap();
+    let k = key(9);
+    let mut m = HDF5Memory::create(MemoryConfig::new(dir.path().join("n.h5"), "a", DIM)).unwrap();
+    m.set_signing_key(k.clone());
+    m.save_batch(vec![
+        entry(0, "inner\0nul"),
+        entry(1, "trailing nul\0"),
+        entry(2, "\0leading"),
+    ])
+    .unwrap();
+    drop(m);
+    let r = verify(&dir.path().join("n.h5"), &k);
+    assert!(r.is_valid(), "{r:?}");
+    let m = HDF5Memory::open(&dir.path().join("n.h5")).unwrap();
+    eprintln!(
+        "reloaded: {:?}",
+        (0..3).map(|i| m.get_chunk(i)).collect::<Vec<_>>()
+    );
+}

crates/clawhdf5-ann/src/hnsw.rs

@@ -13,7 +13,7 @@ use clawhdf5_format::filter_pipeline::FilterPipeline;
 use clawhdf5_format::group_v2::resolve_path_any;
 use clawhdf5_format::message_type::MessageType;
 use clawhdf5_format::object_header::ObjectHeader;
-use clawhdf5_format::signature::find_signature;
+use clawhdf5_format::signature::split_user_block;
 use clawhdf5_format::superblock::Superblock;
 use clawhdf5_io::FileWriter as IoFileWriter;
 
@@ -861,8 +861,9 @@ impl HnswIndex {
     /// The HDF5 data must contain the `/ann/vectors`, `/ann/graph_layer_*`,
     /// and `/ann/config` datasets as produced by [`to_hdf5_bytes`].
     pub fn load_from_hdf5(data: &[u8]) -> Result<Self, FormatError> {
-        let sig_offset = find_signature(data)?;
-        let sb = Superblock::parse(data, sig_offset)?;
+        // Addresses are relative to the superblock: skip any user block.
+        let (_, data) = split_user_block(data)?;
+        let sb = Superblock::parse(data, 0)?;
 
         // Read config dataset and its attributes
         let config_attrs = read_dataset_attrs(data, &sb, "ann/config")?;

crates/clawhdf5-bench/Cargo.toml

@@ -34,6 +34,10 @@ path = "src/bin/consolidation_efficiency.rs"
 name = "ephemeral_perf"
 path = "src/bin/ephemeral_perf.rs"
 
+[[bin]]
+name = "concurrent_read"
+path = "src/bin/concurrent_read.rs"
+
 [[bin]]
 name = "mpi_io_bench"
 path = "src/bin/mpi_io_bench.rs"
@@ -64,6 +68,10 @@ clawhdf5-io = { path = "../clawhdf5-io" }
 mpi = { version = "0.8", optional = true }
 serde = { workspace = true }
 serde_json = "1"
+# concurrent_read: size the decode pool (--decode-threads) and evict files
+# from the page cache (--cold, posix_fadvise). Both pure Rust / bindings only.
+rayon = "1"
+libc = "0.2"
 tempfile = { workspace = true }
 # Optional: libhdf5 C wrapper for side-by-side comparison (requires system libhdf5).
 # Enable with: cargo bench -p clawhdf5-bench --features libhdf5-compare

crates/clawhdf5-bench/scripts/compare_concurrent_read.py

@@ -0,0 +1,70 @@
+#!/usr/bin/env python3
+"""Tabulate concurrent_read JSON results (clawhdf5, h5py threads/processes).
+
+    python compare_concurrent_read.py clawhdf5.json h5py-threads.json h5py-procs.json
+
+Prints one Markdown table: for each layout, mode and thread count, every
+tool's MB/s and scaling efficiency, and the first file's MB/s relative to each
+of the others. Refuses to compare runs whose workload parameters differ.
+"""
+
+import json
+import sys
+
+COMPARED = ("datasets", "rows", "cols", "chunk", "deflate_level", "slab", "slabs", "seed")
+
+
+def main(paths):
+    if len(paths) < 2:
+        sys.exit(__doc__)
+    docs = []
+    for p in paths:
+        with open(p) as fh:
+            docs.append(json.load(fh))
+    ref = docs[0]
+    for d, p in zip(docs[1:], paths[1:]):
+        diff = [k for k in COMPARED if d["params"].get(k) != ref["params"].get(k)]
+        if diff:
+            sys.exit(f"{p}: workload differs from {paths[0]} in {', '.join(diff)}")
+        if d["cache"] != ref["cache"]:
+            print(f"warning: {p} ran {d['cache']!r}, {paths[0]} ran {ref['cache']!r}",
+                  file=sys.stderr)
+        if d.get("host") != ref.get("host"):
+            print(f"warning: {p} ran on {d.get('host')}, {paths[0]} on {ref.get('host')}",
+                  file=sys.stderr)
+
+    names = [d["tool"] for d in docs]
+    for d in docs:
+        extra = f", HDF5 {d['hdf5_version']}" if "hdf5_version" in d else ""
+        print(f"- {d['tool']} {d['version']}{extra}: host {d.get('host')}, "
+              f"{d.get('cpus')} CPUs, cache {d['cache']}, decode threads per read "
+              f"{d.get('decode_threads')}")
+    p = ref["params"]
+    print(f"\n{p['datasets']} datasets of {p['rows']} x {p['cols']} f32, chunks "
+          f"{p['chunk'][0]} x {p['chunk'][1]} (deflate {p['deflate_level']}); "
+          f"`same`: {p['slabs']} slabs of {p['slab']} x {p['slab']}\n")
+
+    index = [{(r["layout"], r["mode"], r["threads"]): r for r in d["results"]} for d in docs]
+    keys = [(r["layout"], r["mode"], r["threads"]) for r in ref["results"]]
+
+    head = ["layout", "mode", "threads"]
+    head += [f"{n} MB/s (eff)" for n in names]
+    head += [f"{names[0]} / {n}" for n in names[1:]]
+    print("| " + " | ".join(head) + " |")
+    print("|---|---|" + "---:|" * (len(head) - 2))
+    for key in keys:
+        cells = [key[0], key[1], str(key[2])]
+        rs = [ix.get(key) for ix in index]
+        for r in rs:
+            if r is None:
+                cells.append("-")
+            else:
+                eff = "-" if r["efficiency"] is None else f"{r['efficiency']:.2f}"
+                cells.append(f"{r['mb_s']:.0f} ({eff})")
+        for r in rs[1:]:
+            cells.append("-" if r is None else f"{rs[0]['mb_s'] / r['mb_s']:.2f}x")
+        print("| " + " | ".join(cells) + " |")
+
+
+if __name__ == "__main__":
+    main(sys.argv[1:])

crates/clawhdf5-bench/scripts/concurrent_read_h5py.py

@@ -0,0 +1,265 @@
+#!/usr/bin/env python3
+"""The concurrent_read workload with h5py, on the files concurrent_read wrote.
+
+libhdf5 serialises every API call under one global lock, and h5py holds its
+own global lock around every call as well, so h5py *threads* cannot decode in
+parallel. h5py users scale with *processes* instead; ``--executor processes``
+measures that (each worker opens the file itself).
+
+The workload mirrors ``crates/clawhdf5-bench/src/bin/concurrent_read.rs``:
+
+* ``distinct``: every dataset read in full once per repetition; worker ``t``
+  of ``T`` reads datasets ``t, t + T, ...``.
+* ``same``: ``--slabs`` random ``--slab`` x ``--slab`` hyperslabs of ``d00``
+  (slab ``j`` to worker ``j % T``), offsets from the same splitmix64 stream.
+
+Each worker times itself from a start barrier; a repetition spans the earliest
+start to the latest finish (CLOCK_MONOTONIC, comparable across processes).
+Threads share one ``h5py.File`` per repetition; process workers open the file
+inside the timed region (a few ms against reads of many MiB).
+
+Generate the files first with the Rust harness (it writes ``manifest.json``),
+then, for example::
+
+    python concurrent_read_h5py.py --dir DIR --executor threads --json h5py-threads.json
+    python concurrent_read_h5py.py --dir DIR --executor processes --json h5py-procs.json
+"""
+
+import argparse
+import json
+import multiprocessing as mp
+import os
+import platform
+import socket
+import sys
+import threading
+import time
+
+import h5py
+import numpy as np
+
+M64 = (1 << 64) - 1
+
+
+def splitmix64(state):
+    """Return (new_state, value); the same stream as the Rust harness."""
+    state = (state + 0x9E3779B97F4A7C15) & M64
+    z = state
+    z = ((z ^ (z >> 30)) * 0xBF58476D1CE4E5B9) & M64
+    z = ((z ^ (z >> 27)) * 0x94D049BB133111EB) & M64
+    return state, z ^ (z >> 31)
+
+
+def value(k, i):
+    """Element i (row-major) of dataset k, exactly as concurrent_read writes it."""
+    _, noise = splitmix64(i ^ (k << 40))
+    return np.float32((((i >> 6) % 16384) + k) + (noise & 0xFF) / 256.0)
+
+
+def slab_offsets(seed, count, rows, cols, slab):
+    s = seed
+    out = []
+    for _ in range(count):
+        s, r = splitmix64(s)
+        s, c = splitmix64(s)
+        out.append((r % (rows - slab + 1), c % (cols - slab + 1)))
+    return out
+
+
+def now():
+    return time.clock_gettime(time.CLOCK_MONOTONIC)
+
+
+def work(f, mode, t, threads, m, slabs, slab, verify):
+    """Worker t's share of one repetition on an open h5py.File."""
+    n = m["rows"] * m["cols"]
+    if mode == "distinct":
+        for k in range(t, m["datasets"], threads):
+            got = f[f"d{k:02d}"][...]
+            assert got.size == n
+            if verify:
+                flat = got.reshape(-1)
+                for i in (0, n // 3, n - 1):
+                    assert flat[i] == value(k, i), f"d{k:02d}[{i}]"
+    else:
+        ds = f["d00"]
+        cols = m["cols"]
+        for r, c in slabs[t::threads]:
+            got = ds[r : r + slab, c : c + slab]
+            assert got.shape == (slab, slab)
+            if verify:
+                assert got[0, 0] == value(0, r * cols + c)
+                last = (r + slab - 1) * cols + c + slab - 1
+                assert got[-1, -1] == value(0, last)
+
+
+# ----- process workers ------------------------------------------------------
+
+_barrier = None
+
+
+def _init(barrier):
+    global _barrier
+    _barrier = barrier
+
+
+def _proc_task(task):
+    path, mode, t, threads, m, slabs, slab = task
+    _barrier.wait()
+    start = now()
+    with h5py.File(path, "r") as f:
+        work(f, mode, t, threads, m, slabs, slab, False)
+    return start, now()
+
+
+def _noop(_):
+    return os.getpid()
+
+
+def run_threads(path, mode, threads, m, slabs, slab):
+    spans = [None] * threads
+    barrier = threading.Barrier(threads)
+    with h5py.File(path, "r") as f:
+
+        def body(t):
+            barrier.wait()
+            start = now()
+            work(f, mode, t, threads, m, slabs, slab, False)
+            spans[t] = (start, now())
+
+        ts = [threading.Thread(target=body, args=(t,)) for t in range(threads)]
+        for th in ts:
+            th.start()
+        for th in ts:
+            th.join()
+    return max(e for _, e in spans) - min(s for s, _ in spans)
+
+
+def run_processes(pool, path, mode, threads, m, slabs, slab):
+    tasks = [(path, mode, t, threads, m, slabs, slab) for t in range(threads)]
+    # One task per worker: each blocks in the barrier until all T have
+    # started, so no worker can take a second task.
+    spans = pool.map(_proc_task, tasks, chunksize=1)
+    return max(e for _, e in spans) - min(s for s, _ in spans)
+
+
+def warm(path):
+    with open(path, "rb") as fh:
+        while fh.read(1 << 24):
+            pass
+
+
+def evict(path):
+    fd = os.open(path, os.O_RDONLY)
+    try:
+        os.posix_fadvise(fd, 0, 0, os.POSIX_FADV_DONTNEED)
+    finally:
+        os.close(fd)
+
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__.split("\n\n")[0])
+    ap.add_argument("--dir", default="concurrent-read-data")
+    ap.add_argument("--executor", choices=["threads", "processes"], default="threads")
+    ap.add_argument("--threads", default="1,2,4,8,16")
+    ap.add_argument("--reps", type=int, default=3)
+    ap.add_argument("--slab", type=int, default=256)
+    ap.add_argument("--slabs", type=int, default=1024)
+    ap.add_argument("--seed", type=int, default=42)
+    ap.add_argument("--cold", action="store_true")
+    ap.add_argument("--modes", default="distinct,same")
+    ap.add_argument("--layouts", default="deflate,contiguous")
+    ap.add_argument("--json")
+    a = ap.parse_args()
+
+    # The Rust harness pins this value (splitmix64_reference).
+    assert splitmix64(42)[1] == 0xBDD732262FEB6E95, "splitmix64 port is wrong"
+
+    try:
+        with open(os.path.join(a.dir, "manifest.json")) as fh:
+            m = json.load(fh)
+    except FileNotFoundError:
+        sys.exit(f"{a.dir}/manifest.json not found: generate the files with "
+                 "`cargo run --release -p clawhdf5-bench --bin concurrent_read -- --dir ...` first")
+    threads_list = [int(x) for x in a.threads.split(",")]
+    modes = a.modes.split(",")
+    layouts = a.layouts.split(",")
+    if a.slab < 1 or a.slab > min(m["rows"], m["cols"]):
+        sys.exit(f"--slab must be 1..={min(m['rows'], m['cols'])}")
+    files = dict(m["files"])
+    slabs = slab_offsets(a.seed, a.slabs, m["rows"], m["cols"], a.slab)
+    dataset_bytes = m["rows"] * m["cols"] * 4
+    tool = f"h5py-{a.executor}"
+
+    ctx = mp.get_context("spawn")  # never fork a process holding HDF5 state
+    pools = {}
+    if a.executor == "processes":
+        for t in threads_list:
+            pool = ctx.Pool(t, initializer=_init, initargs=(ctx.Barrier(t),))
+            pool.map(_noop, range(t))  # start the workers outside the timing
+            pools[t] = pool
+
+    rows = []
+    print("| layout | mode | threads | MB/s | efficiency | median s |")
+    print("|---|---|---:|---:|---:|---:|")
+    try:
+        for layout in layouts:
+            path = os.path.join(a.dir, files[layout])
+            if not a.cold:
+                warm(path)
+            for mode in modes:
+                with h5py.File(path, "r") as f:  # untimed, checked pass
+                    work(f, mode, 0, 1, m, slabs, a.slab, True)
+                nbytes = (dataset_bytes * m["datasets"] if mode == "distinct"
+                          else a.slab * a.slab * 4 * a.slabs)
+                base = None
+                for t in threads_list:
+                    times = []
+                    for _ in range(a.reps):
+                        if a.cold:
+                            evict(path)
+                        if a.executor == "threads":
+                            times.append(run_threads(path, mode, t, m, slabs, a.slab))
+                        else:
+                            times.append(run_processes(pools[t], path, mode, t, m, slabs, a.slab))
+                    med = sorted(times)[len(times) // 2]
+                    mb_s = nbytes / (1 << 20) / med
+                    if t == 1:
+                        base = mb_s
+                    eff = mb_s / (t * base) if base else None
+                    print(f"| {layout} | {mode} | {t} | {mb_s:.0f} | "
+                          f"{'-' if eff is None else f'{eff:.2f}'} | {med:.4f} |")
+                    rows.append({
+                        "layout": layout, "mode": mode, "threads": t, "bytes": nbytes,
+                        "times_s": times, "median_s": med, "mb_s": mb_s, "efficiency": eff,
+                    })
+    finally:
+        for pool in pools.values():
+            pool.terminate()
+
+    if a.json:
+        doc = {
+            "tool": tool,
+            "version": h5py.__version__,
+            "hdf5_version": h5py.version.hdf5_version,
+            "python": platform.python_version(),
+            "host": socket.gethostname(),
+            "cpus": os.cpu_count(),
+            "unix_time": int(time.time()),
+            "cache": ("cold (posix_fadvise DONTNEED before each repetition)"
+                      if a.cold else "warm"),
+            "decode_threads": 1,
+            "params": {
+                "datasets": m["datasets"], "rows": m["rows"], "cols": m["cols"],
+                "chunk": m["chunk"], "deflate_level": m["deflate_level"],
+                "mib": dataset_bytes // (1 << 20), "slab": a.slab, "slabs": a.slabs,
+                "seed": a.seed, "reps": a.reps, "dir": a.dir,
+            },
+            "results": rows,
+        }
+        with open(a.json, "w") as fh:
+            json.dump(doc, fh, indent=2)
+
+
+if __name__ == "__main__":
+    main()

crates/clawhdf5-bench/src/bin/concurrent_read.rs

@@ -0,0 +1,523 @@
+//! Concurrent-read harness: how does decoded read throughput scale with the
+//! number of threads reading one open file?
+//!
+//! libhdf5 (threadsafe build) serialises every API call under one global
+//! mutex, and h5py holds it too, so threads cannot decode in parallel there.
+//! A clawhdf5 [`File`] is `Send + Sync`; this harness measures what that buys.
+//! `crates/clawhdf5-bench/scripts/concurrent_read_h5py.py` runs the same
+//! workload on the same files with h5py (threads, and processes), and
+//! `compare_concurrent_read.py` tabulates the JSON both write.
+//!
+//! Files (generated on first use, reused while `manifest.json` matches):
+//!
+//! * `<dir>/deflate.h5`: `--datasets` datasets `d00`, `d01`, ... of `f32`,
+//!   `--mib` MiB decoded each, shape `[mib * 256, 1024]`, chunks `256 x 256`,
+//!   deflate level 4.
+//! * `<dir>/contiguous.h5`: the same datasets, contiguous.
+//!
+//! Modes, for each layout and each thread count `T` (strong scaling: the total
+//! work per repetition is fixed, split among the threads):
+//!
+//! * `distinct`: every dataset is read in full once; thread `t` reads datasets
+//!   `t, t + T, t + 2T, ...`.
+//! * `same`: all threads read `d00`, `--slabs` random `--slab` x `--slab`
+//!   hyperslabs in total (slab `j` goes to thread `j % T`). The offsets come
+//!   from a splitmix64 stream seeded with `--seed`, identical in the h5py
+//!   script.
+//!
+//! One `File` per layout per repetition is shared by all threads (opened
+//! fresh each repetition, so no chunk cache carries over). Page cache:
+//! `warm` (default) reads every file once before timing; `--cold` evicts the
+//! files from the page cache with `posix_fadvise(POSIX_FADV_DONTNEED)` before
+//! every repetition (no root needed; it only evicts clean, unmapped pages, so
+//! it is best effort — the JSON says which was used).
+//!
+//! Decode inside one read is itself parallel when clawhdf5-format's `parallel`
+//! feature is on (it is in this binary, via clawhdf5-agent). `--decode-threads
+//! N` sizes that rayon pool; `--decode-threads 1` measures the API's own
+//! thread scaling, comparable with h5py where each call decodes on the
+//! calling thread.
+//!
+//! ```text
+//! cargo run --release -p clawhdf5-bench --bin concurrent_read -- \
+//!     --dir /data/concurrent-read --json clawhdf5.json
+//! cargo run --release -p clawhdf5-bench --bin concurrent_read -- \
+//!     --dir /tmp/cr --datasets 4 --mib 1 --threads 1,2 --slabs 16 --reps 1   # smoke
+//! ```
+
+use std::path::{Path, PathBuf};
+use std::sync::Barrier;
+use std::time::Instant;
+
+use clawhdf5::{File, FileBuilder, Selection};
+use serde::{Deserialize, Serialize};
+
+const COLS: u64 = 1024;
+const ROWS_PER_MIB: u64 = 256; // 256 rows x 1024 cols x 4 bytes = 1 MiB
+const CHUNK: u64 = 256;
+const DEFLATE_LEVEL: u32 = 4;
+const LAYOUTS: [&str; 2] = ["deflate", "contiguous"];
+const MANIFEST_VERSION: u32 = 1;
+
+/// splitmix64 — shared with the h5py script, which must produce the same
+/// stream (both the data and the hyperslab offsets depend on it).
+fn splitmix64(state: &mut u64) -> u64 {
+    *state = state.wrapping_add(0x9E37_79B9_7F4A_7C15);
+    let mut z = *state;
+    z = (z ^ (z >> 30)).wrapping_mul(0xBF58_476D_1CE4_E5B9);
+    z = (z ^ (z >> 27)).wrapping_mul(0x94D0_49BB_1331_11EB);
+    z ^ (z >> 31)
+}
+
+/// Element `i` (row-major) of dataset `k`: a slowly varying integer part plus
+/// 8 bits of noise, so deflate has real work to do (about 3.1x) and every value
+/// is exact in `f32` (< 2^15 with 8 fraction bits), which lets both harnesses
+/// check what they read against this formula.
+fn value(k: u64, i: u64) -> f32 {
+    let mut s = i ^ (k << 40);
+    let noise = splitmix64(&mut s) & 0xff;
+    (((i >> 6) % 16384) + k) as f32 + noise as f32 / 256.0
+}
+
+#[derive(Serialize, Deserialize, PartialEq, Debug, Clone)]
+struct Manifest {
+    version: u32,
+    datasets: u64,
+    rows: u64,
+    cols: u64,
+    chunk: [u64; 2],
+    deflate_level: u32,
+    files: Vec<(String, String)>, // (layout, file name)
+    writer: String,
+}
+
+fn manifest_for(datasets: u64, mib: u64) -> Manifest {
+    Manifest {
+        version: MANIFEST_VERSION,
+        datasets,
+        rows: mib * ROWS_PER_MIB,
+        cols: COLS,
+        chunk: [CHUNK, CHUNK],
+        deflate_level: DEFLATE_LEVEL,
+        files: LAYOUTS
+            .iter()
+            .map(|l| (l.to_string(), format!("{l}.h5")))
+            .collect(),
+        writer: format!("clawhdf5 {}", env!("CARGO_PKG_VERSION")),
+    }
+}
+
+fn dataset_values(k: u64, n: u64) -> Vec<f32> {
+    (0..n).map(|i| value(k, i)).collect()
+}
+
+/// Write the files unless `dir` already holds ones matching `want`.
+fn ensure_files(dir: &Path, want: &Manifest) -> std::io::Result<bool> {
+    let manifest_path = dir.join("manifest.json");
+    if let Ok(text) = std::fs::read_to_string(&manifest_path)
+        && let Ok(have) = serde_json::from_str::<Manifest>(&text)
+        && have.version == want.version
+        && have.datasets == want.datasets
+        && have.rows == want.rows
+        && have.cols == want.cols
+        && have.chunk == want.chunk
+        && have.deflate_level == want.deflate_level
+        && have.files == want.files
+        && want.files.iter().all(|(_, f)| dir.join(f).exists())
+    {
+        return Ok(false);
+    }
+    std::fs::create_dir_all(dir)?;
+    // A stale manifest must not survive a half-written regeneration.
+    let _ = std::fs::remove_file(&manifest_path);
+    let n = want.rows * want.cols;
+    for (layout, file) in &want.files {
+        // One layout at a time keeps the peak memory to about twice one
+        // file's decoded size.
+        let mut b = FileBuilder::new();
+        for k in 0..want.datasets {
+            let ds = b.create_dataset(&format!("d{k:02}"));
+            ds.with_f32_data(&dataset_values(k, n))
+                .with_shape(&[want.rows, want.cols]);
+            if layout == "deflate" {
+                ds.with_chunks(&[CHUNK.min(want.rows), CHUNK])
+                    .with_deflate(DEFLATE_LEVEL);
+            }
+        }
+        b.write(dir.join(file)).map_err(std::io::Error::other)?;
+    }
+    std::fs::write(
+        &manifest_path,
+        serde_json::to_string_pretty(want).map_err(std::io::Error::other)?,
+    )?;
+    Ok(true)
+}
+
+fn slab_offsets(seed: u64, count: usize, rows: u64, cols: u64, slab: u64) -> Vec<(u64, u64)> {
+    let mut s = seed;
+    (0..count)
+        .map(|_| {
+            let r = splitmix64(&mut s) % (rows - slab + 1);
+            let c = splitmix64(&mut s) % (cols - slab + 1);
+            (r, c)
+        })
+        .collect()
+}
+
+/// Warm the page cache by reading every byte of `path`.
+fn warm(path: &Path) -> std::io::Result<()> {
+    let mut f = std::fs::File::open(path)?;
+    std::io::copy(&mut f, &mut std::io::sink())?;
+    Ok(())
+}
+
+/// Ask the kernel to drop `path`'s pages from the page cache.
+fn evict(path: &Path) -> std::io::Result<()> {
+    use std::os::fd::AsRawFd;
+    let f = std::fs::File::open(path)?;
+    // SAFETY: plain syscall on a valid, open file descriptor.
+    let rc = unsafe { libc::posix_fadvise(f.as_raw_fd(), 0, 0, libc::POSIX_FADV_DONTNEED) };
+    if rc != 0 {
+        return Err(std::io::Error::from_raw_os_error(rc));
+    }
+    Ok(())
+}
+
+#[derive(Serialize)]
+struct Row {
+    layout: String,
+    mode: String,
+    threads: usize,
+    /// Decoded (selected) bytes read per repetition.
+    bytes: u64,
+    times_s: Vec<f64>,
+    median_s: f64,
+    mb_s: f64,
+    /// `mb_s / (threads * mb_s at threads = 1)`; null without a 1-thread row.
+    efficiency: Option<f64>,
+}
+
+struct Args {
+    dir: PathBuf,
+    datasets: u64,
+    mib: u64,
+    threads: Vec<usize>,
+    reps: usize,
+    slab: u64,
+    slabs: usize,
+    seed: u64,
+    cold: bool,
+    decode_threads: usize,
+    modes: Vec<String>,
+    layouts: Vec<String>,
+    json: Option<PathBuf>,
+}
+
+const USAGE: &str = "\
+usage: concurrent_read [--dir DIR] [--datasets N] [--mib N] [--threads 1,2,4,8,16]
+                       [--reps N] [--slab N] [--slabs N] [--seed N] [--cold]
+                       [--decode-threads N] [--modes distinct,same]
+                       [--layouts deflate,contiguous] [--json FILE]";
+
+fn parse_list<T: std::str::FromStr>(s: &str) -> Result<Vec<T>, String> {
+    s.split(',')
+        .map(|x| x.trim().parse().map_err(|_| format!("bad list item {x:?}")))
+        .collect()
+}
+
+fn parse_args() -> Result<Args, String> {
+    let mut a = Args {
+        dir: PathBuf::from("concurrent-read-data"),
+        datasets: 64,
+        mib: 64,
+        threads: vec![1, 2, 4, 8, 16],
+        reps: 3,
+        slab: 256,
+        slabs: 1024,
+        seed: 42,
+        cold: false,
+        decode_threads: 0,
+        modes: vec!["distinct".into(), "same".into()],
+        layouts: LAYOUTS.iter().map(|s| s.to_string()).collect(),
+        json: None,
+    };
+    let mut it = std::env::args().skip(1);
+    while let Some(flag) = it.next() {
+        if flag == "--cold" {
+            a.cold = true;
+            continue;
+        }
+        if flag == "-h" || flag == "--help" {
+            return Err(USAGE.into());
+        }
+        let v = it.next().ok_or(format!("{flag} needs a value\n{USAGE}"))?;
+        let num = |v: &str| {
+            v.parse::<u64>()
+                .map_err(|_| format!("{flag}: bad number {v:?}"))
+        };
+        match flag.as_str() {
+            "--dir" => a.dir = v.into(),
+            "--datasets" => a.datasets = num(&v)?,
+            "--mib" => a.mib = num(&v)?,
+            "--threads" => a.threads = parse_list(&v)?,
+            "--reps" => a.reps = num(&v)? as usize,
+            "--slab" => a.slab = num(&v)?,
+            "--slabs" => a.slabs = num(&v)? as usize,
+            "--seed" => a.seed = num(&v)?,
+            "--decode-threads" => a.decode_threads = num(&v)? as usize,
+            "--modes" => a.modes = parse_list(&v)?,
+            "--layouts" => a.layouts = parse_list(&v)?,
+            "--json" => a.json = Some(v.into()),
+            _ => return Err(format!("unknown flag {flag}\n{USAGE}")),
+        }
+    }
+    if a.datasets == 0 || a.datasets > 100 {
+        return Err("--datasets must be 1..=100".into());
+    }
+    if a.mib == 0 || a.reps == 0 || a.slabs == 0 || a.threads.contains(&0) {
+        return Err("--mib, --reps, --slabs and every --threads value must be > 0".into());
+    }
+    if a.slab == 0 || a.slab > COLS || a.slab > a.mib * ROWS_PER_MIB {
+        return Err(format!(
+            "--slab must be 1..={}",
+            COLS.min(a.mib * ROWS_PER_MIB)
+        ));
+    }
+    for m in &a.modes {
+        if m != "distinct" && m != "same" {
+            return Err(format!("unknown mode {m:?}"));
+        }
+    }
+    for l in &a.layouts {
+        if !LAYOUTS.contains(&l.as_str()) {
+            return Err(format!("unknown layout {l:?}"));
+        }
+    }
+    Ok(a)
+}
+
+/// One timed repetition: `T` threads on one shared `File`. Returns seconds.
+fn run_once(
+    path: &Path,
+    mode: &str,
+    threads: usize,
+    m: &Manifest,
+    slabs: &[(u64, u64)],
+    slab: u64,
+    verify: bool,
+) -> f64 {
+    let file = File::open(path).expect("open");
+    let barrier = Barrier::new(threads + 1); // + the spawning thread
+    let n = m.rows * m.cols;
+    // Each thread times itself from the barrier; the repetition spans the
+    // earliest start to the latest finish (timing on the spawning thread
+    // instead undercounts whenever it is scheduled after the workers ran).
+    let spans: Vec<(Instant, Instant)> = std::thread::scope(|s| {
+        let handles: Vec<_> = (0..threads)
+            .map(|t| {
+                let (file, barrier) = (&file, &barrier);
+                s.spawn(move || {
+                    barrier.wait();
+                    let start = Instant::now();
+                    match mode {
+                        "distinct" => {
+                            for k in (t as u64..m.datasets).step_by(threads) {
+                                let got = file.dataset(&format!("d{k:02}")).unwrap().read_f32();
+                                let got = got.unwrap();
+                                assert_eq!(got.len() as u64, n);
+                                if verify {
+                                    for i in [0, n / 3, n - 1] {
+                                        assert_eq!(got[i as usize], value(k, i), "d{k:02}[{i}]");
+                                    }
+                                }
+                                std::hint::black_box(got);
+                            }
+                        }
+                        _ => {
+                            let ds = file.dataset("d00").unwrap();
+                            for &(r, c) in slabs.iter().skip(t).step_by(threads) {
+                                let sel = Selection::Hyperslab {
+                                    start: vec![r, c],
+                                    stride: vec![1, 1],
+                                    count: vec![slab, slab],
+                                    block: vec![1, 1],
+                                };
+                                let got = ds.read_f32_selection(&sel).unwrap();
+                                assert_eq!(got.len() as u64, slab * slab);
+                                if verify {
+                                    let last = (r + slab - 1) * m.cols + c + slab - 1;
+                                    assert_eq!(got[0], value(0, r * m.cols + c));
+                                    assert_eq!(*got.last().unwrap(), value(0, last));
+                                }
+                                std::hint::black_box(got);
+                            }
+                        }
+                    }
+                    (start, Instant::now())
+                })
+            })
+            .collect();
+        barrier.wait();
+        handles.into_iter().map(|h| h.join().unwrap()).collect()
+    });
+    let start = spans.iter().map(|s| s.0).min().unwrap();
+    let end = spans.iter().map(|s| s.1).max().unwrap();
+    (end - start).as_secs_f64()
+}
+
+fn median(v: &[f64]) -> f64 {
+    let mut s = v.to_vec();
+    s.sort_by(f64::total_cmp);
+    s[s.len() / 2]
+}
+
+fn hostname() -> String {
+    std::fs::read_to_string("/proc/sys/kernel/hostname")
+        .map(|s| s.trim().to_string())
+        .unwrap_or_else(|_| "unknown".into())
+}
+
+fn main() {
+    let args = match parse_args() {
+        Ok(a) => a,
+        Err(e) => {
+            eprintln!("{e}");
+            std::process::exit(2);
+        }
+    };
+    if cfg!(debug_assertions) {
+        eprintln!("warning: debug build — numbers are meaningless. Use --release.");
+    }
+    if args.decode_threads > 0 {
+        rayon::ThreadPoolBuilder::new()
+            .num_threads(args.decode_threads)
+            .build_global()
+            .expect("configure rayon pool");
+    }
+
+    let manifest = manifest_for(args.datasets, args.mib);
+    let t = Instant::now();
+    match ensure_files(&args.dir, &manifest) {
+        Ok(true) => eprintln!(
+            "generated {} in {:.1} s",
+            args.dir.display(),
+            t.elapsed().as_secs_f64()
+        ),
+        Ok(false) => eprintln!("reusing {}", args.dir.display()),
+        Err(e) => {
+            eprintln!("cannot write test files in {}: {e}", args.dir.display());
+            std::process::exit(1);
+        }
+    }
+    let path_of = |layout: &str| args.dir.join(format!("{layout}.h5"));
+    let slabs = slab_offsets(
+        args.seed,
+        args.slabs,
+        manifest.rows,
+        manifest.cols,
+        args.slab,
+    );
+    let dataset_bytes = manifest.rows * manifest.cols * 4;
+
+    let mut rows: Vec<Row> = Vec::new();
+    println!("| layout | mode | threads | MB/s | efficiency | median s |");
+    println!("|---|---|---:|---:|---:|---:|");
+    for layout in &args.layouts {
+        let path = path_of(layout);
+        // Untimed pass: page cache warm (unless --cold), results checked.
+        if !args.cold {
+            warm(&path).expect("warm page cache");
+        }
+        for mode in &args.modes {
+            run_once(&path, mode, 1, &manifest, &slabs, args.slab, true);
+            let bytes = match mode.as_str() {
+                "distinct" => dataset_bytes * manifest.datasets,
+                _ => args.slab * args.slab * 4 * args.slabs as u64,
+            };
+            let mut base: Option<f64> = None;
+            for &threads in &args.threads {
+                let times: Vec<f64> = (0..args.reps)
+                    .map(|_| {
+                        if args.cold {
+                            evict(&path).expect("posix_fadvise");
+                        }
+                        run_once(&path, mode, threads, &manifest, &slabs, args.slab, false)
+                    })
+                    .collect();
+                let med = median(&times);
+                let mb_s = bytes as f64 / (1 << 20) as f64 / med;
+                if threads == 1 {
+                    base = Some(mb_s);
+                }
+                let efficiency = base.map(|b| mb_s / (threads as f64 * b));
+                println!(
+                    "| {layout} | {mode} | {threads} | {mb_s:.0} | {} | {med:.4} |",
+                    efficiency.map_or("-".into(), |e| format!("{e:.2}"))
+                );
+                rows.push(Row {
+                    layout: layout.clone(),
+                    mode: mode.clone(),
+                    threads,
+                    bytes,
+                    times_s: times,
+                    median_s: med,
+                    mb_s,
+                    efficiency,
+                });
+            }
+        }
+    }
+
+    if let Some(out) = &args.json {
+        let doc = serde_json::json!({
+            "tool": "clawhdf5",
+            "version": env!("CARGO_PKG_VERSION"),
+            "host": hostname(),
+            "cpus": std::thread::available_parallelism().map_or(0, |n| n.get()),
+            "unix_time": std::time::SystemTime::now()
+                .duration_since(std::time::UNIX_EPOCH)
+                .map_or(0, |d| d.as_secs()),
+            "cache": if args.cold { "cold (posix_fadvise DONTNEED before each repetition)" } else { "warm" },
+            "decode_threads": rayon::current_num_threads(),
+            "params": {
+                "datasets": manifest.datasets,
+                "mib": args.mib,
+                "rows": manifest.rows,
+                "cols": manifest.cols,
+                "chunk": manifest.chunk,
+                "deflate_level": manifest.deflate_level,
+                "slab": args.slab,
+                "slabs": args.slabs,
+                "seed": args.seed,
+                "reps": args.reps,
+                "dir": args.dir,
+            },
+            "results": rows,
+        });
+        std::fs::write(out, serde_json::to_string_pretty(&doc).unwrap()).expect("write json");
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn values_are_exact_in_f32() {
+        for k in [0, 7, 63] {
+            for i in [0u64, 1, 4095, 1 << 20, (1 << 24) - 1] {
+                let v = value(k, i);
+                assert_eq!(v, (v as f64) as f32);
+                assert!(v < 32768.0);
+                assert_eq!((v * 256.0).fract(), 0.0);
+            }
+        }
+    }
+
+    /// The h5py script hard-codes this vector to check its splitmix64 port.
+    #[test]
+    fn splitmix64_reference() {
+        let mut s = 42;
+        assert_eq!(splitmix64(&mut s), 0xBDD7_3226_2FEB_6E95);
+    }
+}

crates/clawhdf5-bench/src/bin/search_harness.rs

@@ -21,6 +21,7 @@
 //! cargo run --release -p clawhdf5-bench --bin search_harness -- --ann-only --uniform
 //! cargo run --release -p clawhdf5-bench --bin search_harness -- --float16-study --full
 //! cargo run --release -p clawhdf5-bench --bin search_harness -- --options-study --full
+//! cargo run --release -p clawhdf5-bench --bin search_harness -- --signing-study --full
 //! ```
 
 use std::time::{Duration, Instant};
@@ -488,6 +489,81 @@ fn bench_end_to_end(n: usize, json: &mut Vec<serde_json::Value>) {
     }));
 }
 
+// ---------------------------------------------------------------------------
+// Signing study: what does an Ed25519-signed checkpoint cost?
+// ---------------------------------------------------------------------------
+
+/// `--signing-study`: checkpoint time unsigned vs signed, `verify` time, and
+/// the file-size cost of the stored per-record hashes. Default store
+/// settings (float16, int8 index). Medians of five checkpoints / three
+/// verifies.
+fn signing_study(n: usize) {
+    use clawhdf5_agent::signing::SigningKey;
+    let data = make_dataset(n, 0x516 ^ n as u64);
+    let mut rng = Rng(9);
+    let entries: Vec<MemoryEntry> = data
+        .vectors
+        .iter()
+        .enumerate()
+        .map(|(i, v)| MemoryEntry {
+            chunk: text_for(data.cluster_of[i], i, &mut rng),
+            embedding: v.clone(),
+            source_channel: "bench".into(),
+            timestamp: i as f64,
+            session_id: format!("s{}", i % 50),
+            tags: format!("t{i}"),
+        })
+        .collect();
+    let dir = tempfile::tempdir().unwrap();
+    let path = dir.path().join("sign.h5");
+    let mut mem = HDF5Memory::create(MemoryConfig::new(path.clone(), "bench", DIM)).unwrap();
+    mem.save_batch(entries).unwrap();
+    std::hint::black_box(mem.hybrid_search(&data.queries[0], "", 1.0, 0.0, K));
+
+    let median = |mut v: Vec<Duration>| {
+        v.sort();
+        v[v.len() / 2]
+    };
+    let checkpoint = |mem: &mut HDF5Memory| {
+        median(
+            (0..5)
+                .map(|_| {
+                    let t = Instant::now();
+                    mem.flush_wal().unwrap();
+                    t.elapsed()
+                })
+                .collect(),
+        )
+    };
+    let unsigned = checkpoint(&mut mem);
+    let unsigned_bytes = std::fs::metadata(&path).unwrap().len();
+    let key = SigningKey::from_bytes(&[7; 32]);
+    mem.set_signing_key(key.clone());
+    let signed = checkpoint(&mut mem);
+    let signed_bytes = std::fs::metadata(&path).unwrap().len();
+    drop(mem);
+    let vk = key.verifying_key();
+    let verify = median(
+        (0..3)
+            .map(|_| {
+                let t = Instant::now();
+                let r = HDF5Memory::verify(&path, &vk).unwrap();
+                let d = t.elapsed();
+                assert!(r.is_valid());
+                d
+            })
+            .collect(),
+    );
+    println!(
+        "| {n} | {:.1} | {:.1} | {:+.1} | {:.1} | {:+.2} |",
+        millis(unsigned),
+        millis(signed),
+        millis(signed) - millis(unsigned),
+        millis(verify),
+        (signed_bytes as f64 - unsigned_bytes as f64) / (1024.0 * 1024.0),
+    );
+}
+
 // ---------------------------------------------------------------------------
 // Search options study: source filters, re-ranking, confidence rejection
 // ---------------------------------------------------------------------------
@@ -960,6 +1036,21 @@ fn main() {
         }
         return;
     }
+    if args.iter().any(|a| a == "--signing-study") {
+        println!("## Signed checkpoints ({DIM}-dim, float16, int8 index)\n");
+        println!(
+            "| N | checkpoint ms, unsigned | checkpoint ms, signed | signing adds ms | verify ms | file MiB added |"
+        );
+        println!("|---:|---:|---:|---:|---:|---:|");
+        for &n in if full {
+            &[1_000, 10_000, 100_000][..]
+        } else {
+            &[1_000, 10_000][..]
+        } {
+            signing_study(n);
+        }
+        return;
+    }
     if args.iter().any(|a| a == "--options-study") {
         println!("## Search options ({DIM}-dim, k = {K}, Hebbian boost off)\n");
         println!("| N | options | filtered recall@10 | p50 ms | p99 ms |");

crates/clawhdf5-bench/tests/concurrent_read_smoke.rs

@@ -0,0 +1,148 @@
+//! Keeps the concurrent-read harnesses working: runs `concurrent_read`, the
+//! h5py script (threads and processes) and the comparison script end to end
+//! on tiny files. h5py reading the files also checks, element by element at
+//! spot positions, that both harnesses generate the same data and slabs.
+//!
+//! The h5py half is skipped when python3 with h5py is unavailable, unless
+//! `CLAWHDF5_REQUIRE_INTEROP=1`; `CLAWHDF5_PYTHON` picks the interpreter.
+
+use std::path::{Path, PathBuf};
+use std::process::Command;
+
+fn python() -> String {
+    std::env::var("CLAWHDF5_PYTHON").unwrap_or_else(|_| "python3".to_string())
+}
+
+fn interop_required() -> bool {
+    std::env::var("CLAWHDF5_REQUIRE_INTEROP").is_ok_and(|v| v == "1")
+}
+
+fn python_available() -> bool {
+    Command::new(python())
+        .args(["-c", "import h5py, numpy"])
+        .output()
+        .map(|o| o.status.success())
+        .unwrap_or(false)
+}
+
+fn scripts() -> PathBuf {
+    Path::new(env!("CARGO_MANIFEST_DIR")).join("scripts")
+}
+
+fn run(cmd: &mut Command) -> String {
+    let out = cmd.output().expect("spawn");
+    assert!(
+        out.status.success(),
+        "{cmd:?} failed\nSTDOUT:\n{}\nSTDERR:\n{}",
+        String::from_utf8_lossy(&out.stdout),
+        String::from_utf8_lossy(&out.stderr)
+    );
+    String::from_utf8_lossy(&out.stdout).into_owned()
+}
+
+const SMALL: [&str; 8] = [
+    "--threads",
+    "1,2",
+    "--slabs",
+    "8",
+    "--reps",
+    "1",
+    "--slab",
+    "64",
+];
+
+fn results(path: &Path) -> serde_json::Value {
+    serde_json::from_str(&std::fs::read_to_string(path).unwrap()).unwrap()
+}
+
+#[test]
+fn harnesses_run_end_to_end_on_tiny_files() {
+    let dir = tempfile::TempDir::new().unwrap();
+    let data = dir.path().join("data");
+    let claw = dir.path().join("claw.json");
+
+    let bin = env!("CARGO_BIN_EXE_concurrent_read");
+    run(Command::new(bin)
+        .arg("--dir")
+        .arg(&data)
+        .args(["--datasets", "3", "--mib", "1"])
+        .args(SMALL)
+        .arg("--json")
+        .arg(&claw));
+    // Second run reuses the files (and exercises --cold).
+    let out = Command::new(bin)
+        .arg("--dir")
+        .arg(&data)
+        .args(["--datasets", "3", "--mib", "1", "--cold"])
+        .args(SMALL)
+        .output()
+        .unwrap();
+    assert!(out.status.success());
+    assert!(String::from_utf8_lossy(&out.stderr).contains("reusing"));
+
+    let doc = results(&claw);
+    assert_eq!(doc["tool"], "clawhdf5");
+    // 2 layouts x 2 modes x 2 thread counts.
+    assert_eq!(doc["results"].as_array().unwrap().len(), 8);
+    for r in doc["results"].as_array().unwrap() {
+        assert!(r["mb_s"].as_f64().unwrap() > 0.0, "{r}");
+    }
+
+    if !python_available() {
+        assert!(
+            !interop_required(),
+            "CLAWHDF5_REQUIRE_INTEROP=1 but {} has no h5py",
+            python()
+        );
+        eprintln!("skipping the h5py half: no h5py in {}", python());
+        return;
+    }
+    let mut jsons = vec![claw];
+    for executor in ["threads", "processes"] {
+        let out = dir.path().join(format!("h5py-{executor}.json"));
+        run(Command::new(python())
+            .arg(scripts().join("concurrent_read_h5py.py"))
+            .arg("--dir")
+            .arg(&data)
+            .args(["--executor", executor])
+            .args(SMALL)
+            .arg("--json")
+            .arg(&out));
+        let doc = results(&out);
+        assert_eq!(doc["tool"], format!("h5py-{executor}"));
+        assert_eq!(doc["results"].as_array().unwrap().len(), 8);
+        jsons.push(out);
+    }
+    let table = run(Command::new(python())
+        .arg(scripts().join("compare_concurrent_read.py"))
+        .args(&jsons));
+    assert!(table.contains("| deflate | same | 2 |"), "{table}");
+    assert!(table.contains("clawhdf5 / h5py-processes"), "{table}");
+
+    // A different workload must not be compared.
+    let other = dir.path().join("other.json");
+    run(Command::new(python())
+        .arg(scripts().join("concurrent_read_h5py.py"))
+        .arg("--dir")
+        .arg(&data)
+        .args([
+            "--threads",
+            "1",
+            "--slabs",
+            "4",
+            "--reps",
+            "1",
+            "--slab",
+            "64",
+        ])
+        .arg("--json")
+        .arg(&other));
+    let out = Command::new(python())
+        .arg(scripts().join("compare_concurrent_read.py"))
+        .arg(&jsons[0])
+        .arg(&other)
+        .output()
+        .unwrap();
+    assert!(!out.status.success());
+    assert!(String::from_utf8_lossy(&out.stderr).contains("slabs"));
+}

crates/clawhdf5-cli/src/main.rs

@@ -1,15 +1,22 @@
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 
 use clap::{Parser, Subcommand};
+use clawhdf5_agent::signing::{self, SigningKey, VerifyingKey};
 use clawhdf5_agent::{AgentMemory, HDF5Memory, MemoryConfig, MemoryEntry};
 
 /// ClawhDF5 — HDF5-backed cognitive memory for AI agents
 #[derive(Parser)]
 #[command(name = "clawhdf5", version, about)]
 struct Cli {
-    /// Path to the .h5 memory file
+    /// Path to the .h5 memory file (not needed for `keygen`)
     #[arg(short, long, env = "CLAWHDF5_PATH")]
-    path: PathBuf,
+    path: Option<PathBuf>,
+
+    /// File holding an Ed25519 signing key (64 hex characters, from
+    /// `keygen`). Every checkpoint this command makes is then signed; a
+    /// signed store refuses to checkpoint without it.
+    #[arg(long, env = "CLAWHDF5_SIGNING_KEY", global = true)]
+    signing_key: Option<PathBuf>,
 
     #[command(subcommand)]
     command: Commands,
@@ -91,6 +98,38 @@ enum Commands {
         /// Destination path
         dest: PathBuf,
     },
+    /// Generate an Ed25519 signing key for signed checkpoints
+    Keygen {
+        /// Where to write the secret key (created new, owner-only on Unix)
+        #[arg(long)]
+        out: PathBuf,
+    },
+    /// Verify a signed store against a public key; exit status 2 if not valid
+    Verify {
+        /// The trusted public key: 64 hex characters, or a file holding them
+        #[arg(long)]
+        public_key: String,
+    },
+}
+
+fn read_signing_key(path: &Path) -> Result<SigningKey, Box<dyn std::error::Error>> {
+    let text = std::fs::read_to_string(path)
+        .map_err(|e| format!("cannot read signing key {}: {e}", path.display()))?;
+    let bytes = signing::from_hex::<32>(&text)
+        .ok_or_else(|| format!("{} is not a 64-hex-character key", path.display()))?;
+    Ok(SigningKey::from_bytes(&bytes))
+}
+
+/// Open for writing, with the signing key applied if one was given.
+fn open_writable(
+    path: &Path,
+    key: &Option<SigningKey>,
+) -> Result<HDF5Memory, Box<dyn std::error::Error>> {
+    let mut mem = HDF5Memory::open(path)?;
+    if let Some(k) = key {
+        mem.set_signing_key(k.clone());
+    }
+    Ok(mem)
 }
 
 fn main() {
@@ -103,6 +142,37 @@ fn main() {
 }
 
 fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
+    if let Commands::Keygen { out } = &cli.command {
+        let key = signing::generate_key();
+        let mut opts = std::fs::OpenOptions::new();
+        opts.write(true).create_new(true);
+        #[cfg(unix)]
+        {
+            use std::os::unix::fs::OpenOptionsExt;
+            opts.mode(0o600);
+        }
+        use std::io::Write;
+        let mut f = opts
+            .open(out)
+            .map_err(|e| format!("cannot create {}: {e}", out.display()))?;
+        writeln!(f, "{}", signing::to_hex(&key.to_bytes()))?;
+        let j = serde_json::json!({
+            "status": "generated",
+            "secret_key_file": out.display().to_string(),
+            "public_key": signing::to_hex(&key.verifying_key().to_bytes()),
+        });
+        println!("{}", serde_json::to_string_pretty(&j)?);
+        return Ok(());
+    }
+    let path = cli
+        .path
+        .clone()
+        .ok_or("--path (or CLAWHDF5_PATH) is required")?;
+    let key = cli
+        .signing_key
+        .as_deref()
+        .map(read_signing_key)
+        .transpose()?;
     match cli.command {
         Commands::Create {
             agent_id,
@@ -113,7 +183,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
             f32,
             float16: _,
         } => {
-            let mut config = MemoryConfig::new(cli.path.clone(), &agent_id, dim);
+            let mut config = MemoryConfig::new(path.clone(), &agent_id, dim);
             config.wal_enabled = wal;
             // As with --f32-index: only ever switch the library default off.
             if f32 {
@@ -127,15 +197,21 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
                 config.quantized_index = false;
             }
             let config_quantized = config.quantized_index;
-            let mem = HDF5Memory::create(config)?;
+            let mut mem = HDF5Memory::create(config)?;
+            // Sign straight away, so the store is never on disk unsigned.
+            if let Some(k) = &key {
+                mem.set_signing_key(k.clone());
+                mem.flush_wal()?;
+            }
             let j = serde_json::json!({
                 "status": "created",
-                "path": cli.path.display().to_string(),
+                "path": path.display().to_string(),
                 "agent_id": agent_id,
                 "embedding_dim": dim,
                 "wal_enabled": wal,
                 "quantized_index": config_quantized,
                 "float16": config_float16,
+                "signed": mem.is_signed(),
                 "count": mem.count(),
             });
             println!("{}", serde_json::to_string_pretty(&j)?);
@@ -152,7 +228,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
                 }
             };
             let entry: MemoryEntry = serde_json::from_str(&input)?;
-            let mut mem = HDF5Memory::open(&cli.path)?;
+            let mut mem = open_writable(&path, &key)?;
             let idx = mem.save(entry)?;
             let j = serde_json::json!({ "status": "saved", "index": idx, "count": mem.count() });
             println!("{}", serde_json::to_string(&j)?);
@@ -166,7 +242,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
             keyword_weight,
         } => {
             let emb: Vec<f32> = serde_json::from_str(&embedding)?;
-            let mut mem = HDF5Memory::open(&cli.path)?;
+            let mut mem = open_writable(&path, &key)?;
             let results = mem.hybrid_search(&emb, &query, vector_weight, keyword_weight, top_k);
             let j: Vec<serde_json::Value> = results
                 .iter()
@@ -184,7 +260,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
         }
 
         Commands::Recall { index } => {
-            let mem = HDF5Memory::open_read_only(&cli.path)?;
+            let mem = HDF5Memory::open_read_only(&path)?;
             match mem.get_chunk(index) {
                 Some(content) => {
                     let j = serde_json::json!({ "index": index, "chunk": content });
@@ -198,22 +274,23 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
         }
 
         Commands::Stats => {
-            let mem = HDF5Memory::open_read_only(&cli.path)?;
+            let mem = HDF5Memory::open_read_only(&path)?;
             let cfg = mem.config();
             let j = serde_json::json!({
-                "path": cli.path.display().to_string(),
+                "path": path.display().to_string(),
                 "agent_id": cfg.agent_id,
                 "embedding_dim": cfg.embedding_dim,
                 "count": mem.count(),
                 "active": mem.count_active(),
                 "wal_enabled": cfg.wal_enabled,
                 "wal_pending": mem.wal_pending_count(),
+                "signed": mem.is_signed(),
             });
             println!("{}", serde_json::to_string_pretty(&j)?);
         }
 
         Commands::FlushWal => {
-            let mut mem = HDF5Memory::open(&cli.path)?;
+            let mut mem = open_writable(&path, &key)?;
             let before = mem.wal_pending_count();
             mem.flush_wal()?;
             let j = serde_json::json!({
@@ -225,7 +302,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
         }
 
         Commands::AgentsMd { output } => {
-            let mem = HDF5Memory::open_read_only(&cli.path)?;
+            let mem = HDF5Memory::open_read_only(&path)?;
             let md = mem.generate_agents_md();
             match output {
                 Some(p) => {
@@ -237,7 +314,7 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
         }
 
         Commands::Export => {
-            let mem = HDF5Memory::open_read_only(&cli.path)?;
+            let mem = HDF5Memory::open_read_only(&path)?;
             for i in 0..mem.count() {
                 if let Some(chunk) = mem.get_chunk(i) {
                     let j = serde_json::json!({ "index": i, "chunk": chunk });
@@ -246,11 +323,44 @@ fn run(cli: Cli) -> Result<(), Box<dyn std::error::Error>> {
             }
         }
 
+        Commands::Keygen { .. } => unreachable!("handled before opening a store"),
+
+        Commands::Verify { public_key } => {
+            let text = if Path::new(&public_key).is_file() {
+                std::fs::read_to_string(&public_key)?
+            } else {
+                public_key
+            };
+            let bytes = signing::from_hex::<32>(&text)
+                .ok_or("--public-key must be 64 hex characters or a file holding them")?;
+            let trusted = VerifyingKey::from_bytes(&bytes)?;
+            let r = HDF5Memory::verify(&path, &trusted)?;
+            let j = serde_json::json!({
+                "valid": r.is_valid(),
+                "signed": r.signed,
+                "key_matches": r.key_matches,
+                "signature_valid": r.signature_valid,
+                "records_match": r.records_match,
+                "settings_match": r.settings_match,
+                "sessions_match": r.sessions_match,
+                "graph_match": r.graph_match,
+                "changed_records": r.changed_records,
+                "record_count": r.record_count,
+                "signed_record_count": r.signed_record_count,
+                "signed_by": r.public_key.map(|k| signing::to_hex(&k)),
+                "wal_entries_unsigned": r.wal_entries_unsigned,
+            });
+            println!("{}", serde_json::to_string_pretty(&j)?);
+            if !r.is_valid() {
+                std::process::exit(2);
+            }
+        }
+
         Commands::Snapshot { dest } => {
-            let _result = clawhdf5_agent::storage::snapshot_file(&cli.path, &dest)?;
+            let _result = clawhdf5_agent::storage::snapshot_file(&path, &dest)?;
             let j = serde_json::json!({
                 "status": "snapshot_created",
-                "source": cli.path.display().to_string(),
+                "source": path.display().to_string(),
                 "dest": dest.display().to_string(),
             });
             println!("{}", serde_json::to_string(&j)?);

crates/clawhdf5-format/Cargo.toml

@@ -22,6 +22,17 @@ zstd = { version = "0.13", optional = true }
 blake3 = { version = "1", optional = true }
 libaec-sys = { path = "../libaec-sys", version = "0.1", optional = true }
 pco = { version = "1.0", optional = true }
+# Pure-Rust Zstandard, for the plugin filters that embed zstd (bitshuffle,
+# blosc). The `zstd` feature (filter 32015) links libzstd instead.
+ruzstd = { version = "0.9", optional = true }
+# bzip2 with its default backend, libbz2-rs-sys: a pure-Rust port of
+# libbzip2 (no C is compiled, despite the -sys name).
+bzip2 = { version = "0.6", optional = true }
+snap = { version = "1", optional = true }
+
+[target.'cfg(target_os = "linux")'.dependencies]
+# madvise(MADV_HUGEPAGE) for large read buffers (see src/bulk_alloc.rs).
+libc = { version = "0.2", default-features = false }
 
 [dev-dependencies]
 half = { workspace = true }
@@ -37,7 +48,7 @@ harness = false
 # Deflate backend: `zlib-rs` (pure Rust) by default. `fast-deflate` selects
 # zlib-ng instead (C, built with cmake); flate2 prefers a C zlib whenever one
 # is enabled, so turning it on anywhere in the build overrides the default.
-default = ["std", "checksum", "deflate", "provenance", "zlib-rs", "system-zlib-decompress"]
+default = ["std", "checksum", "deflate", "provenance", "zlib-rs", "system-zlib-decompress", "lzf"]
 std = []
 checksum = []
 deflate = ["flate2"]
@@ -56,6 +67,17 @@ zstd = ["dep:zstd"]
 blake3_hash = ["blake3"]
 szip = ["libaec-sys"]
 pcodec = ["dep:pco"]
+# Plugin filters, pure Rust. LZF (32000) is h5py's built-in compression; it
+# has no dependencies, so it is on by default.
+lzf = []
+# Bitshuffle (32008), with its LZ4 and Zstandard modes.
+bitshuffle = ["lz4_flex", "ruzstd"]
+# bzip2 (307).
+bzip2 = ["dep:bzip2", "std"]
+# Blosc 1 (32001) with its BloscLZ, LZ4, Snappy, Zlib and Zstandard codecs.
+blosc = ["lz4_flex", "ruzstd", "snap", "deflate", "std"]
+# Every plugin filter above.
+plugin-filters = ["lzf", "bitshuffle", "bzip2", "blosc"]
 
 [[bench]]
 name = "parallel_decompress_bench"

crates/clawhdf5-format/src/attribute.rs

@@ -97,7 +97,7 @@ impl AttributeMessage {
             return Ok(Cow::Borrowed(bytes));
         }
         let (file_data, offset_size) = file.ok_or(FormatError::UnresolvedSharedMessage)?;
-        let shared_ref = shared_message::parse_shared_ref(bytes, offset_size)?;
+        let shared_ref = shared_message::parse_shared_ref_sized(bytes, offset_size, length_size)?;
         shared_message::resolve_shared_message(
             file_data,
             &shared_ref,
@@ -362,6 +362,18 @@ fn extract_name(bytes: &[u8]) -> String {
     String::from_utf8_lossy(&bytes[..end]).into_owned()
 }
 
+/// An attribute's datatype gets libhdf5's extra check for a header without
+/// a checksum (see [`Datatype::check_unused_bits`]).
+fn check_in_header(
+    attr: AttributeMessage,
+    header: &ObjectHeader,
+) -> Result<AttributeMessage, FormatError> {
+    if header.version == 1 {
+        attr.datatype.check_unused_bits()?;
+    }
+    Ok(attr)
+}
+
 /// Extract all attribute messages from an object header.
 pub fn extract_attributes(
     header: &ObjectHeader,
@@ -371,7 +383,7 @@ pub fn extract_attributes(
     for msg in &header.messages {
         if msg.msg_type == MessageType::Attribute {
             let attr = AttributeMessage::parse(&msg.data, length_size)?;
-            attrs.push(attr);
+            attrs.push(check_in_header(attr, header)?);
         }
     }
     Ok(attrs)
@@ -394,42 +406,81 @@ pub fn find_attribute<'a>(
 ///
 /// Use this instead of `extract_attributes` when reading files that may use dense storage
 /// (e.g., objects with many attributes, typically >8).
+///
+/// Fails if any attribute cannot be read; see [`extract_attributes_tolerant`]
+/// to read the others.
 pub fn extract_attributes_full(
     file_data: &[u8],
     header: &ObjectHeader,
     offset_size: u8,
     length_size: u8,
+) -> Result<Vec<AttributeMessage>, FormatError> {
+    extract_attributes_with(file_data, header, offset_size, length_size, &mut Err)
+}
+
+/// Like [`extract_attributes_full`], but an attribute that cannot be read
+/// (a corrupt or unsupported attribute message, or a heap object that cannot
+/// be located) is left out and its error returned alongside the attributes
+/// that could be read, instead of failing them all.
+///
+/// Errors in the structures that index the attributes (the Attribute Info
+/// message, the dense-storage heap header or B-tree) still fail the call:
+/// then it is unknown which attributes exist at all.
+pub fn extract_attributes_tolerant(
+    file_data: &[u8],
+    header: &ObjectHeader,
+    offset_size: u8,
+    length_size: u8,
+) -> Result<(Vec<AttributeMessage>, Vec<FormatError>), FormatError> {
+    let mut errors = Vec::new();
+    let attrs = extract_attributes_with(file_data, header, offset_size, length_size, &mut |e| {
+        errors.push(e);
+        Ok(())
+    })?;
+    Ok((attrs, errors))
+}
+
+/// Read every attribute; each one that fails goes to `on_error`, which
+/// either stops the read (returns the error) or skips that attribute.
+fn extract_attributes_with(
+    file_data: &[u8],
+    header: &ObjectHeader,
+    offset_size: u8,
+    length_size: u8,
+    on_error: &mut dyn FnMut(FormatError) -> Result<(), FormatError>,
 ) -> Result<Vec<AttributeMessage>, FormatError> {
     let mut attrs = Vec::new();
 
     // Collect compact attributes (inline in OH)
     for msg in &header.messages {
         if msg.msg_type == MessageType::Attribute {
-            if shared_message::is_shared(msg.flags) {
+            let attr = if shared_message::is_shared(msg.flags) {
                 // Shared attribute: resolve the reference to get actual attribute data
-                let shared_ref = shared_message::parse_shared_ref(&msg.data, offset_size)?;
-                let resolved_data = shared_message::resolve_shared_message(
+                shared_message::parse_shared_ref_sized(&msg.data, offset_size, length_size)
+                    .and_then(|shared_ref| {
+                        shared_message::resolve_shared_message(
                             file_data,
                             &shared_ref,
                             MessageType::Attribute,
                             offset_size,
                             length_size,
-                )?;
-                let attr = AttributeMessage::parse_in_file(
-                    &resolved_data,
+                        )
+                    })
+                    .and_then(|resolved| {
+                        AttributeMessage::parse_in_file(
+                            &resolved,
                             file_data,
                             offset_size,
                             length_size,
-                )?;
-                attrs.push(attr);
+                        )
+                    })
             } else {
-                let attr = AttributeMessage::parse_in_file(
-                    &msg.data,
-                    file_data,
-                    offset_size,
-                    length_size,
-                )?;
-                attrs.push(attr);
+                AttributeMessage::parse_in_file(&msg.data, file_data, offset_size, length_size)
+            };
+            let attr = attr.and_then(|a| check_in_header(a, header));
+            match attr {
+                Ok(attr) => attrs.push(attr),
+                Err(e) => on_error(e)?,
             }
         }
     }
@@ -439,9 +490,15 @@ pub fn extract_attributes_full(
     if let Some(info) = attr_info
         && let Some(fh_addr) = info.fractal_heap_address
     {
-        let dense_attrs =
-            extract_dense_attributes(file_data, &info, fh_addr, offset_size, length_size)?;
-        attrs.extend(dense_attrs);
+        extract_dense_attributes(
+            file_data,
+            &info,
+            fh_addr,
+            offset_size,
+            length_size,
+            &mut attrs,
+            on_error,
+        )?;
     }
 
     Ok(attrs)
@@ -468,7 +525,9 @@ fn extract_dense_attributes(
     fh_addr: u64,
     offset_size: u8,
     length_size: u8,
-) -> Result<Vec<AttributeMessage>, FormatError> {
+    attrs: &mut Vec<AttributeMessage>,
+    on_error: &mut dyn FnMut(FormatError) -> Result<(), FormatError>,
+) -> Result<(), FormatError> {
     // Parse fractal heap
     let fh = FractalHeapHeader::parse(file_data, fh_addr as usize, offset_size, length_size)?;
 
@@ -482,28 +541,32 @@ fn extract_dense_attributes(
     let btree_hdr = BTreeV2Header::parse(file_data, btree_addr as usize, offset_size, length_size)?;
     let records = collect_btree_v2_records(file_data, &btree_hdr, offset_size, length_size)?;
 
-    let mut attrs = Vec::new();
     for record in &records {
         // Per HDF5 spec, both type 8 and type 9 records start with heap_id:
         //   Type 8: heap_id(8) + msg_flags(1) + creation_order(4) + hash(4)
         //   Type 9: heap_id(8) + msg_flags(1) + creation_order(4)
-        let id_offset = 0;
-
-        if record.data.len() < id_offset + fh.heap_id_length as usize {
+        let id_len = fh.heap_id_length as usize;
+        let Some(id_bytes) = record.data.get(..id_len) else {
+            on_error(FormatError::UnexpectedEof {
+                expected: id_len,
+                available: record.data.len(),
+            })?;
             continue;
-        }
-        let id_bytes = &record.data[id_offset..id_offset + fh.heap_id_length as usize];
-
-        // Read attribute message from fractal heap
-        let attr_data = fh.read_managed_object(file_data, id_bytes, offset_size)?;
+        };
 
         // The data in the heap is a complete attribute message
-        let attr =
-            AttributeMessage::parse_in_file(&attr_data, file_data, offset_size, length_size)?;
-        attrs.push(attr);
+        let attr = fh
+            .read_managed_object(file_data, id_bytes, offset_size)
+            .and_then(|attr_data| {
+                AttributeMessage::parse_in_file(&attr_data, file_data, offset_size, length_size)
+            });
+        match attr {
+            Ok(attr) => attrs.push(attr),
+            Err(e) => on_error(e)?,
+        }
     }
 
-    Ok(attrs)
+    Ok(())
 }
 
 #[cfg(test)]
@@ -523,7 +586,8 @@ mod tests {
 
     /// Build an f64 LE datatype message.
     fn build_f64_dt() -> Vec<u8> {
-        let mut buf = build_dt_header(1, 1, [0x00, 0x00, 0x02], 8);
+        // Sign bit 63 (bits 8-15 of the class bits).
+        let mut buf = build_dt_header(1, 1, [0x20, 63, 0x00], 8);
         let mut props = [0u8; 12];
         props[2..4].copy_from_slice(&64u16.to_le_bytes()); // bit_precision
         props[4] = 52; // exp_location

crates/clawhdf5-format/src/btree_v2.rs

@@ -323,39 +323,21 @@ fn collect_internal_records(
     let records_start = pos;
     pos += records_total;
 
-    // Compute sizes for child pointers
-    // max_records at child depth - for variable-width nrec encoding
+    // Child pointer layout, as libhdf5 computes it (H5B2__hdr_init): the
+    // child's record count is always encoded in the width needed for a
+    // *leaf's* maximum, and — below the first internal level — the child
+    // subtree's total record count in the width needed for the most records
+    // a subtree of that depth can hold.
     let child_depth = depth - 1;
-    let max_nrec_child = if child_depth == 0 {
-        max_leaf_nrec
-    } else {
-        // For internal nodes at child_depth, the true max_nrec depends on the
-        // node size, record size, and the recursive width of child pointer
-        // entries (which themselves depend on max_nrec at deeper levels).
-        // Computing the exact value requires iterating from the leaf level
-        // upward, as described in the HDF5 spec (III.A.2 "Computing the Size
-        // of B-tree Nodes").
-        //
-        // We use `max_leaf_nrec * 2` as a conservative upper bound. This
-        // over-estimates the nrec encoding width, which means we may read
-        // slightly more bytes per child pointer than strictly necessary, but
-        // never fewer. The over-read bytes are harmless because we only
-        // decode `num_records` entries (the actual count from the node header).
-        //
-        // Known limitation: for very deep trees (depth > 3) with small record
-        // sizes, the true max could exceed this estimate, causing us to
-        // under-allocate the nrec encoding width and misparse child pointers.
-        // In practice, HDF5 B-tree v2 depths rarely exceed 2-3.
-        max_leaf_nrec * 2
-    };
-    let nrec_width = bytes_for_max_records(max_nrec_child);
-
-    // Total records in subtree width (only if depth > 1)
+    let nrec_width = bytes_for_max_records(max_leaf_nrec);
     let total_nrec_width = if depth > 1 {
-        // Width to hold total records in a subtree
-        // We compute max possible total records at this subtree depth
-        let max_total = header_max_total_records(max_leaf_nrec, depth - 1);
-        bytes_for_max_records(max_total)
+        bytes_for_max_records(cum_max_records(
+            node_size,
+            record_size,
+            offset_size,
+            max_leaf_nrec,
+            child_depth,
+        ))
     } else {
         0
     };
@@ -435,14 +417,36 @@ fn collect_internal_records(
     Ok(())
 }
 
-/// Estimate maximum total records at a given depth (for variable-width encoding).
-fn header_max_total_records(max_leaf_nrec: u64, depth: u16) -> u64 {
-    // Conservative: branching factor * max_leaf at each level
-    let mut total = max_leaf_nrec;
-    for _ in 0..depth {
-        total = total.saturating_mul(max_leaf_nrec.max(2));
+/// Most records a subtree whose root is at `depth` can hold (libhdf5's
+/// `cum_max_nrec`): a leaf holds `max_leaf_nrec`; an internal node at depth
+/// `d` holds `max_nrec(d)` records and `max_nrec(d) + 1` subtrees of depth
+/// `d - 1`, where `max_nrec(d)` is what fits in a node once each record is
+/// paired with a child pointer of the width depth `d` needs.
+fn cum_max_records(
+    node_size: u32,
+    record_size: u16,
+    offset_size: u8,
+    max_leaf_nrec: u64,
+    depth: u16,
+) -> u64 {
+    // Internal node overhead: signature(4) + version(1) + type(1) + checksum(4).
+    const PREFIX: u64 = 10;
+    let nrec_width = bytes_for_max_records(max_leaf_nrec) as u64;
+    let mut cum = max_leaf_nrec;
+    let mut cum_width = 0u64;
+    for d in 1..=depth {
+        let ptr = u64::from(offset_size) + nrec_width + if d > 1 { cum_width } else { 0 };
+        let max_nrec = u64::from(node_size)
+            .saturating_sub(PREFIX)
+            .saturating_sub(ptr)
+            / (u64::from(record_size) + ptr).max(1);
+        cum = max_nrec
+            .saturating_add(1)
+            .saturating_mul(cum)
+            .saturating_add(max_nrec);
+        cum_width = bytes_for_max_records(cum) as u64;
     }
-    total
+    cum
 }
 
 #[cfg(test)]
@@ -512,9 +516,15 @@ mod tests {
         child_nrec: u64,
     ) -> Vec<u8> {
         let max_leaf = max_records_leaf(node_size, record_size);
-        let nrec_width = bytes_for_max_records(if depth == 1 { max_leaf } else { max_leaf * 2 });
+        let nrec_width = bytes_for_max_records(max_leaf);
         let total_width = if depth > 1 {
-            bytes_for_max_records(header_max_total_records(max_leaf, depth - 1))
+            bytes_for_max_records(cum_max_records(
+                node_size,
+                record_size,
+                8,
+                max_leaf,
+                depth - 1,
+            ))
         } else {
             0
         };
@@ -673,4 +683,18 @@ mod tests {
         let records = collect_btree_v2_records(&header, &hdr, 8, 8).unwrap();
         assert!(records.is_empty());
     }
+
+    #[test]
+    fn subtree_capacity_matches_libhdf5() {
+        // A link-name index (11-byte records, 512-byte nodes, 8-byte
+        // addresses): libhdf5's H5B2__hdr_init gives 45 records per leaf,
+        // then cum_max_nrec 1 149 at depth 1 and 26 449 at depth 2 — two
+        // bytes of subtree count in a depth-3 root's child pointers, where
+        // leaf_max^3 = 91 125 would need three.
+        let leaf = max_records_leaf(512, 11);
+        assert_eq!(leaf, 45);
+        assert_eq!(cum_max_records(512, 11, 8, leaf, 0), 45);
+        assert_eq!(cum_max_records(512, 11, 8, leaf, 1), 1_149);
+        assert_eq!(cum_max_records(512, 11, 8, leaf, 2), 26_449);
+    }
 }

crates/clawhdf5-format/src/bulk_alloc.rs

@@ -0,0 +1,79 @@
+//! Large output buffers backed by transparent huge pages where the OS offers
+//! them.
+//!
+//! A fresh multi-megabyte `Vec` is mapped lazily by the kernel: the first
+//! write to each 4 KiB page takes a page fault, and the kernel zeroes the page
+//! before handing it over. For a 64 MiB read that is 16384 faults, and they
+//! cost far more than the copy that fills the buffer — single-threaded
+//! contiguous reads ran at about a quarter of h5py's speed because of them.
+//! numpy (so h5py) avoids this by asking for transparent huge pages
+//! (`madvise(MADV_HUGEPAGE)`) on every allocation of 4 MiB or more, which
+//! turns 512 faults into one; this module does the same.
+//!
+//! The advice only changes how the pages are backed, never their contents, so
+//! it is harmless when it cannot be honoured (THP disabled, not Linux, a
+//! region that is part of the heap): the buffer is then exactly what it would
+//! have been without it.
+
+#[cfg(not(feature = "std"))]
+use alloc::vec::Vec;
+
+/// Buffers smaller than this are left alone (numpy uses the same threshold).
+#[cfg(any(target_os = "linux", test))]
+pub(crate) const HUGE_PAGE_THRESHOLD: usize = 4 << 20;
+
+/// Advise the kernel to back `[ptr, ptr + len)` with transparent huge pages,
+/// when `len` is large enough to benefit. Call it before the first write so
+/// the faults happen at huge-page granularity.
+#[inline]
+pub(crate) fn advise_huge_pages(ptr: *const u8, len: usize) {
+    #[cfg(target_os = "linux")]
+    if len >= HUGE_PAGE_THRESHOLD {
+        const PAGE: usize = 4096;
+        let start = (ptr as usize).next_multiple_of(PAGE);
+        let end = (ptr as usize + len) & !(PAGE - 1);
+        if end > start {
+            // SAFETY: `[start, end)` lies inside an allocation of `len` bytes
+            // at `ptr` that the caller owns, and is page aligned as madvise
+            // requires. MADV_HUGEPAGE does not change the memory's contents or
+            // validity; on failure (EINVAL when THP is compiled out, etc.) the
+            // region is simply left as it was, so the result is ignored.
+            unsafe {
+                libc::madvise(start as *mut libc::c_void, end - start, libc::MADV_HUGEPAGE);
+            }
+        }
+    }
+    #[cfg(not(target_os = "linux"))]
+    let _ = (ptr, len);
+}
+
+/// `Vec::with_capacity(count)` for a buffer about to be filled in bulk, with
+/// huge-page advice when it is large (see the module docs).
+#[inline]
+pub(crate) fn vec_for_bulk<T>(count: usize) -> Vec<T> {
+    let v: Vec<T> = Vec::with_capacity(count);
+    advise_huge_pages(
+        v.as_ptr().cast::<u8>(),
+        v.capacity().saturating_mul(core::mem::size_of::<T>()),
+    );
+    v
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn bulk_vec_is_an_ordinary_vec() {
+        for count in [0usize, 1, 1000, HUGE_PAGE_THRESHOLD / 4 + 3] {
+            let mut v: Vec<u32> = vec_for_bulk(count);
+            assert!(v.capacity() >= count);
+            v.extend((0..count as u32).map(|i| i.wrapping_mul(2654435761)));
+            assert!(
+                v.iter()
+                    .enumerate()
+                    .all(|(i, &x)| x == (i as u32).wrapping_mul(2654435761))
+            );
+        }
+    }
+}

crates/clawhdf5-format/src/chunk_cache.rs

@@ -223,13 +223,32 @@ pub const DEFAULT_CACHE_BYTES: usize = 16 * 1024 * 1024; // 16 MiB
 /// coordinate map and reduces collision chains compared to power-of-two sizes.
 pub const DEFAULT_MAX_SLOTS: usize = 521;
 
+/// Most datasets whose chunk index a [`ChunkCache`] keeps at once.
+pub const MAX_INDEXED_DATASETS: usize = 64;
+
+/// Most chunk-index entries, summed over all datasets, a [`ChunkCache`] keeps.
+/// Least-recently-used datasets' indexes are dropped past this (the dataset
+/// being read is always kept), so a file with many or huge chunked datasets
+/// cannot grow the cache without bound.
+pub const MAX_INDEXED_CHUNKS: usize = 1 << 20;
+
+/// The dataset key the address-less (legacy) methods use when
+/// [`ChunkCache::ensure_dataset`] has not been called.
+#[cfg(feature = "std")]
+const UNBOUND_DATASET: u64 = u64::MAX;
+
 // ---------------------------------------------------------------------------
 // LRU entry
 // ---------------------------------------------------------------------------
 
+/// Decompressed chunks are keyed by dataset *and* coordinate: every chunked
+/// dataset has a chunk at (0, 0, ...), so the coordinate alone is ambiguous.
+#[cfg(feature = "std")]
+type SlotKey = (u64, ChunkCoord);
+
 #[cfg(feature = "std")]
 struct CachedChunk {
-    coord: ChunkCoord,
+    key: SlotKey,
     /// Shared so a cache hit is a refcount bump, not a copy of the whole
     /// (potentially large) decompressed chunk.
     data: Arc<CacheAlignedBuffer>,
@@ -237,21 +256,48 @@ struct CachedChunk {
     last_access: u64,
 }
 
+/// Per-dataset index state.
+#[cfg(feature = "std")]
+#[derive(Default)]
+struct DatasetEntry {
+    /// Chunk coordinate -> ChunkInfo (offset + size in file).
+    index: Option<Arc<HashMap<ChunkCoord, ChunkInfo>>>,
+    /// Pre-built chunk index for O(1) coordinate lookups.
+    chunk_index: Option<Arc<ChunkIndex>>,
+    /// Pre-computed chunk layout for fast assembly.
+    chunk_layout: Option<Arc<ChunkLayout>>,
+    /// Tick of the last use, for dropping the least recently used dataset.
+    last_used: u64,
+}
+
+#[cfg(feature = "std")]
+impl DatasetEntry {
+    fn weight(&self) -> usize {
+        self.index.as_ref().map_or(0, |m| m.len())
+            + self.chunk_index.as_ref().map_or(0, |c| c.num_chunks())
+    }
+}
+
 // ---------------------------------------------------------------------------
 // ChunkCache
 // ---------------------------------------------------------------------------
 
-/// A per-dataset chunk cache with hash-based index and LRU eviction.
+/// A per-file chunk cache: chunk indexes per dataset, plus an LRU of
+/// decompressed chunks, all keyed by dataset.
 ///
-/// # Usage
+/// A dataset is identified by the address of its chunk index (B-tree, fixed
+/// or extensible array, ...), which is unique within a file. Every method
+/// that takes an `addr` works on that dataset only, so threads reading
+/// different datasets through one shared cache never see each other's
+/// chunks. The address-less methods (`has_index`, `populate_index`,
+/// `get_decompressed`, ...) act on the dataset last bound with
+/// [`Self::ensure_dataset`]; that binding is shared state, so concurrent
+/// readers must use the `*_in` / `*_for` methods instead (the chunked
+/// readers in [`crate::chunked_read`] do).
 ///
-/// ```ignore
-/// let cache = ChunkCache::new();
-/// // Pass &cache to read_chunked_data — it will populate the index lazily.
-/// ```
-///
-/// The cache is wrapped in `Mutex` internally so it can be mutated through
-/// shared references (thread-safe).
+/// Memory is bounded: decompressed data by `max_bytes`/`max_slots` across
+/// all datasets, indexes by [`MAX_INDEXED_DATASETS`] and
+/// [`MAX_INDEXED_CHUNKS`].
 ///
 /// Only available with the `std` feature because it requires `std::sync::Mutex`.
 #[cfg(feature = "std")]
@@ -261,26 +307,20 @@ pub struct ChunkCache {
 
 #[cfg(feature = "std")]
 struct CacheInner {
-    /// Hash index: chunk coordinate -> ChunkInfo (offset + size in file).
-    /// Populated once per dataset on first access.
-    index: Option<HashMap<ChunkCoord, ChunkInfo>>,
+    /// Per-dataset chunk indexes, keyed by chunk-index address.
+    datasets: HashMap<u64, DatasetEntry>,
 
-    /// Address of the dataset (its chunk-index base address) that the cached
-    /// index, chunk index, layout, and decompressed slots currently belong to.
-    /// The cache is shared per file across datasets, so every cached-read entry
-    /// checks this and resets the per-dataset state when the dataset changes —
-    /// otherwise one dataset's chunk index (with its own rank) would be reused
-    /// for another, corrupting reads.
-    index_addr: Option<u64>,
+    /// Dataset the address-less methods act on (see `ensure_dataset`).
+    current: Option<u64>,
 
     /// LRU cache of decompressed chunk data.
     slots: Vec<CachedChunk>,
 
-    /// Coordinate -> index into `slots`, for O(1) lookup instead of a linear
+    /// Key -> index into `slots`, for O(1) lookup instead of a linear
     /// scan. Kept in sync with `slots` on every insert/evict/clear — in
     /// particular, `slots.swap_remove(i)` moves the last element into slot
     /// `i`, so the moved element's index entry must be updated too.
-    slot_index: HashMap<ChunkCoord, usize>,
+    slot_index: HashMap<SlotKey, usize>,
 
     /// Current total bytes of cached decompressed data.
     current_bytes: usize,
@@ -294,17 +334,145 @@ struct CacheInner {
     /// Monotonic counter for LRU ordering.
     tick: u64,
 
-    /// Last accessed chunk coordinate (for sequential detection).
-    last_coord: Option<ChunkCoord>,
+    /// Last accessed chunk (for sequential detection).
+    last_coord: Option<SlotKey>,
 
     /// Access pattern statistics.
     stats: AccessStats,
+}
 
-    /// Pre-built chunk index for O(1) coordinate lookups.
-    chunk_index: Option<ChunkIndex>,
+#[cfg(feature = "std")]
+impl CacheInner {
+    fn current(&self) -> u64 {
+        self.current.unwrap_or(UNBOUND_DATASET)
+    }
 
-    /// Pre-computed chunk layout for fast assembly.
-    chunk_layout: Option<ChunkLayout>,
+    fn touch(&mut self, addr: u64) -> &mut DatasetEntry {
+        self.tick += 1;
+        let tick = self.tick;
+        let entry = self.datasets.entry(addr).or_default();
+        entry.last_used = tick;
+        entry
+    }
+
+    fn entry(&self, addr: u64) -> Option<&DatasetEntry> {
+        self.datasets.get(&addr)
+    }
+
+    /// Drop least-recently-used datasets' indexes (never `keep`'s) until the
+    /// dataset and chunk-entry budgets hold.
+    fn trim_datasets(&mut self, keep: u64) {
+        loop {
+            let total: usize = self.datasets.values().map(DatasetEntry::weight).sum();
+            if self.datasets.len() <= MAX_INDEXED_DATASETS && total <= MAX_INDEXED_CHUNKS {
+                return;
+            }
+            let victim = self
+                .datasets
+                .iter()
+                .filter(|(a, _)| **a != keep)
+                .min_by_key(|(_, e)| e.last_used)
+                .map(|(a, _)| *a);
+            match victim {
+                Some(a) => {
+                    self.datasets.remove(&a);
+                }
+                None => return,
+            }
+        }
+    }
+
+    fn get_decompressed(&mut self, addr: u64, coord: &[u64]) -> Option<Arc<CacheAlignedBuffer>> {
+        self.tick += 1;
+        let tick = self.tick;
+
+        // Track sequential vs random access
+        let is_sequential = self.last_coord.as_ref().is_some_and(|(prev_addr, prev)| {
+            // Sequential if exactly one dimension changed
+            let changes: usize = prev
+                .iter()
+                .zip(coord.iter())
+                .filter(|(a, b)| a != b)
+                .count();
+            *prev_addr == addr && changes <= 1
+        });
+        if is_sequential {
+            self.stats.sequential_count += 1;
+        } else if self.last_coord.is_some() {
+            self.stats.random_count += 1;
+        }
+        let key: SlotKey = (addr, coord.to_vec());
+        let found = if let Some(&idx) = self.slot_index.get(&key) {
+            self.slots[idx].last_access = tick;
+            Some(Arc::clone(&self.slots[idx].data))
+        } else {
+            None
+        };
+        self.last_coord = Some(key);
+        if let Some(ref data) = found {
+            self.stats.hits += 1;
+            self.stats.bytes_read += data.len() as u64;
+        } else {
+            self.stats.misses += 1;
+        }
+        found
+    }
+
+    fn put_decompressed(
+        &mut self,
+        key: SlotKey,
+        data: Arc<CacheAlignedBuffer>,
+    ) -> Arc<CacheAlignedBuffer> {
+        let data_len = data.len();
+
+        // Don't cache if single chunk exceeds budget — still return the data
+        // to the caller, just don't retain it.
+        if data_len > self.max_bytes {
+            return data;
+        }
+
+        // Check if already present
+        self.tick += 1;
+        let tick = self.tick;
+        if let Some(&idx) = self.slot_index.get(&key) {
+            self.slots[idx].last_access = tick;
+            return Arc::clone(&self.slots[idx].data); // already cached
+        }
+
+        // Evict until we have room
+        while self.slots.len() >= self.max_slots
+            || (self.current_bytes + data_len > self.max_bytes && !self.slots.is_empty())
+        {
+            // Find LRU slot
+            let lru_idx = self
+                .slots
+                .iter()
+                .enumerate()
+                .min_by_key(|(_, s)| s.last_access)
+                .map(|(i, _)| i)
+                .unwrap();
+            let removed = self.slots.swap_remove(lru_idx);
+            self.slot_index.remove(&removed.key);
+            // swap_remove moved the former last element into `lru_idx` (unless
+            // it *was* the last element) — fix up that element's index entry.
+            if lru_idx < self.slots.len() {
+                let moved_key = self.slots[lru_idx].key.clone();
+                self.slot_index.insert(moved_key, lru_idx);
+            }
+            self.current_bytes -= removed.data.len();
+            self.stats.evictions += 1;
+        }
+
+        self.current_bytes += data_len;
+        let new_idx = self.slots.len();
+        self.slot_index.insert(key.clone(), new_idx);
+        self.slots.push(CachedChunk {
+            key,
+            data: Arc::clone(&data),
+            last_access: tick,
+        });
+        data
+    }
 }
 
 /// Access pattern statistics tracked by the chunk cache.
@@ -356,8 +524,8 @@ impl ChunkCache {
     pub fn with_capacity(max_bytes: usize, max_slots: usize) -> Self {
         Self {
             inner: std::sync::Mutex::new(CacheInner {
-                index: None,
-                index_addr: None,
+                datasets: HashMap::new(),
+                current: None,
                 slots: Vec::with_capacity(max_slots.min(64)),
                 slot_index: HashMap::with_capacity(max_slots.min(64)),
                 current_bytes: 0,
@@ -366,340 +534,331 @@ impl ChunkCache {
                 tick: 0,
                 last_coord: None,
                 stats: AccessStats::default(),
-                chunk_index: None,
-                chunk_layout: None,
             }),
         }
     }
 
-    // ----- Index operations -----
+    fn lock(&self) -> std::sync::MutexGuard<'_, CacheInner> {
+        self.inner.lock().unwrap_or_else(|e| e.into_inner())
+    }
 
     /// The most decompressed bytes this cache will hold.
     pub fn max_bytes(&self) -> usize {
-        self.inner.lock().map(|g| g.max_bytes).unwrap_or(0)
+        self.lock().max_bytes
     }
 
-    /// Bind the cache to the dataset at chunk-index address `addr`.
+    // ----- Dataset-keyed operations (safe to use concurrently) -----
+
+    /// The chunk list of the dataset whose chunk index is at `addr`.
     ///
-    /// The cache is shared per file across all of its datasets. If the cache
-    /// currently holds state for a different dataset, all per-dataset state
-    /// (chunk index, chunk-index map, layout, and decompressed slots) is
-    /// dropped so the next access rebuilds it for this dataset. Reading the
-    /// same dataset again is a no-op, preserving the cache's benefit for
-    /// repeated/sequential access. Returns `true` if a reset occurred.
+    /// On the first call for a dataset, `build` scans its chunk index; the
+    /// result is kept (offsets truncated to `rank` for the lookup key), so
+    /// later calls skip the scan. `build` runs without the cache lock held;
+    /// if two threads race to build the same dataset's index, the first
+    /// stored one wins and both return equivalent lists.
+    pub fn chunks_for<E>(
+        &self,
+        addr: u64,
+        rank: usize,
+        build: impl FnOnce() -> Result<Vec<ChunkInfo>, E>,
+    ) -> Result<Vec<ChunkInfo>, E> {
+        Ok(self
+            .index_for(addr, rank, build)?
+            .values()
+            .cloned()
+            .collect())
+    }
+
+    fn index_for<E>(
+        &self,
+        addr: u64,
+        rank: usize,
+        build: impl FnOnce() -> Result<Vec<ChunkInfo>, E>,
+    ) -> Result<Arc<HashMap<ChunkCoord, ChunkInfo>>, E> {
+        if let Some(index) = self.lock().touch(addr).index.clone() {
+            return Ok(index);
+        }
+        let chunks = build()?;
+        let map: HashMap<ChunkCoord, ChunkInfo> = chunks
+            .into_iter()
+            .map(|ci| (ci.offsets.iter().take(rank).copied().collect(), ci))
+            .collect();
+        let mut inner = self.lock();
+        let entry = inner.touch(addr);
+        let index = Arc::clone(entry.index.get_or_insert_with(|| Arc::new(map)));
+        inner.trim_datasets(addr);
+        Ok(index)
+    }
+
+    /// The pre-computed assembly layout of the dataset at `addr`, building
+    /// its chunk index (via `build`, as in [`Self::chunks_for`]) and layout on
+    /// first use.
+    pub fn chunk_layout_for<E>(
+        &self,
+        addr: u64,
+        rank: usize,
+        build: impl FnOnce() -> Result<Vec<ChunkInfo>, E>,
+        ds_dims: &[usize],
+        chunk_dims: &[usize],
+        elem_size: usize,
+    ) -> Result<Arc<ChunkLayout>, E> {
+        let (layout, chunk_index) = {
+            let mut inner = self.lock();
+            let entry = inner.touch(addr);
+            (entry.chunk_layout.clone(), entry.chunk_index.clone())
+        };
+        if let Some(layout) = layout {
+            return Ok(layout);
+        }
+        let chunk_index = match chunk_index {
+            Some(ci) => ci,
+            None => {
+                let index = self.index_for(addr, rank, build)?;
+                let chunks: Vec<ChunkInfo> = index.values().cloned().collect();
+                Arc::new(ChunkIndex::build(&chunks, rank))
+            }
+        };
+        let layout = ChunkLayout::build(&chunk_index, ds_dims, chunk_dims, elem_size);
+        let mut inner = self.lock();
+        let entry = inner.touch(addr);
+        entry.chunk_index.get_or_insert(chunk_index);
+        let layout = Arc::clone(entry.chunk_layout.get_or_insert_with(|| Arc::new(layout)));
+        inner.trim_datasets(addr);
+        Ok(layout)
+    }
+
+    /// Cached decompressed chunk at `coord` of the dataset at `addr`.
+    ///
+    /// O(1) lookup; the clone is an `Arc` refcount bump, not a copy of the
+    /// underlying decompressed data.
+    pub fn get_decompressed_in(&self, addr: u64, coord: &[u64]) -> Option<Arc<CacheAlignedBuffer>> {
+        self.lock().get_decompressed(addr, coord)
+    }
+
+    /// Cache decompressed chunk data for `coord` of the dataset at `addr`.
+    /// Returns the `Arc`-shared buffer now cached (or already cached).
+    pub fn put_decompressed_in(
+        &self,
+        addr: u64,
+        coord: ChunkCoord,
+        data: Vec<u8>,
+    ) -> Arc<CacheAlignedBuffer> {
+        self.put_decompressed_aligned_in(addr, coord, CacheAlignedBuffer::from_vec(data))
+    }
+
+    /// [`Self::put_decompressed_in`] for an already-aligned buffer.
+    pub fn put_decompressed_aligned_in(
+        &self,
+        addr: u64,
+        coord: ChunkCoord,
+        data: CacheAlignedBuffer,
+    ) -> Arc<CacheAlignedBuffer> {
+        let data = Arc::new(data);
+        self.lock().put_decompressed((addr, coord), data)
+    }
+
+    /// Record that the given chunk coordinates of the dataset at `addr` are
+    /// predicted to be accessed soon (bookkeeping only).
+    ///
+    /// This does **not** prefetch or pre-decompress anything — it only
+    /// checks whether each coordinate is already in the chunk index and
+    /// updates access-pattern stats accordingly.
+    pub fn prefetch_hint_in(&self, addr: u64, next_coords: &[ChunkCoord]) {
+        let mut inner = self.lock();
+        let Some(index) = inner.entry(addr).and_then(|e| e.index.clone()) else {
+            return;
+        };
+        let known = next_coords
+            .iter()
+            .filter(|c| index.contains_key(*c))
+            .count();
+        inner.stats.sequential_count += known as u64;
+    }
+
+    // ----- Address-less operations on the bound dataset -----
+
+    /// Bind the address-less methods to the dataset at chunk-index address
+    /// `addr`. Returns `true` if this changed the bound dataset.
+    ///
+    /// Each dataset's state is kept separately, so switching loses nothing
+    /// and never exposes one dataset's index or chunks to another. The
+    /// binding itself is shared, though: concurrent readers should use the
+    /// `addr`-taking methods rather than bind and then call these.
     pub fn ensure_dataset(&self, addr: u64) -> bool {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        if inner.index_addr == Some(addr) {
-            return false;
-        }
-        inner.index = None;
-        inner.chunk_index = None;
-        inner.chunk_layout = None;
-        inner.slots.clear();
-        inner.slot_index.clear();
-        inner.current_bytes = 0;
-        inner.last_coord = None;
-        inner.index_addr = Some(addr);
-        true
+        let mut inner = self.lock();
+        let changed = inner.current != Some(addr);
+        inner.current = Some(addr);
+        changed
     }
 
-    /// Returns `true` if the chunk index has been built.
+    /// Returns `true` if the bound dataset's chunk index has been built.
     pub fn has_index(&self) -> bool {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .index
-            .is_some()
+        let inner = self.lock();
+        inner
+            .entry(inner.current())
+            .is_some_and(|e| e.index.is_some())
     }
 
-    /// Build the chunk index from a pre-collected list of `ChunkInfo`.
+    /// Build the bound dataset's chunk index from a pre-collected list of
+    /// `ChunkInfo`.
     ///
     /// The `rank` parameter is used to truncate offsets to spatial dims only
     /// (B-tree v1 stores rank+1 offsets).
     pub fn populate_index(&self, chunks: &[ChunkInfo], rank: usize) {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        if inner.index.is_some() {
-            return; // already populated
-        }
-        let mut map = HashMap::with_capacity(chunks.len());
-
-        for ci in chunks {
-            let coord: ChunkCoord = ci.offsets.iter().take(rank).copied().collect();
-            map.insert(coord, ci.clone());
-        }
-        inner.index = Some(map);
+        let addr = self.lock().current();
+        let _ = self.index_for::<core::convert::Infallible>(addr, rank, || Ok(chunks.to_vec()));
     }
 
-    /// Look up a chunk by its spatial coordinate in the index.
+    /// Look up a chunk by its spatial coordinate in the bound dataset's index.
     pub fn lookup_index(&self, coord: &[u64]) -> Option<ChunkInfo> {
-        let inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        inner.index.as_ref()?.get(coord).cloned()
+        let inner = self.lock();
+        inner
+            .entry(inner.current())?
+            .index
+            .as_ref()?
+            .get(coord)
+            .cloned()
     }
 
-    /// Return all indexed chunks as a `Vec<ChunkInfo>` (order unspecified).
+    /// Return all of the bound dataset's indexed chunks (order unspecified).
     pub fn all_indexed_chunks(&self) -> Option<Vec<ChunkInfo>> {
-        let inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        inner.index.as_ref().map(|m| m.values().cloned().collect())
+        let inner = self.lock();
+        let index = inner.entry(inner.current())?.index.as_ref()?;
+        Some(index.values().cloned().collect())
     }
 
-    // ----- Chunk index (pre-built coordinate → ChunkInfo map) -----
-
-    /// Returns `true` if the chunk B-tree index has been built.
+    /// Returns `true` if the bound dataset's `ChunkIndex` has been built.
     pub fn has_chunk_index(&self) -> bool {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .chunk_index
-            .is_some()
+        let inner = self.lock();
+        inner
+            .entry(inner.current())
+            .is_some_and(|e| e.chunk_index.is_some())
     }
 
-    /// Build and store the chunk B-tree index from a pre-collected list of `ChunkInfo`.
+    /// Build and store the bound dataset's `ChunkIndex`.
     pub fn populate_chunk_index(&self, chunks: &[ChunkInfo], rank: usize) {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        if inner.chunk_index.is_some() {
-            return;
-        }
-        inner.chunk_index = Some(ChunkIndex::build(chunks, rank));
+        let built = Arc::new(ChunkIndex::build(chunks, rank));
+        let mut inner = self.lock();
+        let addr = inner.current();
+        inner.touch(addr).chunk_index.get_or_insert(built);
+        inner.trim_datasets(addr);
     }
 
-    // ----- Chunk layout (pre-computed assembly plan) -----
-
-    /// Returns `true` if the chunk layout has been computed.
+    /// Returns `true` if the bound dataset's chunk layout has been computed.
     pub fn has_chunk_layout(&self) -> bool {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .chunk_layout
-            .is_some()
+        let inner = self.lock();
+        inner
+            .entry(inner.current())
+            .is_some_and(|e| e.chunk_layout.is_some())
     }
 
-    /// Build and store the pre-computed chunk layout for fast assembly.
+    /// Build and store the bound dataset's chunk layout (needs its
+    /// `ChunkIndex`; does nothing without one).
     pub fn populate_chunk_layout(&self, ds_dims: &[usize], chunk_dims: &[usize], elem_size: usize) {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        if inner.chunk_layout.is_some() {
+        let mut inner = self.lock();
+        let addr = inner.current();
+        let entry = inner.touch(addr);
+        if entry.chunk_layout.is_some() {
             return;
         }
-        if let Some(ref idx) = inner.chunk_index {
-            inner.chunk_layout = Some(ChunkLayout::build(idx, ds_dims, chunk_dims, elem_size));
+        if let Some(idx) = entry.chunk_index.clone() {
+            entry.chunk_layout = Some(Arc::new(ChunkLayout::build(
+                &idx, ds_dims, chunk_dims, elem_size,
+            )));
         }
     }
 
-    /// Execute a function with a reference to the chunk layout.
-    ///
-    /// Returns `None` if the layout hasn't been computed yet.
+    /// Execute a function with a reference to the bound dataset's chunk
+    /// layout. Returns `None` if the layout hasn't been computed yet.
     pub fn with_chunk_layout<F, R>(&self, f: F) -> Option<R>
     where
         F: FnOnce(&ChunkLayout) -> R,
     {
-        let inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        inner.chunk_layout.as_ref().map(f)
+        let layout = {
+            let inner = self.lock();
+            inner.entry(inner.current())?.chunk_layout.clone()?
+        };
+        Some(f(&layout))
     }
 
-    // ----- Decompressed data cache (LRU) -----
-
-    /// Try to get cached decompressed data for a chunk coordinate.
+    /// Try to get cached decompressed data for a chunk of the bound dataset.
     ///
-    /// O(1) lookup. Returns an owned copy for API compatibility with callers
-    /// that need a `Vec<u8>`; prefer [`Self::get_decompressed_aligned`] when
-    /// an `Arc`-shared buffer works for the caller, since that avoids the
-    /// copy entirely.
+    /// Returns an owned copy; prefer [`Self::get_decompressed_aligned`] when
+    /// an `Arc`-shared buffer works for the caller.
     pub fn get_decompressed(&self, coord: &[u64]) -> Option<Vec<u8>> {
         self.get_decompressed_aligned(coord)
             .map(|arc| arc.as_slice().to_vec())
     }
 
-    /// Try to get a reference-counted clone of the aligned buffer for a chunk.
-    ///
-    /// O(1) index lookup; the clone is an `Arc` refcount bump, not a copy of
-    /// the underlying decompressed data.
+    /// Reference-counted cached buffer for a chunk of the bound dataset.
     pub fn get_decompressed_aligned(&self, coord: &[u64]) -> Option<Arc<CacheAlignedBuffer>> {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        inner.tick += 1;
-        let tick = inner.tick;
-
-        // Track sequential vs random access
-        let is_sequential = inner.last_coord.as_ref().is_some_and(|prev| {
-            // Sequential if exactly one dimension changed
-            let changes: usize = prev
-                .iter()
-                .zip(coord.iter())
-                .filter(|(a, b)| a != b)
-                .count();
-            changes <= 1
-        });
-        if is_sequential {
-            inner.stats.sequential_count += 1;
-        } else if inner.last_coord.is_some() {
-            inner.stats.random_count += 1;
-        }
-        inner.last_coord = Some(coord.to_vec());
-
-        let found = if let Some(&idx) = inner.slot_index.get(coord) {
-            inner.slots[idx].last_access = tick;
-            Some(Arc::clone(&inner.slots[idx].data))
-        } else {
-            None
-        };
-        if let Some(ref data) = found {
-            inner.stats.hits += 1;
-            inner.stats.bytes_read += data.len() as u64;
-        } else {
-            inner.stats.misses += 1;
-        }
-        found
+        let mut inner = self.lock();
+        let addr = inner.current();
+        inner.get_decompressed(addr, coord)
     }
 
-    /// Insert decompressed chunk data into the LRU cache.
-    ///
-    /// The data is stored in a [`CacheAlignedBuffer`] so subsequent reads
-    /// return cache-line-aligned memory. Returns the `Arc`-shared buffer that
-    /// is now cached (or already was), so the caller can reuse it directly
-    /// instead of holding a separate copy of the same data.
+    /// Insert decompressed chunk data for the bound dataset into the LRU
+    /// cache, returning the `Arc`-shared buffer now cached.
     pub fn put_decompressed(&self, coord: ChunkCoord, data: Vec<u8>) -> Arc<CacheAlignedBuffer> {
-        let aligned = CacheAlignedBuffer::from_vec(data);
-        self.put_decompressed_aligned(coord, aligned)
+        self.put_decompressed_aligned(coord, CacheAlignedBuffer::from_vec(data))
     }
 
-    /// Insert an already-aligned buffer into the LRU cache.
-    ///
-    /// Returns the `Arc`-shared buffer now held by the cache (the one just
-    /// inserted, or the existing cached copy if `coord` was already present).
+    /// Insert an already-aligned buffer for the bound dataset.
     pub fn put_decompressed_aligned(
         &self,
         coord: ChunkCoord,
         data: CacheAlignedBuffer,
     ) -> Arc<CacheAlignedBuffer> {
         let data = Arc::new(data);
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        let data_len = data.len();
-
-        // Don't cache if single chunk exceeds budget — still return the data
-        // to the caller, just don't retain it.
-        if data_len > inner.max_bytes {
-            return data;
+        let mut inner = self.lock();
+        let addr = inner.current();
+        inner.put_decompressed((addr, coord), data)
     }
 
-        // Check if already present
-        inner.tick += 1;
-        let tick = inner.tick;
-        if let Some(&idx) = inner.slot_index.get(&coord) {
-            inner.slots[idx].last_access = tick;
-            return Arc::clone(&inner.slots[idx].data); // already cached
+    /// [`Self::prefetch_hint_in`] for the bound dataset.
+    pub fn prefetch_hint(&self, next_coords: &[ChunkCoord]) {
+        let addr = self.lock().current();
+        self.prefetch_hint_in(addr, next_coords);
     }
 
-        // Evict until we have room
-        while inner.slots.len() >= inner.max_slots
-            || (inner.current_bytes + data_len > inner.max_bytes && !inner.slots.is_empty())
-        {
-            // Find LRU slot
-            let lru_idx = inner
-                .slots
-                .iter()
-                .enumerate()
-                .min_by_key(|(_, s)| s.last_access)
-                .map(|(i, _)| i)
-                .unwrap();
-            let removed = inner.slots.swap_remove(lru_idx);
-            inner.slot_index.remove(&removed.coord);
-            // swap_remove moved the former last element into `lru_idx` (unless
-            // it *was* the last element) — fix up that element's index entry.
-            if lru_idx < inner.slots.len() {
-                let moved_coord = inner.slots[lru_idx].coord.clone();
-                inner.slot_index.insert(moved_coord, lru_idx);
-            }
-            inner.current_bytes -= removed.data.len();
-            inner.stats.evictions += 1;
-        }
+    // ----- Whole-cache operations -----
 
-        inner.current_bytes += data_len;
-        let new_idx = inner.slots.len();
-        inner.slot_index.insert(coord.clone(), new_idx);
-        inner.slots.push(CachedChunk {
-            coord,
-            data: Arc::clone(&data),
-            last_access: tick,
-        });
-        data
-    }
-
-    /// Clear the entire cache (index + decompressed data).
+    /// Clear the entire cache (indexes + decompressed data + stats).
     pub fn clear(&self) {
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        inner.index = None;
-        inner.index_addr = None;
+        let mut inner = self.lock();
+        inner.datasets.clear();
+        inner.current = None;
         inner.slots.clear();
         inner.slot_index.clear();
         inner.current_bytes = 0;
         inner.tick = 0;
         inner.last_coord = None;
         inner.stats = AccessStats::default();
-        inner.chunk_index = None;
-        inner.chunk_layout = None;
-    }
-
-    /// Record that the given chunk coordinates are predicted to be accessed
-    /// soon (bookkeeping only).
-    ///
-    /// This does **not** prefetch or pre-decompress anything — it only
-    /// checks whether each coordinate is already in the chunk index and
-    /// updates access-pattern stats accordingly. Real prefetching (e.g.
-    /// background pre-decompression) is not implemented.
-    pub fn prefetch_hint(&self, next_coords: &[ChunkCoord]) {
-        let inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        if inner.index.is_none() {
-            return;
-        }
-        drop(inner);
-        // For each predicted coordinate, verify it exists in the index.
-        // The index is already populated, so this is a no-op for known chunks.
-        // The purpose is to signal intent — callers can pre-decompress if needed.
-        // We touch the stats to record that prefetch hints were issued.
-        let mut inner = self.inner.lock().unwrap_or_else(|e| e.into_inner());
-        for coord in next_coords {
-            let exists = inner
-                .index
-                .as_ref()
-                .map(|idx| idx.contains_key(coord))
-                .unwrap_or(false);
-            if exists {
-                inner.stats.sequential_count += 1;
-            }
-        }
     }
 
     /// Return the current access pattern statistics.
     pub fn access_stats(&self) -> AccessStats {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .stats
-            .clone()
+        self.lock().stats.clone()
     }
 
     /// Update the sweep direction label in the access stats.
     pub fn set_sweep_direction(&self, direction: &'static str) {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .stats
-            .sweep_direction = Some(direction);
+        self.lock().stats.sweep_direction = Some(direction);
     }
 
-    /// Number of decompressed chunks currently cached.
+    /// Number of decompressed chunks currently cached (all datasets).
     pub fn cached_chunk_count(&self) -> usize {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .slots
-            .len()
+        self.lock().slots.len()
     }
 
-    /// Total bytes of decompressed data currently cached.
+    /// Total bytes of decompressed data currently cached (all datasets).
     pub fn cached_bytes(&self) -> usize {
-        self.inner
-            .lock()
-            .unwrap_or_else(|e| e.into_inner())
-            .current_bytes
+        self.lock().current_bytes
+    }
+
+    /// Number of datasets whose chunk index is currently kept.
+    pub fn indexed_dataset_count(&self) -> usize {
+        self.lock().datasets.len()
     }
 }
 
@@ -808,6 +967,92 @@ mod tests {
         assert_eq!(cache.cached_bytes(), 0);
     }
 
+    #[test]
+    fn datasets_sharing_coordinates_stay_separate() {
+        let cache = ChunkCache::new();
+        let a = vec![make_chunk(vec![0, 0], 0x100, 8)];
+        let b = vec![make_chunk(vec![0, 0], 0x900, 8)];
+        let got_a = cache.chunks_for::<()>(1, 1, || Ok(a.clone())).unwrap();
+        let got_b = cache.chunks_for::<()>(2, 1, || Ok(b.clone())).unwrap();
+        assert_eq!(got_a[0].address, 0x100);
+        assert_eq!(got_b[0].address, 0x900);
+        // Built once per dataset: a second lookup doesn't call the builder.
+        let again = cache
+            .chunks_for::<()>(1, 1, || panic!("index rebuilt"))
+            .unwrap();
+        assert_eq!(again[0].address, 0x100);
+
+        cache.put_decompressed_in(1, vec![0], vec![1; 4]);
+        cache.put_decompressed_in(2, vec![0], vec![2; 4]);
+        assert_eq!(
+            cache.get_decompressed_in(1, &[0]).unwrap().as_slice(),
+            &[1; 4]
+        );
+        assert_eq!(
+            cache.get_decompressed_in(2, &[0]).unwrap().as_slice(),
+            &[2; 4]
+        );
+        assert!(cache.get_decompressed_in(3, &[0]).is_none());
+        assert_eq!(cache.cached_chunk_count(), 2);
+
+        // The bound-dataset methods see only the bound dataset.
+        cache.ensure_dataset(2);
+        assert_eq!(cache.lookup_index(&[0]).unwrap().address, 0x900);
+        assert_eq!(cache.get_decompressed(&[0]).unwrap(), vec![2; 4]);
+    }
+
+    #[test]
+    fn dataset_indexes_are_bounded() {
+        let cache = ChunkCache::new();
+        for addr in 0..(MAX_INDEXED_DATASETS as u64 + 10) {
+            cache
+                .chunks_for::<()>(addr, 1, || Ok(vec![make_chunk(vec![0], addr, 8)]))
+                .unwrap();
+        }
+        assert_eq!(cache.indexed_dataset_count(), MAX_INDEXED_DATASETS);
+
+        // One huge index evicts the others but is itself kept.
+        let huge: Vec<ChunkInfo> = (0..MAX_INDEXED_CHUNKS as u64)
+            .map(|i| make_chunk(vec![i], i, 8))
+            .collect();
+        let got = cache.chunks_for::<()>(9999, 1, || Ok(huge)).unwrap();
+        assert_eq!(got.len(), MAX_INDEXED_CHUNKS);
+        assert_eq!(cache.indexed_dataset_count(), 1);
+    }
+
+    #[test]
+    fn concurrent_readers_of_different_datasets_see_their_own_chunks() {
+        let cache = std::sync::Arc::new(ChunkCache::with_capacity(1 << 20, 64));
+        let handles: Vec<_> = (0..8u64)
+            .map(|t| {
+                let cache = std::sync::Arc::clone(&cache);
+                std::thread::spawn(move || {
+                    for round in 0..500u64 {
+                        let addr = (t + round) % 16;
+                        let coord = vec![round % 4];
+                        let chunks = cache
+                            .chunks_for::<()>(addr, 1, || {
+                                Ok((0..4).map(|c| make_chunk(vec![c], addr, 8)).collect())
+                            })
+                            .unwrap();
+                        assert!(chunks.iter().all(|c| c.address == addr));
+                        let want = vec![addr as u8; 8];
+                        let got = match cache.get_decompressed_in(addr, &coord) {
+                            Some(hit) => hit.to_vec(),
+                            None => cache
+                                .put_decompressed_in(addr, coord, want.clone())
+                                .to_vec(),
+                        };
+                        assert_eq!(got, want);
+                    }
+                })
+            })
+            .collect();
+        for h in handles {
+            h.join().unwrap();
+        }
+    }
+
     #[test]
     fn duplicate_insert_is_noop() {
         let cache = ChunkCache::new();

crates/clawhdf5-format/src/chunk_grid.rs

@@ -0,0 +1,200 @@
+//! Chunk-index linearisation shared by the Fixed Array and Extensible Array
+//! chunk indexes (reader and writer).
+//!
+//! Both indexes store one element per chunk at a *linear* index, and the
+//! library derives that index from the chunk's scaled coordinates
+//! (`offset / chunk_dim`) using the dataset's **maximum** dimensions, not its
+//! current ones (`H5D__farray_idx_get_addr` / `H5D__earray_idx_get_addr`,
+//! via `layout->max_down_chunks`). A dataset whose current shape is smaller
+//! than its maxshape therefore has gaps in the index, and laying it out by the
+//! current shape puts every chunk after the first row in the wrong place.
+//!
+//! The Extensible Array adds one more step: its one unlimited dimension has no
+//! finite chunk count, so the library *swizzles* the coordinates to make that
+//! dimension the slowest-varying one (`H5VM_swizzle_coords`, which moves
+//! `coords[unlim_dim]` to the front and shifts the dimensions before it right
+//! by one) before linearising with `swizzled_max_down_chunks`. When the
+//! unlimited dimension is already dimension 0 no swizzle happens.
+
+#[cfg(not(feature = "std"))]
+extern crate alloc;
+
+#[cfg(not(feature = "std"))]
+use alloc::{vec, vec::Vec};
+
+use crate::error::FormatError;
+
+/// How a chunk index maps linear element indexes to chunk coordinates.
+#[derive(Debug, Clone)]
+pub(crate) struct ChunkGrid {
+    /// Spatial chunk dimensions, in dataset order.
+    chunk_dims: Vec<u64>,
+    /// Chunks per dimension covering the *current* extent, in dataset order.
+    cur_chunks: Vec<u64>,
+    /// Dataset dimension stored at each linearisation position (slowest
+    /// first). The identity except for a swizzled Extensible Array.
+    order: Vec<usize>,
+    /// Linear stride of each linearisation position.
+    down: Vec<u64>,
+}
+
+impl ChunkGrid {
+    /// Grid for a Fixed Array index: row-major over the chunk counts of the
+    /// maximum dimensions (`max_dims`, falling back to the current dimensions
+    /// when the dataspace records none).
+    pub(crate) fn fixed_array(
+        cur_dims: &[u64],
+        max_dims: Option<&[u64]>,
+        chunk_dims: &[u64],
+    ) -> Result<Self, FormatError> {
+        Self::build(cur_dims, max_dims, chunk_dims, None)
+    }
+
+    /// Grid for an Extensible Array index: like the Fixed Array, but the
+    /// unlimited dimension (the one whose maximum is `H5S_UNLIMITED`) is moved
+    /// to the slowest-varying position first.
+    pub(crate) fn extensible_array(
+        cur_dims: &[u64],
+        max_dims: Option<&[u64]>,
+        chunk_dims: &[u64],
+    ) -> Result<Self, FormatError> {
+        let unlim = max_dims.and_then(|m| m.iter().position(|&d| d == u64::MAX));
+        Self::build(cur_dims, max_dims, chunk_dims, unlim)
+    }
+
+    fn build(
+        cur_dims: &[u64],
+        max_dims: Option<&[u64]>,
+        chunk_dims: &[u64],
+        unlim: Option<usize>,
+    ) -> Result<Self, FormatError> {
+        let rank = chunk_dims.len();
+        if cur_dims.len() != rank || max_dims.is_some_and(|m| m.len() != rank) {
+            return Err(FormatError::ChunkedReadError(
+                "chunk index rank does not match the dataspace".into(),
+            ));
+        }
+        if chunk_dims.contains(&0) {
+            return Err(FormatError::ChunkedReadError(
+                "chunk dimension is zero".into(),
+            ));
+        }
+        let cur_chunks: Vec<u64> = cur_dims
+            .iter()
+            .zip(chunk_dims)
+            .map(|(&d, &c)| d.div_ceil(c))
+            .collect();
+        // Chunk counts of the maximum extent. An unlimited dimension has no
+        // finite count; it only ever sits in the slowest position, where its
+        // count never enters a stride. A (corrupt) maximum smaller than the
+        // current extent is widened so no allocated chunk becomes unreachable.
+        let max_chunks: Vec<u64> = (0..rank)
+            .map(|d| {
+                let max = max_dims.map_or(cur_dims[d], |m| m[d]);
+                if max == u64::MAX {
+                    u64::MAX
+                } else {
+                    max.div_ceil(chunk_dims[d]).max(cur_chunks[d])
+                }
+            })
+            .collect();
+
+        let mut order: Vec<usize> = (0..rank).collect();
+        if let Some(u) = unlim {
+            order.remove(u);
+            order.insert(0, u);
+        }
+        let mut down = vec![1u64; rank];
+        for p in (0..rank.saturating_sub(1)).rev() {
+            let next = max_chunks[order[p + 1]];
+            if next == u64::MAX {
+                // Only reachable with more than one unlimited dimension, which
+                // neither index type can describe.
+                return Err(FormatError::ChunkedReadError(
+                    "array chunk index with more than one unlimited dimension".into(),
+                ));
+            }
+            down[p] = down[p + 1].checked_mul(next).ok_or_else(|| {
+                FormatError::Overflow("chunk index linear stride overflows u64".into())
+            })?;
+        }
+        Ok(Self {
+            chunk_dims: chunk_dims.to_vec(),
+            cur_chunks,
+            order,
+            down,
+        })
+    }
+
+    /// Dataset-space offsets of the chunk stored at linear `index`, or `None`
+    /// when that chunk lies outside the current extent (the index still has a
+    /// slot for it; the library ignores such chunks on read).
+    pub(crate) fn offsets(&self, index: u64) -> Option<Vec<u64>> {
+        let rank = self.chunk_dims.len();
+        let mut offsets = vec![0u64; rank];
+        let mut rem = index;
+        for p in 0..rank {
+            let d = self.order[p];
+            let scaled = rem / self.down[p];
+            rem %= self.down[p];
+            if scaled >= self.cur_chunks[d] {
+                return None;
+            }
+            offsets[d] = scaled * self.chunk_dims[d];
+        }
+        Some(offsets)
+    }
+
+    /// Linear index of the chunk with scaled coordinates `scaled`
+    /// (`offset / chunk_dim` per dimension, in dataset order).
+    pub(crate) fn linear_index(&self, scaled: &[u64]) -> u64 {
+        self.order
+            .iter()
+            .zip(&self.down)
+            .map(|(&d, &stride)| scaled[d] * stride)
+            .sum()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn fixed_array_uses_max_dims() {
+        // shape (4, 6), chunks (2, 3), maxshape (20, 10): 10 x 4 chunk grid.
+        let g = ChunkGrid::fixed_array(&[4, 6], Some(&[20, 10]), &[2, 3]).unwrap();
+        assert_eq!(g.offsets(0), Some(vec![0, 0]));
+        assert_eq!(g.offsets(1), Some(vec![0, 3]));
+        assert_eq!(g.offsets(2), None); // column chunk 2 is beyond the extent
+        assert_eq!(g.offsets(4), Some(vec![2, 0]));
+        assert_eq!(g.offsets(5), Some(vec![2, 3]));
+        assert_eq!(g.offsets(8), None); // row chunk 2 is beyond the extent
+        assert_eq!(g.linear_index(&[1, 1]), 5);
+    }
+
+    #[test]
+    fn extensible_array_swizzles_unlimited_dim() {
+        // maxshape (10, None): dim 1 is unlimited and becomes slowest.
+        let g = ChunkGrid::extensible_array(&[4, 6], Some(&[10, u64::MAX]), &[2, 3]).unwrap();
+        // max chunks of dim 0 = 5, so index = c1 * 5 + c0.
+        assert_eq!(g.linear_index(&[1, 0]), 1);
+        assert_eq!(g.linear_index(&[0, 1]), 5);
+        assert_eq!(g.offsets(5), Some(vec![0, 3]));
+        assert_eq!(g.offsets(6), Some(vec![2, 3]));
+        assert_eq!(g.offsets(2), None);
+    }
+
+    #[test]
+    fn extensible_array_unlimited_first_is_row_major() {
+        let g = ChunkGrid::extensible_array(&[4, 6], Some(&[u64::MAX, 30]), &[2, 3]).unwrap();
+        // max chunks of dim 1 = 10.
+        assert_eq!(g.linear_index(&[1, 1]), 11);
+        assert_eq!(g.offsets(11), Some(vec![2, 3]));
+    }
+
+    #[test]
+    fn rejects_two_unlimited_dims_after_the_first() {
+        assert!(ChunkGrid::fixed_array(&[4, 6], Some(&[u64::MAX, u64::MAX]), &[2, 3]).is_err());
+    }
+}

crates/clawhdf5-format/src/chunked_write.rs

@@ -4,15 +4,17 @@
 extern crate alloc;
 
 #[cfg(not(feature = "std"))]
-use alloc::{vec, vec::Vec};
+use alloc::{format, vec, vec::Vec};
 
 use crate::checksum::jenkins_lookup3;
 use crate::chunk_cache::{CACHE_LINE_SIZE, align_to_cache_line};
+use crate::chunk_grid::ChunkGrid;
 use crate::ea_writer;
 use crate::error::FormatError;
 use crate::filter_pipeline::{
-    FILTER_DEFLATE, FILTER_FLETCHER32, FILTER_LZ4, FILTER_PCODEC, FILTER_SHUFFLE, FILTER_ZSTD,
-    FilterDescription, FilterPipeline,
+    FILTER_BITSHUFFLE, FILTER_BLOSC, FILTER_BZIP2, FILTER_DEFLATE, FILTER_FLETCHER32, FILTER_LZ4,
+    FILTER_LZF, FILTER_PCODEC, FILTER_PCODEC_NAME, FILTER_SHUFFLE, FILTER_ZSTD, FilterDescription,
+    FilterPipeline,
 };
 use crate::filters::compress_chunk;
 /// Round a file offset up to the next cache-line boundary.
@@ -44,8 +46,170 @@ pub struct ChunkOptions {
     pub lz4: bool,
     /// Zstandard compression level (1-22), None = no zstd. Filter ID 32015.
     pub zstd_level: Option<u32>,
-    /// Pcodec lossless numerical compression. Filter ID 32023.
+    /// Pcodec lossless numerical compression. Private, unregistered filter
+    /// ID [`FILTER_PCODEC`] (480): only clawhdf5 can read it.
     pub pcodec: bool,
+    /// A plugin compression filter (LZF, ...). Takes priority over the
+    /// codecs above. Each needs its cargo feature to be written.
+    pub plugin: Option<PluginFilter>,
+}
+
+/// A compression filter from the common HDF5 plugin set, written in the
+/// format the libhdf5 plugin (h5py / hdf5plugin) reads.
+#[derive(Debug, Clone, PartialEq, Eq)]
+#[non_exhaustive]
+pub enum PluginFilter {
+    /// LZF (filter 32000), h5py's built-in `compression="lzf"`. Needs the
+    /// `lzf` feature.
+    Lzf,
+    /// Bitshuffle (filter 32008): a bit transpose of each block of
+    /// `block_size` elements (0 = bitshuffle's default, else a multiple of
+    /// 8), optionally compressed. Needs the `bitshuffle` feature.
+    Bitshuffle {
+        /// Block size in elements; 0 for the default.
+        block_size: u32,
+        /// Compression after the transpose.
+        compression: BitshuffleCompression,
+    },
+    /// bzip2 (filter 307) at block size `level` (1-9). Needs the `bzip2`
+    /// feature.
+    Bzip2 {
+        /// Block size 1-9 (9 = hdf5plugin's default).
+        level: u32,
+    },
+    /// Blosc 1 (filter 32001): `codec` at `level` (0-9; 0 stores), after
+    /// `shuffle`. Needs the `blosc` feature.
+    Blosc {
+        /// The codec inside the Blosc frame.
+        codec: BloscCodec,
+        /// Compression level 0-9 (0 stores the data uncompressed).
+        level: u32,
+        /// The shuffle Blosc applies first.
+        shuffle: BloscShuffle,
+    },
+}
+
+/// The codec inside a Blosc frame that clawhdf5 can write. (It reads
+/// BloscLZ too, but cannot write it.)
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum BloscCodec {
+    /// LZ4.
+    Lz4,
+    /// Snappy.
+    Snappy,
+    /// Zlib, at the Blosc level.
+    Zlib,
+    /// Zstandard (clawhdf5's pure-Rust encoder has one level, about zstd 1).
+    Zstd,
+}
+
+/// The shuffle Blosc applies before compressing.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum BloscShuffle {
+    /// None.
+    None,
+    /// Byte shuffle (Blosc's default).
+    Byte,
+    /// Bit shuffle.
+    Bit,
+}
+
+/// What bitshuffle compresses its blocks with.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum BitshuffleCompression {
+    /// Transpose only.
+    None,
+    /// LZ4 (bitshuffle's `cname="lz4"`, the common choice).
+    Lz4,
+    /// Zstandard. clawhdf5's pure-Rust encoder has a single level (about
+    /// zstd's level 1); `level` is recorded in the file for other writers.
+    Zstd {
+        /// Level recorded in `cd_values[5]`.
+        level: u32,
+    },
+}
+
+impl PluginFilter {
+    /// Whether the filter reorders bytes itself, so the automatic shuffle
+    /// pre-filter would only get in its way.
+    fn shuffles_itself(&self) -> bool {
+        match self {
+            PluginFilter::Lzf => false,
+            PluginFilter::Bitshuffle { .. } => true,
+            PluginFilter::Bzip2 { .. } => false,
+            PluginFilter::Blosc { .. } => true,
+        }
+    }
+
+    /// The pipeline entry for this filter. `chunk_bytes` is one chunk's
+    /// uncompressed size (0 if unknown).
+    fn description(&self, element_size: u32, chunk_bytes: u32) -> FilterDescription {
+        match self {
+            // h5py's lzf_set_local: filter version, liblzf version, chunk
+            // size in bytes. Optional, as h5py flags it: a chunk the filter
+            // cannot shrink may then be stored unfiltered.
+            PluginFilter::Lzf => FilterDescription {
+                filter_id: FILTER_LZF,
+                name: Some("lzf".into()),
+                flags: 1,
+                client_data: vec![4, 0x0105, chunk_bytes],
+            },
+            // bshuf_h5_set_local: version 0.4, element size, block size,
+            // compression (0 none, 2 LZ4, 3 Zstandard), Zstandard level.
+            // hdf5-blosc's blosc_set_local: filter revision 2, Blosc format
+            // 2, type size, chunk size, then level, shuffle, compressor.
+            PluginFilter::Blosc {
+                codec,
+                level,
+                shuffle,
+            } => FilterDescription {
+                filter_id: FILTER_BLOSC,
+                name: Some("blosc".into()),
+                flags: 1,
+                client_data: vec![
+                    2,
+                    2,
+                    element_size,
+                    chunk_bytes,
+                    (*level).min(9),
+                    match shuffle {
+                        BloscShuffle::None => 0,
+                        BloscShuffle::Byte => 1,
+                        BloscShuffle::Bit => 2,
+                    },
+                    match codec {
+                        BloscCodec::Lz4 => 1,
+                        BloscCodec::Snappy => 3,
+                        BloscCodec::Zlib => 4,
+                        BloscCodec::Zstd => 5,
+                    },
+                ],
+            },
+            PluginFilter::Bzip2 { level } => FilterDescription {
+                filter_id: FILTER_BZIP2,
+                name: Some("bzip2".into()),
+                flags: 1,
+                client_data: vec![(*level).clamp(1, 9)],
+            },
+            PluginFilter::Bitshuffle {
+                block_size,
+                compression,
+            } => {
+                let mut cd = vec![0, 4, element_size, *block_size];
+                match compression {
+                    BitshuffleCompression::None => cd.push(0),
+                    BitshuffleCompression::Lz4 => cd.push(2),
+                    BitshuffleCompression::Zstd { level } => cd.extend([3, *level]),
+                }
+                FilterDescription {
+                    filter_id: FILTER_BITSHUFFLE,
+                    name: Some("bitshuffle; see https://github.com/kiyo-masui/bitshuffle".into()),
+                    flags: 1,
+                    client_data: cd,
+                }
+            }
+        }
+    }
 }
 
 /// Largest chunk the automatic choice produces, in bytes.
@@ -90,14 +254,33 @@ impl ChunkOptions {
             || self.lz4
             || self.zstd_level.is_some()
             || self.pcodec
+            || self.plugin.is_some()
     }
 
     /// Build a FilterPipeline from the options.
     pub fn build_pipeline(&self, element_size: u32) -> Option<FilterPipeline> {
+        self.build_pipeline_for_chunk(element_size, 0)
+    }
+
+    /// Build a FilterPipeline for chunks of `chunk_bytes` uncompressed bytes
+    /// (0 if unknown). Some plugin filters record the chunk size in their
+    /// client data.
+    pub fn build_pipeline_for_chunk(
+        &self,
+        element_size: u32,
+        chunk_bytes: u32,
+    ) -> Option<FilterPipeline> {
         let mut filters = Vec::new();
 
-        let has_compression =
-            self.deflate_level.is_some() || self.zstd_level.is_some() || self.lz4 || self.pcodec;
+        let plugin_shuffles = self
+            .plugin
+            .as_ref()
+            .is_some_and(PluginFilter::shuffles_itself);
+        let has_compression = self.deflate_level.is_some()
+            || self.zstd_level.is_some()
+            || self.lz4
+            || self.pcodec
+            || (self.plugin.is_some() && !plugin_shuffles);
 
         // Shuffle before compression. Applied if explicitly requested OR if compression
         // is active and the caller hasn't disabled it — matches h5py default behavior
@@ -111,11 +294,14 @@ impl ChunkOptions {
             });
         }
 
-        // Compression filters (mutually exclusive, priority: pcodec > zstd > lz4 > deflate)
-        if self.pcodec {
+        // Compression filters (mutually exclusive, priority: plugin > pcodec >
+        // zstd > lz4 > deflate)
+        if let Some(plugin) = &self.plugin {
+            filters.push(plugin.description(element_size, chunk_bytes));
+        } else if self.pcodec {
             filters.push(FilterDescription {
                 filter_id: FILTER_PCODEC,
-                name: Some("pcodec".into()),
+                name: Some(FILTER_PCODEC_NAME.into()),
                 flags: 0,
                 client_data: vec![element_size],
             });
@@ -381,39 +567,7 @@ fn serialize_v4_single_chunk(
     let ndims = chunk_dims.len() as u8 + 1;
     buf.push(ndims);
 
-    // dim_size_encoded_length: how many bytes per dimension
-    // We need to figure out the minimum encoding width
-    let max_dim = chunk_dims
-        .iter()
-        .map(|&d| d as u64)
-        .chain(core::iter::once(element_size as u64))
-        .max()
-        .unwrap_or(1);
-    let dim_encoded_len: u8 = if max_dim <= 0xFF {
-        1
-    } else if max_dim <= 0xFFFF {
-        2
-    } else {
-        4
-    };
-    buf.push(dim_encoded_len);
-
-    // dimension sizes (chunk dims + element size)
-    for &d in chunk_dims {
-        match dim_encoded_len {
-            1 => buf.push(d as u8),
-            2 => buf.extend_from_slice(&(d as u16).to_le_bytes()),
-            4 => buf.extend_from_slice(&d.to_le_bytes()),
-            _ => {}
-        }
-    }
-    // Element size dimension
-    match dim_encoded_len {
-        1 => buf.push(element_size as u8),
-        2 => buf.extend_from_slice(&(element_size as u16).to_le_bytes()),
-        4 => buf.extend_from_slice(&element_size.to_le_bytes()),
-        _ => {}
-    }
+    push_v4_chunk_dims(&mut buf, chunk_dims, element_size);
 
     // chunk index type = 1 (single chunk)
     buf.push(1);
@@ -443,45 +597,7 @@ fn serialize_v4_fixed_array(
     element_size: u32,
     max_bits: u8,
 ) -> Vec<u8> {
-    let mut buf = Vec::new();
-    buf.push(4); // version
-    buf.push(2); // class = chunked
-
-    let flags: u8 = 0x00;
-    buf.push(flags);
-
-    let ndims = chunk_dims.len() as u8 + 1;
-    buf.push(ndims);
-
-    let max_dim = chunk_dims
-        .iter()
-        .map(|&d| d as u64)
-        .chain(core::iter::once(element_size as u64))
-        .max()
-        .unwrap_or(1);
-    let dim_encoded_len: u8 = if max_dim <= 0xFF {
-        1
-    } else if max_dim <= 0xFFFF {
-        2
-    } else {
-        4
-    };
-    buf.push(dim_encoded_len);
-
-    for &d in chunk_dims {
-        match dim_encoded_len {
-            1 => buf.push(d as u8),
-            2 => buf.extend_from_slice(&(d as u16).to_le_bytes()),
-            4 => buf.extend_from_slice(&d.to_le_bytes()),
-            _ => {}
-        }
-    }
-    match dim_encoded_len {
-        1 => buf.push(element_size as u8),
-        2 => buf.extend_from_slice(&(element_size as u16).to_le_bytes()),
-        4 => buf.extend_from_slice(&element_size.to_le_bytes()),
-        _ => {}
-    }
+    let mut buf = layout_v4_chunked_prefix(chunk_dims, element_size);
 
     // chunk index type = 3 (Fixed Array)
     buf.push(3);
@@ -499,107 +615,175 @@ fn serialize_v4_fixed_array(
     buf
 }
 
+/// The part of a v4 chunked layout message before the chunk index type:
+/// version, class, flags and the chunk dimensions (plus the element size).
+/// Append a v4 layout's dimension width and its dimensions (the chunk
+/// dimensions, then the element size). Each takes the fewest bytes that hold
+/// the largest, as libhdf5 computes it (`H5D__chunk_set_sizes`:
+/// `(log2(dim) + 8) / 8`); HDF5 2.0.0 refuses any other width.
+pub(crate) fn push_v4_chunk_dims(buf: &mut Vec<u8>, chunk_dims: &[u32], element_size: u32) {
+    let max_dim = chunk_dims
+        .iter()
+        .copied()
+        .chain(core::iter::once(element_size))
+        .max()
+        .unwrap_or(1)
+        .max(1);
+    let width = (32 - max_dim.leading_zeros()).div_ceil(8) as usize;
+    buf.push(width as u8);
+    for &d in chunk_dims.iter().chain(core::iter::once(&element_size)) {
+        buf.extend_from_slice(&d.to_le_bytes()[..width]);
+    }
+}
+
+fn layout_v4_chunked_prefix(chunk_dims: &[u32], element_size: u32) -> Vec<u8> {
+    let mut buf = Vec::new();
+    buf.push(4); // version
+    buf.push(2); // class = chunked
+
+    let flags: u8 = 0x00;
+    buf.push(flags);
+
+    let ndims = chunk_dims.len() as u8 + 1;
+    buf.push(ndims);
+
+    push_v4_chunk_dims(&mut buf, chunk_dims, element_size);
+    buf
+}
+
+/// log2 of the elements per Fixed Array data block page (the library's
+/// default, `H5D_FARRAY_MAX_DBLK_PAGE_NELMTS_BITS`).
+const FA_PAGE_BITS: u8 = 10;
+
+pub(crate) fn push_addr(buf: &mut Vec<u8>, addr: u64, offset_size: u8) {
+    match offset_size {
+        4 => buf.extend_from_slice(&(addr as u32).to_le_bytes()),
+        _ => buf.extend_from_slice(&addr.to_le_bytes()),
+    }
+}
+
+/// Width of the chunk-size field of a filtered chunk index element. Must
+/// match the library's `H5D_FARRAY_FILT_COMPUTE_CHUNK_SIZE_LEN` (the EA and
+/// B-tree v2 indexes use the same formula):
+/// `1 + ((log2(unfiltered chunk bytes) + 8) / 8)`, capped at 8.
+pub(crate) fn filtered_chunk_size_len(slots: &[Option<WrittenChunk>]) -> usize {
+    let max_raw = slots
+        .iter()
+        .flatten()
+        .map(|c| c.raw_size)
+        .max()
+        .unwrap_or(1);
+    let log2_val = if max_raw <= 1 {
+        0
+    } else {
+        63 - max_raw.leading_zeros()
+    };
+    (1 + ((log2_val + 8) / 8) as usize).min(8)
+}
+
+/// Append one chunk index element: the chunk's address, plus its stored size
+/// and filter mask when the dataset is filtered. `None` is an unallocated
+/// chunk (undefined address, zero size and mask).
+pub(crate) fn push_index_element(
+    buf: &mut Vec<u8>,
+    slot: Option<&WrittenChunk>,
+    offset_size: u8,
+    chunk_size_bytes: Option<usize>,
+) {
+    match slot {
+        Some(c) => {
+            push_addr(buf, c.address, offset_size);
+            if let Some(n) = chunk_size_bytes {
+                buf.extend_from_slice(&c.compressed_size.to_le_bytes()[..n]);
+                buf.extend_from_slice(&c.filter_mask.to_le_bytes());
+            }
+        }
+        None => {
+            buf.extend(core::iter::repeat_n(0xFF, offset_size as usize));
+            if let Some(n) = chunk_size_bytes {
+                buf.extend(core::iter::repeat_n(0x00, n + 4));
+            }
+        }
+    }
+}
+
 /// Build a complete Fixed Array at a known absolute address.
+///
+/// `slots` holds one entry per element of the array, i.e. per chunk of the
+/// dataset's *maximum* extent in the order [`crate::chunk_grid`] defines;
+/// `None` marks a chunk that is not allocated. An array with more elements
+/// than fit in one page (`2^FA_PAGE_BITS`) gets a paged data block: a
+/// page-init bitmap after the prefix, then one checksummed page per
+/// `2^FA_PAGE_BITS` elements, the last one short (`H5FA__dblock_create`).
 pub fn build_fixed_array_at(
-    chunks: &[WrittenChunk],
+    slots: &[Option<WrittenChunk>],
     offset_size: u8,
     length_size: u8,
     has_filters: bool,
     fa_base_address: u64,
 ) -> Vec<u8> {
     let os = offset_size as usize;
-    let num_elements = chunks.len();
-
-    // For filtered chunks, compute chunk_size encoding width.
-    // Must match the HDF5 C library's H5D_FARRAY_FILT_COMPUTE_CHUNK_SIZE_LEN macro:
-    //   chunk_size_len = 1 + ((H5VM_log2_gen(chunk.size) + 8) / 8)
-    // where chunk.size is the unfiltered chunk size in bytes (product of all chunk dims).
-    let chunk_size_bytes: usize = if has_filters {
-        let max_raw = chunks.iter().map(|c| c.raw_size).max().unwrap_or(1);
-        let log2_val = if max_raw <= 1 {
-            0
-        } else {
-            63 - max_raw.leading_zeros()
-        };
-        let len = 1 + ((log2_val + 8) / 8) as usize;
-        len.min(8)
-    } else {
-        0
-    };
-
-    let elem_size = if has_filters {
-        os + chunk_size_bytes + 4
-    } else {
-        os
-    };
+    let num_elements = slots.len();
 
+    let chunk_size_bytes = has_filters.then(|| filtered_chunk_size_len(slots));
+    let elem_size = os + chunk_size_bytes.map_or(0, |n| n + 4);
     let client_id: u8 = if has_filters { 1 } else { 0 };
 
     // FAHD total size
-    let nelmts_field_size = length_size as usize;
-    let fahd_total_size = 4 + 1 + 1 + 1 + 1 + nelmts_field_size + os + 4;
+    let fahd_total_size = 4 + 1 + 1 + 1 + 1 + length_size as usize + os + 4;
     let fadb_address = fa_base_address + fahd_total_size as u64;
 
-    // Build FAHD
     let mut fahd = Vec::with_capacity(fahd_total_size);
     fahd.extend_from_slice(b"FAHD");
     fahd.push(0); // version
     fahd.push(client_id);
     fahd.push(elem_size as u8);
-
-    // max_nelmts_bits: use 10 as default (page_size = 1024), matching h5py convention
-    let max_bits: u8 = 10;
-    fahd.push(max_bits);
-
+    fahd.push(FA_PAGE_BITS);
     match length_size {
         4 => fahd.extend_from_slice(&(num_elements as u32).to_le_bytes()),
-        8 => fahd.extend_from_slice(&(num_elements as u64).to_le_bytes()),
         _ => fahd.extend_from_slice(&(num_elements as u64).to_le_bytes()),
     }
-
-    match offset_size {
-        4 => fahd.extend_from_slice(&(fadb_address as u32).to_le_bytes()),
-        8 => fahd.extend_from_slice(&fadb_address.to_le_bytes()),
-        _ => fahd.extend_from_slice(&fadb_address.to_le_bytes()),
-    }
-
-    // Checksum
+    push_addr(&mut fahd, fadb_address, offset_size);
     let checksum = jenkins_lookup3(&fahd);
     fahd.extend_from_slice(&checksum.to_le_bytes());
-
     assert_eq!(fahd.len(), fahd_total_size);
 
-    // Build FADB
+    // FADB prefix
     let mut fadb = Vec::new();
     fadb.extend_from_slice(b"FADB");
     fadb.push(0); // version
     fadb.push(client_id);
+    push_addr(&mut fadb, fa_base_address, offset_size);
 
-    // header address
-    match offset_size {
-        4 => fadb.extend_from_slice(&(fa_base_address as u32).to_le_bytes()),
-        8 => fadb.extend_from_slice(&fa_base_address.to_le_bytes()),
-        _ => fadb.extend_from_slice(&fa_base_address.to_le_bytes()),
+    let page_nelmts = 1usize << FA_PAGE_BITS;
+    if num_elements <= page_nelmts {
+        // Unpaged: the elements follow the prefix, one checksum over both.
+        for slot in slots {
+            push_index_element(&mut fadb, slot.as_ref(), offset_size, chunk_size_bytes);
         }
-
-    // Element data
-    for chunk in chunks {
-        match offset_size {
-            4 => fadb.extend_from_slice(&(chunk.address as u32).to_le_bytes()),
-            8 => fadb.extend_from_slice(&chunk.address.to_le_bytes()),
-            _ => fadb.extend_from_slice(&chunk.address.to_le_bytes()),
-        }
-        if has_filters {
-            // Write compressed size using chunk_size_bytes (variable width)
-            let cs_bytes = chunk.compressed_size.to_le_bytes();
-            fadb.extend_from_slice(&cs_bytes[..chunk_size_bytes]);
-            fadb.extend_from_slice(&chunk.filter_mask.to_le_bytes());
-        }
-    }
-
-    // FADB checksum
         let fadb_checksum = jenkins_lookup3(&fadb);
         fadb.extend_from_slice(&fadb_checksum.to_le_bytes());
+    } else {
+        // Paged: every page is written, so every page-init bit is set
+        // (MSB-first, as `H5VM_bit_set` packs them). The prefix and bitmap
+        // share a checksum; each page carries its own.
+        let npages = num_elements.div_ceil(page_nelmts);
+        let mut bitmap = vec![0u8; npages.div_ceil(8)];
+        for p in 0..npages {
+            bitmap[p / 8] |= 0x80 >> (p % 8);
+        }
+        fadb.extend_from_slice(&bitmap);
+        let prefix_checksum = jenkins_lookup3(&fadb);
+        fadb.extend_from_slice(&prefix_checksum.to_le_bytes());
+        for page in slots.chunks(page_nelmts) {
+            let start = fadb.len();
+            for slot in page {
+                push_index_element(&mut fadb, slot.as_ref(), offset_size, chunk_size_bytes);
+            }
+            let page_checksum = jenkins_lookup3(&fadb[start..]);
+            fadb.extend_from_slice(&page_checksum.to_le_bytes());
+        }
+    }
 
     let mut combined = fahd;
     combined.extend_from_slice(&fadb);
@@ -634,7 +818,12 @@ pub fn precompress_chunks(
     element_size: usize,
     options: &ChunkOptions,
 ) -> Result<PrecompressedChunks, FormatError> {
-    let pipeline = options.build_pipeline(element_size as u32);
+    let chunk_bytes = chunk_dims
+        .iter()
+        .try_fold(element_size as u64, |acc, &d| acc.checked_mul(d))
+        .and_then(|b| u32::try_from(b).ok())
+        .unwrap_or(0);
+    let pipeline = options.build_pipeline_for_chunk(element_size as u32, chunk_bytes);
     let has_filters = pipeline.is_some();
     let pipeline_message = pipeline.as_ref().map(|pl| pl.serialize());
 
@@ -667,7 +856,8 @@ pub fn build_chunked_data_from_precompressed(
     pre: &PrecompressedChunks,
     base_address: u64,
     maxshape: Option<&[u64]>,
-) -> ChunkedDataResult {
+) -> Result<ChunkedDataResult, FormatError> {
+    let index = ChunkIndexPlan::new(&pre.shape, maxshape, &pre.chunk_dims)?;
     let offset_size: u8 = 8;
     let length_size: u8 = 8;
     let num_chunks = pre.chunks.len();
@@ -693,17 +883,18 @@ pub fn build_chunked_data_from_precompressed(
     }
 
     let chunk_dims_u32: Vec<u32> = pre.chunk_dims.iter().map(|&d| d as u32).collect();
-    let use_extensible = maxshape.is_some_and(|ms| ms.contains(&u64::MAX));
 
     let aligned_idx = align_to_cache_line(data_buf.len());
     if aligned_idx > data_buf.len() {
         data_buf.resize(aligned_idx, 0u8);
     }
 
-    let layout_message = if use_extensible {
+    let layout_message = match &index {
+        ChunkIndexPlan::ExtensibleArray(grid) => {
             let ea_address = base_address + data_buf.len() as u64;
+            let slots = index_slots(grid, &pre.shape, &pre.chunk_dims, &written_chunks, None)?;
             let ea_bytes = ea_writer::build_extensible_array_at(
-            &written_chunks,
+                &slots,
                 offset_size,
                 length_size,
                 pre.has_filters,
@@ -716,7 +907,8 @@ pub fn build_chunked_data_from_precompressed(
                 offset_size,
                 element_size as u32,
             )
-    } else if num_chunks == 1 {
+        }
+        ChunkIndexPlan::SingleChunk => {
             let chunk_addr = written_chunks[0].address;
             let filtered_size = if pre.has_filters {
                 Some(written_chunks[0].compressed_size)
@@ -732,10 +924,18 @@ pub fn build_chunked_data_from_precompressed(
                 offset_size,
                 element_size as u32,
             )
-    } else {
+        }
+        ChunkIndexPlan::FixedArray(grid, nslots) => {
             let fa_address = base_address + data_buf.len() as u64;
-        let fa_bytes = build_fixed_array_at(
+            let slots = index_slots(
+                grid,
+                &pre.shape,
+                &pre.chunk_dims,
                 &written_chunks,
+                Some(*nslots),
+            )?;
+            let fa_bytes = build_fixed_array_at(
+                &slots,
                 offset_size,
                 length_size,
                 pre.has_filters,
@@ -747,15 +947,263 @@ pub fn build_chunked_data_from_precompressed(
                 fa_address,
                 offset_size,
                 element_size as u32,
-            10, // max_nelmts_bits — matches h5py convention
+                FA_PAGE_BITS,
             )
+        }
+        ChunkIndexPlan::BTreeV2 => {
+            let bt_address = base_address + data_buf.len() as u64;
+            let records: Vec<(Vec<u64>, &WrittenChunk)> = written_chunks
+                .iter()
+                .enumerate()
+                .map(|(i, c)| (scaled_coords(&pre.shape, &pre.chunk_dims, i), c))
+                .collect();
+            let (bt_bytes, node_size) = build_btree_v2_chunk_index_at(
+                pre.shape.len(),
+                &records,
+                offset_size,
+                length_size,
+                pre.has_filters,
+                bt_address,
+            )?;
+            data_buf.extend_from_slice(&bt_bytes);
+            serialize_v4_btree_v2(
+                &chunk_dims_u32,
+                bt_address,
+                offset_size,
+                element_size as u32,
+                node_size,
+            )
+        }
     };
 
-    ChunkedDataResult {
+    Ok(ChunkedDataResult {
         data_bytes: data_buf,
         layout_message,
         pipeline_message: pre.pipeline_message.clone(),
+    })
+}
+
+/// Most slots a Fixed Array index may have before we refuse to build it: its
+/// data block holds one element per chunk of the *maximum* extent, so a huge
+/// finite maxshape with small chunks would otherwise exhaust memory.
+const MAX_FIXED_ARRAY_SLOTS: u64 = 1 << 26;
+
+/// Which chunk index a dataset gets, following the library's choice in
+/// `H5D__layout_set_latest_indexing`: version-2 B-tree for more than one
+/// unlimited dimension, Extensible Array for exactly one, Fixed Array for a
+/// finite maxshape, Single Chunk when the whole maximum extent is one chunk.
+enum ChunkIndexPlan {
+    SingleChunk,
+    /// The grid and the number of array elements (chunks of the max extent).
+    FixedArray(ChunkGrid, usize),
+    ExtensibleArray(ChunkGrid),
+    BTreeV2,
+}
+
+impl ChunkIndexPlan {
+    fn new(
+        shape: &[u64],
+        maxshape: Option<&[u64]>,
+        chunk_dims: &[u64],
+    ) -> Result<Self, FormatError> {
+        let bad = |what: &str| FormatError::ChunkedReadError(format!("maxshape: {what}"));
+        if let Some(ms) = maxshape {
+            if ms.len() != shape.len() {
+                return Err(bad("rank differs from the shape"));
             }
+            if ms.iter().zip(shape).any(|(&m, &s)| m < s) {
+                return Err(bad("smaller than the shape"));
+            }
+        }
+        let max = maxshape.unwrap_or(shape);
+        let nunlim = max.iter().filter(|&&d| d == u64::MAX).count();
+        match nunlim {
+            0 => {
+                let nslots = max
+                    .iter()
+                    .zip(chunk_dims)
+                    .try_fold(1u64, |acc, (&m, &c)| acc.checked_mul(m.div_ceil(c.max(1))))
+                    .filter(|&n| n <= MAX_FIXED_ARRAY_SLOTS)
+                    .ok_or_else(|| {
+                        bad("too many chunks for a Fixed Array index; \
+                             use larger chunks or an unlimited dimension")
+                    })?;
+                // A Single Chunk index needs that one chunk to exist; an
+                // empty dataset gets an all-unallocated Fixed Array instead.
+                let empty = shape.contains(&0);
+                if nslots == 1 && !empty {
+                    Ok(Self::SingleChunk)
+                } else {
+                    let grid = ChunkGrid::fixed_array(shape, Some(max), chunk_dims)?;
+                    Ok(Self::FixedArray(grid, nslots as usize))
+                }
+            }
+            1 => Ok(Self::ExtensibleArray(ChunkGrid::extensible_array(
+                shape,
+                Some(max),
+                chunk_dims,
+            )?)),
+            _ => Ok(Self::BTreeV2),
+        }
+    }
+}
+
+/// Place each written chunk at its linear index in `grid`. `chunks` are in
+/// row-major order over the chunks of the current extent (`split_into_chunks`).
+/// `len` fixes the slot count (Fixed Array); otherwise it is one past the
+/// highest index used.
+fn index_slots(
+    grid: &ChunkGrid,
+    shape: &[u64],
+    chunk_dims: &[u64],
+    chunks: &[WrittenChunk],
+    len: Option<usize>,
+) -> Result<Vec<Option<WrittenChunk>>, FormatError> {
+    let mut placed: Vec<(usize, &WrittenChunk)> = Vec::with_capacity(chunks.len());
+    for (i, chunk) in chunks.iter().enumerate() {
+        let scaled = scaled_coords(shape, chunk_dims, i);
+        let idx = usize::try_from(grid.linear_index(&scaled))
+            .map_err(|_| FormatError::Overflow("chunk index slot".into()))?;
+        placed.push((idx, chunk));
+    }
+    let n = len.unwrap_or_else(|| placed.iter().map(|&(i, _)| i + 1).max().unwrap_or(0));
+    let mut slots = vec![None; n];
+    for (idx, chunk) in placed {
+        *slots
+            .get_mut(idx)
+            .ok_or_else(|| FormatError::Overflow("chunk index slot".into()))? = Some(chunk.clone());
+    }
+    Ok(slots)
+}
+
+/// Scaled coordinates (`offset / chunk_dim`) of the `i`-th chunk in the
+/// row-major order `split_into_chunks` produces over the current extent.
+fn scaled_coords(shape: &[u64], chunk_dims: &[u64], i: usize) -> Vec<u64> {
+    let rank = shape.len();
+    let mut scaled = vec![0u64; rank];
+    let mut rem = i as u64;
+    for d in (0..rank).rev() {
+        let n = shape[d].div_ceil(chunk_dims[d]);
+        scaled[d] = rem % n;
+        rem /= n;
+    }
+    scaled
+}
+
+/// Node size the library gives a chunk index B-tree (`H5D_BT2_NODE_SIZE`),
+/// with its split and merge percentages.
+const BT2_NODE_SIZE: u32 = 2048;
+const BT2_SPLIT_PERCENT: u8 = 100;
+const BT2_MERGE_PERCENT: u8 = 40;
+/// B-tree v2 record types for chunk indexes (`H5B2_CDSET_ID`,
+/// `H5B2_CDSET_FILT_ID`).
+const BT2_CHUNK_UNFILTERED: u8 = 10;
+const BT2_CHUNK_FILTERED: u8 = 11;
+
+/// Build a version-2 B-tree chunk index (the library's index for datasets
+/// with more than one unlimited dimension) at a known absolute address.
+///
+/// `records` are `(scaled coordinates, chunk)` in lexicographic order of the
+/// coordinates, which is the order the library's comparator
+/// (`H5VM_vector_cmp_u`) keeps them in. The tree is a single leaf: the
+/// library's 2048-byte node when the records fit, otherwise a leaf node
+/// sized to hold them all (the root's record count is 16-bit, so at most
+/// 65535 chunks). Returns the bytes and the node size the layout message
+/// must record.
+fn build_btree_v2_chunk_index_at(
+    rank: usize,
+    records: &[(Vec<u64>, &WrittenChunk)],
+    offset_size: u8,
+    length_size: u8,
+    has_filters: bool,
+    base_address: u64,
+) -> Result<(Vec<u8>, u32), FormatError> {
+    let os = offset_size as usize;
+    let nrec = u16::try_from(records.len()).map_err(|_| {
+        FormatError::ChunkedReadError(
+            "more than 65535 chunks with more than one unlimited dimension: \
+             use larger chunks"
+                .into(),
+        )
+    })?;
+    let chunk_size_bytes = has_filters.then(|| {
+        let slots: Vec<Option<WrittenChunk>> =
+            records.iter().map(|(_, c)| Some((*c).clone())).collect();
+        filtered_chunk_size_len(&slots)
+    });
+    let record_size = os + chunk_size_bytes.map_or(0, |n| n + 4) + 8 * rank;
+    // Leaf: signature, version, type, records, checksum.
+    let leaf_len = 4 + 1 + 1 + records.len() * record_size + 4;
+    let node_size = u32::try_from(leaf_len)
+        .map_err(|_| FormatError::Overflow("B-tree v2 leaf size".into()))?
+        .max(BT2_NODE_SIZE);
+    let tree_type = if has_filters {
+        BT2_CHUNK_FILTERED
+    } else {
+        BT2_CHUNK_UNFILTERED
+    };
+
+    let hdr_len = 4 + 1 + 1 + 4 + 2 + 2 + 1 + 1 + os + 2 + length_size as usize + 4;
+    let leaf_address = base_address + hdr_len as u64;
+
+    let mut out = Vec::with_capacity(hdr_len + node_size as usize);
+    out.extend_from_slice(b"BTHD");
+    out.push(0); // version
+    out.push(tree_type);
+    out.extend_from_slice(&node_size.to_le_bytes());
+    out.extend_from_slice(&(record_size as u16).to_le_bytes());
+    out.extend_from_slice(&0u16.to_le_bytes()); // depth
+    out.push(BT2_SPLIT_PERCENT);
+    out.push(BT2_MERGE_PERCENT);
+    if records.is_empty() {
+        out.extend(core::iter::repeat_n(0xFF, os));
+    } else {
+        push_addr(&mut out, leaf_address, offset_size);
+    }
+    out.extend_from_slice(&nrec.to_le_bytes());
+    match length_size {
+        4 => out.extend_from_slice(&(records.len() as u32).to_le_bytes()),
+        _ => out.extend_from_slice(&(records.len() as u64).to_le_bytes()),
+    }
+    let sum = jenkins_lookup3(&out);
+    out.extend_from_slice(&sum.to_le_bytes());
+    debug_assert_eq!(out.len(), hdr_len);
+    if records.is_empty() {
+        return Ok((out, node_size));
+    }
+
+    let leaf_start = out.len();
+    out.extend_from_slice(b"BTLF");
+    out.push(0); // version
+    out.push(tree_type);
+    for (scaled, chunk) in records {
+        push_index_element(&mut out, Some(chunk), offset_size, chunk_size_bytes);
+        for &c in scaled {
+            out.extend_from_slice(&c.to_le_bytes());
+        }
+    }
+    let sum = jenkins_lookup3(&out[leaf_start..]);
+    out.extend_from_slice(&sum.to_le_bytes());
+    // The library reads whole nodes; pad the leaf out to the node size.
+    out.resize(leaf_start + node_size as usize, 0);
+    Ok((out, node_size))
+}
+
+/// Serialize a v4 layout message for a version-2 B-tree chunk index.
+fn serialize_v4_btree_v2(
+    chunk_dims: &[u32],
+    btree_address: u64,
+    offset_size: u8,
+    element_size: u32,
+    node_size: u32,
+) -> Vec<u8> {
+    let mut buf = layout_v4_chunked_prefix(chunk_dims, element_size);
+    buf.push(5); // chunk index type = 5 (version-2 B-tree)
+    buf.extend_from_slice(&node_size.to_le_bytes());
+    buf.push(BT2_SPLIT_PERCENT);
+    buf.push(BT2_MERGE_PERCENT);
+    push_addr(&mut buf, btree_address, offset_size);
+    buf
 }
 
 /// Build chunked data with absolute addresses.
@@ -790,11 +1238,7 @@ pub fn build_chunked_data_at_ext(
     maxshape: Option<&[u64]>,
 ) -> Result<ChunkedDataResult, FormatError> {
     let pre = precompress_chunks(raw_data, shape, chunk_dims, element_size, options)?;
-    Ok(build_chunked_data_from_precompressed(
-        &pre,
-        base_address,
-        maxshape,
-    ))
+    build_chunked_data_from_precompressed(&pre, base_address, maxshape)
 }
 
 /// Write selected elements into an existing in-memory dataset buffer.
@@ -1273,6 +1717,35 @@ mod tests {
         assert_eq!(pl.filters[1].client_data, vec![3]);
     }
 
+    #[test]
+    fn chunk_options_pipeline_lzf() {
+        let options = ChunkOptions {
+            plugin: Some(PluginFilter::Lzf),
+            ..Default::default()
+        };
+        assert!(options.is_chunked());
+        let pl = options.build_pipeline_for_chunk(8, 800).unwrap();
+        assert_eq!(pl.filters.len(), 2);
+        assert_eq!(pl.filters[0].filter_id, FILTER_SHUFFLE);
+        assert_eq!(pl.filters[1].filter_id, FILTER_LZF);
+        assert_eq!(pl.filters[1].client_data, vec![4, 0x0105, 800]);
+    }
+
+    #[test]
+    fn chunk_options_pipeline_bitshuffle_has_no_auto_shuffle() {
+        let options = ChunkOptions {
+            plugin: Some(PluginFilter::Bitshuffle {
+                block_size: 0,
+                compression: BitshuffleCompression::Zstd { level: 5 },
+            }),
+            ..Default::default()
+        };
+        let pl = options.build_pipeline(4).unwrap();
+        assert_eq!(pl.filters.len(), 1);
+        assert_eq!(pl.filters[0].filter_id, FILTER_BITSHUFFLE);
+        assert_eq!(pl.filters[0].client_data, vec![0, 4, 4, 0, 3, 5]);
+    }
+
     #[test]
     fn chunk_options_zstd_priority_over_deflate() {
         let options = ChunkOptions {
@@ -1314,6 +1787,7 @@ mod tests {
                 chunk_index_type,
                 single_chunk_filtered_size,
                 single_chunk_filter_mask,
+                ..
             } => {
                 assert_eq!(version, 4);
                 assert_eq!(chunk_index_type, Some(1));
@@ -1382,7 +1856,8 @@ mod tests {
                 filter_mask: 0,
             },
         ];
-        let fa = build_fixed_array_at(&chunks, 8, 8, false, 0x2000);
+        let slots: Vec<_> = chunks.into_iter().map(Some).collect();
+        let fa = build_fixed_array_at(&slots, 8, 8, false, 0x2000);
         // Should start with FAHD
         assert_eq!(&fa[0..4], b"FAHD");
         // FAHD size = 4+1+1+1+1+8+8+4 = 28
@@ -1429,7 +1904,8 @@ mod tests {
                 filter_mask: 0,
             },
         ];
-        let ea = ea_writer::build_extensible_array_at(&chunks, 8, 8, false, 0x2000);
+        let slots: Vec<_> = chunks.into_iter().map(Some).collect();
+        let ea = ea_writer::build_extensible_array_at(&slots, 8, 8, false, 0x2000);
         assert_eq!(&ea[0..4], b"EAHD");
         // Find EAIB after EAHD: 12 fixed + 6*8 stats + 8 addr + 4 checksum = 72
         let aehd_size = 4 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 6 * 8 + 8 + 4;

crates/clawhdf5-format/src/data_layout.rs

@@ -1,7 +1,7 @@
 //! HDF5 Data Layout message parsing (message type 0x0008).
 
 #[cfg(not(feature = "std"))]
-use alloc::{string::String, vec::Vec};
+use alloc::{format, string::String, vec::Vec};
 
 #[cfg(feature = "std")]
 use std::string::String;
@@ -24,6 +24,34 @@ pub struct VdsMapping {
     pub virtual_selection: Vec<u8>,
 }
 
+/// Most dimensions a layout message can list (libhdf5 `H5O_LAYOUT_NDIMS`):
+/// 32 dataspace dimensions plus the element size.
+const MAX_LAYOUT_NDIMS: usize = 33;
+
+/// libhdf5's checks on a chunked layout message's dimensions
+/// (`H5O__layout_decode`): at most [`MAX_LAYOUT_NDIMS`], no dimension 0, and
+/// before version 4 at least one dataspace dimension plus the element size.
+/// A zero chunk dimension used to read the dataset as all fill values.
+fn check_chunk_dims(dims: Vec<u32>, layout_version: u8) -> Result<Vec<u32>, FormatError> {
+    if dims.len() > MAX_LAYOUT_NDIMS {
+        return Err(FormatError::InvalidChunkDimensions(
+            "dimensionality is too large".into(),
+        ));
+    }
+    if layout_version < 4 && dims.len() < 2 {
+        return Err(FormatError::InvalidChunkDimensions(
+            "bad dimensions for chunked storage".into(),
+        ));
+    }
+    if let Some(u) = dims.iter().position(|&d| d == 0) {
+        return Err(FormatError::InvalidChunkDimensions(format!(
+            "bad chunk dimension value when parsing layout message - chunk dimension must be \
+             positive: mesg->u.chunk.dim[{u}] = 0"
+        )));
+    }
+    Ok(dims)
+}
+
 /// Parsed HDF5 data layout message.
 #[derive(Debug, Clone, PartialEq)]
 pub enum DataLayout {
@@ -45,7 +73,9 @@ pub enum DataLayout {
         chunk_dimensions: Vec<u32>,
         /// B-tree address, or `None` if undefined.
         btree_address: Option<u64>,
-        /// Layout version (3 or 4).
+        /// Layout version (3 or 4). Version 1/2 messages (HDF5 1.4/1.6-era)
+        /// use the same version-1 B-tree chunk index as version 3 and are
+        /// reported as 3.
         version: u8,
         /// Chunk index type (v4 only).
         chunk_index_type: Option<u8>,
@@ -53,6 +83,11 @@ pub enum DataLayout {
         single_chunk_filtered_size: Option<u64>,
         /// Filter mask for v4 single chunk with filters.
         single_chunk_filter_mask: Option<u32>,
+        /// Layout v4 flag bit 0 (`H5D_CHUNK_DONT_FILTER_PARTIAL_CHUNKS`):
+        /// partial edge chunks — those extending past the dataset's current
+        /// extent in some dimension — are stored without the filter pipeline,
+        /// even though their filter mask is 0. Always `false` for v3.
+        dont_filter_partial_edge_chunks: bool,
     },
     /// Virtual dataset layout (v4 only).
     Virtual {
@@ -67,21 +102,33 @@ pub enum DataLayout {
     },
 }
 
+/// Version-1 VDS mapping flag: the source file name is stored by an earlier
+/// entry, whose index follows in place of the name.
+const VDS_SOURCE_FILE_SHARED: u8 = 0x01;
+/// Version-1 VDS mapping flag: likewise for the source dataset name.
+const VDS_SOURCE_DSET_SHARED: u8 = 0x02;
+/// Version-1 VDS mapping flag: the source is in the virtual file itself
+/// (`"."`); no file name is stored.
+const VDS_SOURCE_SAME_FILE: u8 = 0x04;
+const VDS_ALL_FLAGS: u8 = VDS_SOURCE_FILE_SHARED | VDS_SOURCE_DSET_SHARED | VDS_SOURCE_SAME_FILE;
+
 /// Parse VDS mappings from global-heap object data.
 ///
 /// The global-heap block holding a VDS mapping list is laid out as
-/// (reverse-engineered and validated against HDF5 2.0):
+/// (`H5D__virtual_store_layout` / `H5D__virtual_load_layout` in libhdf5):
 ///
 /// ```text
 /// version(1) · nused(length_size, LE) · entry[nused] · checksum(4)
 /// ```
 ///
 /// Each entry is:
-/// - source file name — a null-terminated string in **block version 0**; in
-///   **block version 1** a same-file reference is encoded as a single `0x04`
-///   marker byte (the source file is the virtual file itself) in place of the
-///   name;
-/// - source dataset name (null-terminated string);
+/// - **block version 1 only:** a flags byte. `0x04`: the source is in the
+///   virtual file itself and no file name is stored; `0x01`/`0x02`: the
+///   source file/dataset name is that of an earlier entry, whose index
+///   (`length_size` bytes) is stored instead of the name. libhdf5 2.0 writes
+///   version 1 when the file's low version bound is 2.0 and it saves space;
+/// - source file name (null-terminated string, unless flagged above);
+/// - source dataset name (null-terminated string, unless flagged above);
 /// - source selection (serialized `H5S` dataspace selection — self-describing
 ///   in length);
 /// - virtual selection (serialized `H5S` dataspace selection).
@@ -107,7 +154,7 @@ pub fn parse_vds_mappings(
     // `nused` is untrusted; don't pre-allocate from it. Each entry consumes at
     // least a few bytes, so the loop is naturally bounded by the heap data and
     // a bogus `nused` simply errors out on the first short read.
-    let mut mappings = Vec::new();
+    let mut mappings: Vec<VdsMapping> = Vec::new();
     // Reads one self-describing selection at `pos`, returning its raw bytes and
     // advancing past it — bounds-checked so a corrupt selection can't overrun.
     let read_selection = |heap_data: &[u8], pos: &mut usize| -> Result<Vec<u8>, FormatError> {
@@ -127,17 +174,57 @@ pub fn parse_vds_mappings(
         Ok(bytes)
     };
 
-    for _ in 0..nused {
-        // Source file name (with the version-1 same-file marker handled).
-        let source_file = if version >= 1 && heap_data.get(pos) == Some(&0x04) {
+    if version > 1 {
+        return Err(FormatError::ChunkedReadError(
+            "unsupported VDS mapping block version".into(),
+        ));
+    }
+    for i in 0..nused {
+        // Version 1 prefixes each entry with a flags byte; a name may then be
+        // omitted (same file) or replaced by the index of an earlier entry
+        // holding the same name (`H5D__virtual_load_layout`).
+        let flags = if version >= 1 {
+            let f = *heap_data.get(pos).ok_or(FormatError::UnexpectedEof {
+                expected: pos + 1,
+                available: heap_data.len(),
+            })?;
             pos += 1;
+            if f & !VDS_ALL_FLAGS != 0 {
+                return Err(FormatError::ChunkedReadError(
+                    "unknown VDS mapping flags".into(),
+                ));
+            }
+            f
+        } else {
+            0
+        };
+        // Index of an earlier entry, for a shared name.
+        let earlier = |pos: &mut usize| -> Result<usize, FormatError> {
+            let idx = read_length(heap_data, *pos, length_size)?;
+            *pos += ls;
+            if idx >= i {
+                return Err(FormatError::ChunkedReadError(
+                    "VDS mapping shares a name with a later entry".into(),
+                ));
+            }
+            Ok(idx as usize)
+        };
+
+        let source_file = if flags & VDS_SOURCE_SAME_FILE != 0 {
             String::from(".")
+        } else if flags & VDS_SOURCE_FILE_SHARED != 0 {
+            let idx = earlier(&mut pos)?;
+            mappings[idx].source_file.clone()
         } else {
             read_null_terminated_string(heap_data, &mut pos)?
         };
 
-        // Source dataset name.
-        let source_dataset = read_null_terminated_string(heap_data, &mut pos)?;
+        let source_dataset = if flags & VDS_SOURCE_DSET_SHARED != 0 {
+            let idx = earlier(&mut pos)?;
+            mappings[idx].source_dataset.clone()
+        } else {
+            read_null_terminated_string(heap_data, &mut pos)?
+        };
 
         // Source selection, then virtual selection (both self-describing length).
         let source_selection = read_selection(heap_data, &mut pos)?;
@@ -256,6 +343,7 @@ impl DataLayout {
         let layout_class = data[1];
 
         match version {
+            1 | 2 => Self::parse_v1_v2(data, offset_size),
             3 => Self::parse_v3(data, layout_class, offset_size, length_size),
             // v5 (emitted by HDF5 1.14+/2.0 with `libver=latest`) uses the same
             // message structure as v4 — only the version number was bumped.
@@ -264,6 +352,87 @@ impl DataLayout {
         }
     }
 
+    /// Layout message versions 1 and 2 (HDF5 before 1.6.3):
+    ///
+    /// ```text
+    /// version(1) · dimensionality(1) · layout class(1) · reserved(5)
+    /// · address(offset_size)      — contiguous and chunked only
+    /// · dimension sizes(4 × dimensionality)
+    /// · compact data size(4) · compact raw data — compact only
+    /// ```
+    ///
+    /// The dimension sizes are the dataset's (contiguous/compact) or the
+    /// chunk's (chunked) extent plus a trailing element-size dimension, as in
+    /// version 3's chunked form. libhdf5 ignores them for contiguous storage
+    /// and sizes the data from the dataspace; the product of the stored
+    /// dimensions is that same size, and a disagreement (a dimension that was
+    /// truncated to 32 bits) is caught by the reader's size check rather than
+    /// returning wrong data.
+    fn parse_v1_v2(data: &[u8], offset_size: u8) -> Result<DataLayout, FormatError> {
+        ensure_len(data, 0, 8)?;
+        let dimensionality = data[1] as usize;
+        let layout_class = data[2];
+        // H5O_LAYOUT_NDIMS: 32 dataspace dimensions + the element-size one.
+        if dimensionality > 33 {
+            return Err(FormatError::Overflow(format!(
+                "data layout dimensionality {dimensionality} exceeds 33"
+            )));
+        }
+        let mut p = 8;
+        let os = offset_size as usize;
+        let address = match layout_class {
+            1 | 2 => {
+                ensure_len(data, p, os)?;
+                let a = if is_undefined(data, p, offset_size) {
+                    None
+                } else {
+                    Some(read_offset(data, p, offset_size)?)
+                };
+                p += os;
+                a
+            }
+            0 => None,
+            _ => return Err(FormatError::InvalidLayoutClass(layout_class)),
+        };
+        ensure_len(data, p, dimensionality * 4)?;
+        let dims: Vec<u32> = data[p..p + dimensionality * 4]
+            .as_chunks::<4>()
+            .0
+            .iter()
+            .map(|c| u32::from_le_bytes(*c))
+            .collect();
+        p += dimensionality * 4;
+        match layout_class {
+            0 => {
+                ensure_len(data, p, 4)?;
+                let size =
+                    u32::from_le_bytes([data[p], data[p + 1], data[p + 2], data[p + 3]]) as usize;
+                ensure_len(data, p + 4, size)?;
+                Ok(DataLayout::Compact {
+                    data: data[p + 4..p + 4 + size].to_vec(),
+                })
+            }
+            1 => {
+                let size = dims
+                    .iter()
+                    .try_fold(1u64, |acc, &d| acc.checked_mul(d as u64))
+                    .ok_or_else(|| {
+                        FormatError::Overflow(format!("contiguous layout size {dims:?}"))
+                    })?;
+                Ok(DataLayout::Contiguous { address, size })
+            }
+            _ => Ok(DataLayout::Chunked {
+                chunk_dimensions: check_chunk_dims(dims, 2)?,
+                btree_address: address,
+                version: 3,
+                chunk_index_type: None,
+                single_chunk_filtered_size: None,
+                single_chunk_filter_mask: None,
+                dont_filter_partial_edge_chunks: false,
+            }),
+        }
+    }
+
     fn parse_v3(
         data: &[u8],
         layout_class: u8,
@@ -316,12 +485,13 @@ impl DataLayout {
                     p += 4;
                 }
                 Ok(DataLayout::Chunked {
-                    chunk_dimensions,
+                    chunk_dimensions: check_chunk_dims(chunk_dimensions, 3)?,
                     btree_address,
                     version: 3,
                     chunk_index_type: None,
                     single_chunk_filtered_size: None,
                     single_chunk_filter_mask: None,
+                    dont_filter_partial_edge_chunks: false,
                 })
             }
             _ => Err(FormatError::InvalidLayoutClass(layout_class)),
@@ -364,47 +534,40 @@ impl DataLayout {
                 let dimensionality = data[pos + 1] as usize;
                 let dim_size_encoded_length = data[pos + 2] as usize;
                 let mut p = pos + 3;
+                if dimensionality > MAX_LAYOUT_NDIMS {
+                    return Err(FormatError::InvalidChunkDimensions(
+                        "dimensionality is too large".into(),
+                    ));
+                }
 
-                // dimension sizes
+                // Each dimension takes 1 to 8 bytes (libhdf5 writes the
+                // fewest that hold the largest one, so 3, 5, 6 and 7 occur:
+                // a chunk dimension of 70 000 takes 3). libhdf5 refuses 0
+                // and more than 8.
+                if dim_size_encoded_length == 0 || dim_size_encoded_length > 8 {
+                    return Err(FormatError::InvalidChunkDimensions(
+                        "encoded chunk dimension size is too large".into(),
+                    ));
+                }
                 ensure_len(data, p, dimensionality * dim_size_encoded_length)?;
                 let mut chunk_dimensions = Vec::with_capacity(dimensionality);
                 for _ in 0..dimensionality {
-                    let val = match dim_size_encoded_length {
-                        1 => data[p] as u32,
-                        2 => u16::from_le_bytes([data[p], data[p + 1]]) as u32,
-                        4 => u32::from_le_bytes([data[p], data[p + 1], data[p + 2], data[p + 3]]),
-                        8 => {
-                            // V4 chunked encodes dimension sizes as 8 bytes, but
-                            // our ChunkedStorageV4 stores them as u32. We read only
-                            // the low 4 bytes (little-endian). This silently
-                            // truncates dimensions > 4 GiB, which are not expected
-                            // in practice (HDF5 chunk dimensions are always small).
-                            // If the high bytes are non-zero, the file is malformed
-                            // or uses dimensions we cannot represent.
-                            let high = u32::from_le_bytes([
-                                data[p + 4],
-                                data[p + 5],
-                                data[p + 6],
-                                data[p + 7],
-                            ]);
-                            if high != 0 {
-                                return Err(FormatError::UnexpectedEof {
-                                    expected: p + 8,
-                                    available: data.len(),
-                                });
-                            }
-                            u32::from_le_bytes([data[p], data[p + 1], data[p + 2], data[p + 3]])
-                        }
-                        _ => {
-                            return Err(FormatError::UnexpectedEof {
-                                expected: p + dim_size_encoded_length,
-                                available: data.len(),
-                            });
-                        }
-                    };
+                    let val = data[p..p + dim_size_encoded_length]
+                        .iter()
+                        .rev()
+                        .fold(0u64, |acc, &b| (acc << 8) | u64::from(b));
+                    // Chunk dimensions are held as u32; HDF5 2.0 can write
+                    // larger ones (layout version 5), which are refused
+                    // rather than truncated.
+                    let val = u32::try_from(val).map_err(|_| {
+                        FormatError::InvalidChunkDimensions(format!(
+                            "chunk dimension {val} is larger than 2^32 - 1, which is not supported"
+                        ))
+                    })?;
                     chunk_dimensions.push(val);
                     p += dim_size_encoded_length;
                 }
+                let chunk_dimensions = check_chunk_dims(chunk_dimensions, 4)?;
 
                 // chunk index type
                 ensure_len(data, p, 1)?;
@@ -505,6 +668,7 @@ impl DataLayout {
                     chunk_index_type: Some(chunk_index_type),
                     single_chunk_filtered_size,
                     single_chunk_filter_mask,
+                    dont_filter_partial_edge_chunks: flags & 0x01 != 0,
                 })
             }
             3 => {
@@ -539,6 +703,202 @@ impl DataLayout {
 mod tests {
     use super::*;
 
+    /// Version 1/2 header: version, dimensionality, class, reserved(5).
+    fn v1v2_header(version: u8, ndims: u8, class: u8) -> Vec<u8> {
+        vec![version, ndims, class, 0, 0, 0, 0, 0]
+    }
+
+    #[test]
+    fn v2_compact() {
+        let mut buf = v1v2_header(2, 2, 0);
+        // dims (3 elements of 2 bytes) — no address for compact
+        buf.extend_from_slice(&3u32.to_le_bytes());
+        buf.extend_from_slice(&2u32.to_le_bytes());
+        buf.extend_from_slice(&6u32.to_le_bytes()); // compact size (u32 in v1/v2)
+        buf.extend_from_slice(&[1, 0, 2, 0, 3, 0]);
+        assert_eq!(
+            DataLayout::parse(&buf, 8, 8).unwrap(),
+            DataLayout::Compact {
+                data: vec![1, 0, 2, 0, 3, 0]
+            }
+        );
+    }
+
+    #[test]
+    fn v1_contiguous_size_from_dimensions() {
+        let mut buf = v1v2_header(1, 3, 1);
+        buf.extend_from_slice(&0x800u32.to_le_bytes()); // 4-byte address
+        for d in [10u32, 20, 4] {
+            buf.extend_from_slice(&d.to_le_bytes());
+        }
+        assert_eq!(
+            DataLayout::parse(&buf, 4, 4).unwrap(),
+            DataLayout::Contiguous {
+                address: Some(0x800),
+                size: 800,
+            }
+        );
+    }
+
+    #[test]
+    fn v1_contiguous_undefined_address() {
+        let mut buf = v1v2_header(1, 2, 1);
+        buf.extend_from_slice(&[0xFF; 8]);
+        buf.extend_from_slice(&5u32.to_le_bytes());
+        buf.extend_from_slice(&8u32.to_le_bytes());
+        assert_eq!(
+            DataLayout::parse(&buf, 8, 8).unwrap(),
+            DataLayout::Contiguous {
+                address: None,
+                size: 40,
+            }
+        );
+    }
+
+    #[test]
+    fn v1_chunked_maps_to_btree_v1_index() {
+        let mut buf = v1v2_header(1, 3, 2);
+        buf.extend_from_slice(&0x1234u64.to_le_bytes());
+        for d in [50u32, 50, 4] {
+            buf.extend_from_slice(&d.to_le_bytes());
+        }
+        assert_eq!(
+            DataLayout::parse(&buf, 8, 8).unwrap(),
+            DataLayout::Chunked {
+                chunk_dimensions: vec![50, 50, 4],
+                btree_address: Some(0x1234),
+                version: 3,
+                chunk_index_type: None,
+                single_chunk_filtered_size: None,
+                single_chunk_filter_mask: None,
+                dont_filter_partial_edge_chunks: false,
+            }
+        );
+    }
+
+    /// A v3 chunked layout message with these dims (element size last).
+    fn v3_chunked_msg(dims: &[u32]) -> Vec<u8> {
+        let mut buf = vec![3u8, 2, dims.len() as u8];
+        buf.extend_from_slice(&0x1000u64.to_le_bytes());
+        for d in dims {
+            buf.extend_from_slice(&d.to_le_bytes());
+        }
+        buf
+    }
+
+    #[test]
+    fn chunk_dimensions_are_checked_when_the_layout_is_parsed() {
+        assert!(DataLayout::parse(&v3_chunked_msg(&[4, 4, 8]), 8, 8).is_ok());
+        // A zero chunk dimension used to read as all fill values.
+        let err = DataLayout::parse(&v3_chunked_msg(&[4, 0, 8]), 8, 8).unwrap_err();
+        assert!(
+            matches!(&err, FormatError::InvalidChunkDimensions(m) if m.contains("dim[1] = 0")),
+            "{err:?}"
+        );
+        // Only the element-size dimension: libhdf5 "bad dimensions".
+        assert_eq!(
+            DataLayout::parse(&v3_chunked_msg(&[8]), 8, 8).unwrap_err(),
+            FormatError::InvalidChunkDimensions("bad dimensions for chunked storage".into())
+        );
+        assert_eq!(
+            DataLayout::parse(&v3_chunked_msg(&[1; 34]), 8, 8).unwrap_err(),
+            FormatError::InvalidChunkDimensions("dimensionality is too large".into())
+        );
+        // v1/v2 and v4 messages get the zero check too.
+        let mut v1 = v1v2_header(1, 2, 2);
+        v1.extend_from_slice(&0x1000u64.to_le_bytes());
+        v1.extend_from_slice(&0u32.to_le_bytes());
+        v1.extend_from_slice(&8u32.to_le_bytes());
+        assert!(matches!(
+            DataLayout::parse(&v1, 8, 8),
+            Err(FormatError::InvalidChunkDimensions(_))
+        ));
+        let mut v4 = vec![4u8, 2, 0, 2, 4];
+        v4.extend_from_slice(&0u32.to_le_bytes());
+        v4.extend_from_slice(&8u32.to_le_bytes());
+        v4.push(3); // fixed array index
+        v4.push(0); // page bits
+        v4.extend_from_slice(&0x1000u64.to_le_bytes());
+        assert!(matches!(
+            DataLayout::parse(&v4, 8, 8),
+            Err(FormatError::InvalidChunkDimensions(_))
+        ));
+    }
+
+    /// A v4 chunked layout (fixed array index) whose `dims` are each
+    /// encoded in `width` bytes.
+    fn v4_chunked_msg(width: u8, dims: &[u64]) -> Vec<u8> {
+        let mut m = vec![4u8, 2, 0, dims.len() as u8, width];
+        for &d in dims {
+            m.extend_from_slice(&d.to_le_bytes()[..width.min(8) as usize]);
+        }
+        m.push(3); // fixed array index
+        m.push(0); // page bits
+        m.extend_from_slice(&0x1000u64.to_le_bytes());
+        m
+    }
+
+    #[test]
+    fn v4_chunk_dimensions_take_1_to_8_bytes() {
+        // libhdf5 encodes each dimension in the fewest bytes that hold the
+        // largest: a chunk dimension of 70 000 takes 3, and 3, 5, 6 and 7
+        // were refused ("UnexpectedEof").
+        for width in 1..=8u8 {
+            let dims = [if width >= 3 { 70_000 } else { 200 }, 8];
+            let layout = DataLayout::parse(&v4_chunked_msg(width, &dims), 8, 8)
+                .unwrap_or_else(|e| panic!("width {width}: {e:?}"));
+            assert!(
+                matches!(&layout, DataLayout::Chunked { chunk_dimensions, .. }
+                    if chunk_dimensions.iter().map(|&d| u64::from(d)).eq(dims)),
+                "width {width}: {layout:?}"
+            );
+        }
+        // libhdf5 refuses 0 and more than 8 bytes.
+        for width in [0u8, 9] {
+            assert_eq!(
+                DataLayout::parse(&v4_chunked_msg(width, &[4, 8]), 8, 8).unwrap_err(),
+                FormatError::InvalidChunkDimensions(
+                    "encoded chunk dimension size is too large".into()
+                )
+            );
+        }
+        // A dimension past u32 cannot be represented and is refused, not
+        // truncated.
+        assert!(matches!(
+            DataLayout::parse(&v4_chunked_msg(5, &[1 << 32, 8]), 8, 8),
+            Err(FormatError::InvalidChunkDimensions(m)) if m.contains("2^32")
+        ));
+    }
+
+    #[test]
+    fn v1v2_rejects_bad_class_dimensionality_and_truncation() {
+        assert_eq!(
+            DataLayout::parse(&v1v2_header(1, 1, 3), 8, 8).unwrap_err(),
+            FormatError::InvalidLayoutClass(3)
+        );
+        assert!(matches!(
+            DataLayout::parse(&v1v2_header(2, 34, 1), 8, 8).unwrap_err(),
+            FormatError::Overflow(_)
+        ));
+        // Chunked, dims cut short.
+        let mut buf = v1v2_header(1, 2, 2);
+        buf.extend_from_slice(&0x10u64.to_le_bytes());
+        buf.extend_from_slice(&7u32.to_le_bytes());
+        assert!(matches!(
+            DataLayout::parse(&buf, 8, 8).unwrap_err(),
+            FormatError::UnexpectedEof { .. }
+        ));
+        // Compact, raw data shorter than its declared size.
+        let mut buf = v1v2_header(2, 1, 0);
+        buf.extend_from_slice(&4u32.to_le_bytes());
+        buf.extend_from_slice(&100u32.to_le_bytes());
+        buf.extend_from_slice(&[0; 4]);
+        assert!(matches!(
+            DataLayout::parse(&buf, 8, 8).unwrap_err(),
+            FormatError::UnexpectedEof { .. }
+        ));
+    }
+
     #[test]
     fn v3_compact() {
         let mut buf = vec![3u8, 0]; // version=3, class=0 (compact)
@@ -602,6 +962,7 @@ mod tests {
                 chunk_index_type: None,
                 single_chunk_filtered_size: None,
                 single_chunk_filter_mask: None,
+                dont_filter_partial_edge_chunks: false,
             }
         );
     }
@@ -679,10 +1040,35 @@ mod tests {
                 chunk_index_type: Some(1),
                 single_chunk_filtered_size: None,
                 single_chunk_filter_mask: None,
+                dont_filter_partial_edge_chunks: false,
             }
         );
     }
 
+    #[test]
+    fn v4_chunked_dont_filter_partial_edge_chunks_flag() {
+        let mut buf = vec![4u8, 2]; // version=4, class=2
+        buf.push(0x01); // flags bit 0 = don't filter partial edge chunks
+        buf.push(2); // dimensionality=2
+        buf.push(4); // dim_size_encoded_length=4
+        buf.extend_from_slice(&5u32.to_le_bytes());
+        buf.extend_from_slice(&4u32.to_le_bytes());
+        buf.push(3); // Fixed Array
+        buf.push(10); // max_dblk_page_nelmts_bits
+        buf.extend_from_slice(&0x3000u64.to_le_bytes());
+        match DataLayout::parse(&buf, 8, 8).unwrap() {
+            DataLayout::Chunked {
+                dont_filter_partial_edge_chunks,
+                btree_address,
+                ..
+            } => {
+                assert!(dont_filter_partial_edge_chunks);
+                assert_eq!(btree_address, Some(0x3000));
+            }
+            other => panic!("expected Chunked, got {other:?}"),
+        }
+    }
+
     #[test]
     fn v4_chunked_single_chunk_with_filters() {
         let mut buf = vec![4u8, 2]; // version=4, class=2
@@ -705,6 +1091,7 @@ mod tests {
                 chunk_index_type: Some(1),
                 single_chunk_filtered_size: Some(1024),
                 single_chunk_filter_mask: Some(0),
+                dont_filter_partial_edge_chunks: false,
             }
         );
     }
@@ -815,6 +1202,62 @@ mod tests {
         assert_eq!(v1.iter_linear_1d(8).unwrap(), vec![4, 5, 6, 7]);
     }
 
+    #[test]
+    fn parse_vds_mappings_v1_shared_names() {
+        // Written by HDF5 2.0 (h5py, libver=("v200", "v200")) for three
+        // mappings from `a_rather_long_source_file.h5:a_rather_long_dataset_name`
+        // and one from the same file: the entries carry flags 0x00, 0x03, 0x03
+        // and 0x06, so names after the first are stored as entry indices.
+        let blob: &[u8] = &[
+            0x01, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x61, 0x5f, 0x72, 0x61,
+            0x74, 0x68, 0x65, 0x72, 0x5f, 0x6c, 0x6f, 0x6e, 0x67, 0x5f, 0x73, 0x6f, 0x75, 0x72,
+            0x63, 0x65, 0x5f, 0x66, 0x69, 0x6c, 0x65, 0x2e, 0x68, 0x35, 0x00, 0x61, 0x5f, 0x72,
+            0x61, 0x74, 0x68, 0x65, 0x72, 0x5f, 0x6c, 0x6f, 0x6e, 0x67, 0x5f, 0x64, 0x61, 0x74,
+            0x61, 0x73, 0x65, 0x74, 0x5f, 0x6e, 0x61, 0x6d, 0x65, 0x00, 0x02, 0x00, 0x00, 0x00,
+            0x03, 0x00, 0x00, 0x00, 0x01, 0x02, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00,
+            0x01, 0x00, 0x04, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x02,
+            0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x01, 0x00, 0x04, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03,
+            0x00, 0x00, 0x00, 0x01, 0x02, 0x01, 0x00, 0x00, 0x00, 0x04, 0x00, 0x01, 0x00, 0x01,
+            0x00, 0x04, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x02, 0x02,
+            0x00, 0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
+            0x00, 0x01, 0x00, 0x04, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00,
+            0x00, 0x00, 0x01, 0x02, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x01, 0x00, 0x01, 0x00,
+            0x04, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x02, 0x02, 0x00,
+            0x00, 0x00, 0x02, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00,
+            0x01, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02,
+            0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x01, 0x02, 0x01, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x01, 0x00, 0x01, 0x00, 0x04, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00,
+            0x00, 0x01, 0x02, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01,
+            0x00, 0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0x04, 0x00, 0x8e, 0xa7, 0xea, 0x7a,
+        ];
+        let mappings = parse_vds_mappings(blob, 8).unwrap();
+        let names: Vec<(&str, &str)> = mappings
+            .iter()
+            .map(|m| (m.source_file.as_str(), m.source_dataset.as_str()))
+            .collect();
+        let (file, dset) = ("a_rather_long_source_file.h5", "a_rather_long_dataset_name");
+        assert_eq!(
+            names,
+            vec![(file, dset), (file, dset), (file, dset), (".", dset)]
+        );
+    }
+
+    #[test]
+    fn parse_vds_mappings_v1_forward_reference_is_error() {
+        // Entry 0 claiming to share entry 0's file name must not index past
+        // the entries decoded so far.
+        let mut blob = vec![0x01u8, 1, 0, 0, 0, 0, 0, 0, 0, 0x01];
+        blob.extend_from_slice(&[0u8; 8]);
+        blob.extend_from_slice(b"d\0");
+        assert!(parse_vds_mappings(&blob, 8).is_err());
+        // Unknown flag bits are refused.
+        let blob = [0x01u8, 1, 0, 0, 0, 0, 0, 0, 0, 0x08, b'd', 0];
+        assert!(parse_vds_mappings(&blob, 8).is_err());
+    }
+
     #[test]
     fn parse_vds_mappings_external_v0() {
         // Block version 0 with an explicit (external) source file name.

crates/clawhdf5-format/src/ea_writer.rs

@@ -7,7 +7,9 @@ extern crate alloc;
 use alloc::{vec, vec::Vec};
 
 use crate::checksum::jenkins_lookup3;
-use crate::chunked_write::WrittenChunk;
+use crate::chunked_write::{
+    WrittenChunk, filtered_chunk_size_len, push_addr, push_index_element, push_v4_chunk_dims,
+};
 
 /// Serialize a v4 Extensible Array layout message.
 pub(crate) fn serialize_v4_extensible_array(
@@ -24,45 +26,17 @@ pub(crate) fn serialize_v4_extensible_array(
     let ndims = chunk_dims.len() as u8 + 1;
     buf.push(ndims);
 
-    let max_dim = chunk_dims
-        .iter()
-        .map(|&d| d as u64)
-        .chain(core::iter::once(element_size as u64))
-        .max()
-        .unwrap_or(1);
-    let dim_encoded_len: u8 = if max_dim <= 0xFF {
-        1
-    } else if max_dim <= 0xFFFF {
-        2
-    } else {
-        4
-    };
-    buf.push(dim_encoded_len);
-
-    for &d in chunk_dims {
-        match dim_encoded_len {
-            1 => buf.push(d as u8),
-            2 => buf.extend_from_slice(&(d as u16).to_le_bytes()),
-            4 => buf.extend_from_slice(&d.to_le_bytes()),
-            _ => unreachable!("unexpected dim_encoded_len: {dim_encoded_len}"),
-        }
-    }
-    match dim_encoded_len {
-        1 => buf.push(element_size as u8),
-        2 => buf.extend_from_slice(&(element_size as u16).to_le_bytes()),
-        4 => buf.extend_from_slice(&element_size.to_le_bytes()),
-        _ => unreachable!("unexpected dim_encoded_len: {dim_encoded_len}"),
-    }
+    push_v4_chunk_dims(&mut buf, chunk_dims, element_size);
 
     // chunk index type = 4 (Extensible Array)
     buf.push(4);
 
     // EA creation parameters (must match AEHD and HDF5 C library defaults)
-    buf.push(32); // max_nelmts_bits
-    buf.push(4); // idx_blk_elmts
-    buf.push(4); // super_blk_min_data_ptrs
-    buf.push(16); // data_blk_min_elmts
-    buf.push(10); // max_dblk_page_nelmts_bits
+    buf.push(MAX_NELMTS_BITS);
+    buf.push(IDX_BLK_ELMTS);
+    buf.push(SUP_BLK_MIN_DATA_PTRS);
+    buf.push(DATA_BLK_MIN_ELMTS);
+    buf.push(MAX_DBLK_PAGE_NELMTS_BITS);
 
     // EA header address
     match offset_size {
@@ -74,304 +48,281 @@ pub(crate) fn serialize_v4_extensible_array(
     buf
 }
 
+// EA creation parameters — the HDF5 library's defaults for chunk indexes
+// (`H5D_EARRAY_*`); the layout message above and the header must agree.
+const MAX_NELMTS_BITS: u8 = 32;
+const IDX_BLK_ELMTS: u8 = 4;
+const SUP_BLK_MIN_DATA_PTRS: u8 = 4;
+const DATA_BLK_MIN_ELMTS: u8 = 16;
+const MAX_DBLK_PAGE_NELMTS_BITS: u8 = 10;
+
+/// One data block of the array: its first element (relative to the end of
+/// the index block's own elements), element count, and address when it is
+/// allocated.
+struct DataBlock {
+    start: usize,
+    nelmts: usize,
+    addr: Option<u64>,
+}
+
 /// Build a complete Extensible Array at a known absolute address.
 ///
-/// For simplicity, we put all elements inline in the index block when the
-/// number of chunks is small (up to idx_blk_elmts), otherwise use inline +
-/// direct data blocks.
+/// `slots[i]` is the element at linear index `i` (see `chunk_grid`); `None`
+/// marks an unallocated chunk. The first `IDX_BLK_ELMTS` elements live in
+/// the index block, the rest in data blocks grouped by super block level
+/// exactly as `H5EA__hdr_init` sizes them: level `u` has `2^(u/2)` data
+/// blocks of `DATA_BLK_MIN_ELMTS * 2^ceil(u/2)` elements. The data blocks of
+/// the first levels are addressed straight from the index block; later
+/// levels go through a super block (EASB). Data blocks larger than a page
+/// (`2^MAX_DBLK_PAGE_NELMTS_BITS` elements) are paged, with their page-init
+/// bits kept in the owning super block. Only blocks holding a defined element
+/// are allocated; the rest keep the undefined address, as in a file the
+/// library wrote.
 pub fn build_extensible_array_at(
-    chunks: &[WrittenChunk],
+    slots: &[Option<WrittenChunk>],
     offset_size: u8,
     length_size: u8,
     has_filters: bool,
     ea_base_address: u64,
 ) -> Vec<u8> {
     let os = offset_size as usize;
-    let num_elements = chunks.len();
-
-    // Compute element encoding size (same logic as Fixed Array)
-    let chunk_size_bytes: usize = if has_filters {
-        let max_raw = chunks.iter().map(|c| c.raw_size).max().unwrap_or(1);
-        let log2_val = if max_raw <= 1 {
-            0
-        } else {
-            63 - max_raw.leading_zeros()
-        };
-        let len = 1 + ((log2_val + 8) / 8) as usize;
-        len.min(8)
-    } else {
-        0
-    };
-
-    let elem_size = if has_filters {
-        os + chunk_size_bytes + 4
-    } else {
-        os
-    };
-
+    let chunk_size_bytes = has_filters.then(|| filtered_chunk_size_len(slots));
+    let elem_size = os + chunk_size_bytes.map_or(0, |n| n + 4);
     let client_id: u8 = if has_filters { 1 } else { 0 };
+    let arr_off_size = (MAX_NELMTS_BITS as usize).div_ceil(8);
+    let page_nelmts = 1usize << MAX_DBLK_PAGE_NELMTS_BITS;
+    let idx_blk = IDX_BLK_ELMTS as usize;
 
-    // EA creation parameters — must match HDF5 C library defaults exactly
-    let max_nelmts_bits: u8 = 32;
-    let idx_blk_elmts: u8 = 4;
-    let min_dblk_nelmts: u8 = 16;
-    let super_blk_min_nelmts: u8 = 4;
-    let max_dblk_nelmts_bits: u8 = 10;
+    // Elements past the last defined one are never realised
+    // (`max_idx_set` is one past the highest index ever set).
+    let max_idx_set = slots.iter().rposition(Option::is_some).map_or(0, |i| i + 1);
+    let slots = &slots[..max_idx_set];
+    let defined_in = |start: usize, n: usize| -> bool {
+        let lo = idx_blk.saturating_add(start).min(slots.len());
+        let hi = idx_blk
+            .saturating_add(start)
+            .saturating_add(n)
+            .min(slots.len());
+        slots[lo..hi].iter().any(Option::is_some)
+    };
 
-    // EAHD size: fixed(12) + 6 stats(6*length_size) + addr(offset_size) + checksum(4)
+    // Super block levels: (ndblks, dblk_nelmts, first element).
+    let log2_dmin = (DATA_BLK_MIN_ELMTS as u32).trailing_zeros() as usize;
+    let nsblks = 1 + MAX_NELMTS_BITS as usize - log2_dmin;
+    let ndblk_addrs = 2 * (SUP_BLK_MIN_DATA_PTRS as usize - 1);
+    let mut levels: Vec<(usize, usize, usize)> = Vec::with_capacity(nsblks);
+    let mut start = 0usize;
+    for u in 0..nsblks {
+        let ndblks = 1usize << (u / 2);
+        let nelmts = (DATA_BLK_MIN_ELMTS as usize) << u.div_ceil(2);
+        levels.push((ndblks, nelmts, start));
+        // Saturate: on 32-bit targets the last levels only need to compare
+        // as "beyond the end".
+        start = start.saturating_add(ndblks.saturating_mul(nelmts));
+    }
+    // Levels whose data blocks the index block addresses directly.
+    let mut direct_levels = 0;
+    let mut n = 0;
+    while n < ndblk_addrs {
+        n += levels[direct_levels].0;
+        direct_levels += 1;
+    }
+    let nsblk_addrs = nsblks - direct_levels;
+
+    let dblk_size = |nelmts: usize| -> usize {
+        let prefix = 4 + 1 + 1 + os + arr_off_size + 4;
+        if nelmts > page_nelmts {
+            prefix + (nelmts / page_nelmts) * (page_nelmts * elem_size + 4)
+        } else {
+            prefix + nelmts * elem_size
+        }
+    };
+    let sblk_bitmap_len = |ndblks: usize, nelmts: usize| -> usize {
+        if nelmts > page_nelmts {
+            ndblks * (nelmts / page_nelmts).div_ceil(8)
+        } else {
+            0
+        }
+    };
+
+    // Plan addresses: header, index block, the direct data blocks, then each
+    // allocated super block followed by its allocated data blocks.
     let aehd_size = 4 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 1 + 6 * length_size as usize + os + 4;
     let aeib_address = ea_base_address + aehd_size as u64;
+    let aeib_size = 4 + 1 + 1 + os + idx_blk * elem_size + ndblk_addrs * os + nsblk_addrs * os + 4;
+    let mut cursor = aeib_address + aeib_size as u64;
 
-    // Determine how many elements go inline vs data blocks
-    let n_inline = (idx_blk_elmts as usize).min(num_elements);
-    let remaining_after_inline = num_elements.saturating_sub(n_inline);
+    let mut ndata_blks = 0u64;
+    let mut data_blk_size = 0u64;
+    let mut nsuper_blks = 0u64;
+    let mut super_blk_size = 0u64;
+    let mut realized = idx_blk as u64;
 
-    // Compute super block layout per HDF5 spec
-    let sblk_min = super_blk_min_nelmts as usize;
-    let log2_dblk_min = if min_dblk_nelmts <= 1 {
-        0
-    } else {
-        (min_dblk_nelmts as u32).trailing_zeros() as usize
-    };
-    let nsblks = (max_nelmts_bits as usize).saturating_sub(log2_dblk_min) + 1;
-
-    // Direct data block addresses (from super blocks 0..sblk_min-1)
-    let mut dblk_sizes: Vec<usize> = Vec::new();
-    for sblk_idx in 0..sblk_min.min(nsblks) {
-        let ndblks = 1usize << (sblk_idx / 2);
-        let dblk_nelmts = (min_dblk_nelmts as usize) * (1 << sblk_idx.div_ceil(2));
-        for _ in 0..ndblks {
-            dblk_sizes.push(dblk_nelmts);
+    let mut plan_dblk = |cursor: &mut u64, start: usize, nelmts: usize| -> DataBlock {
+        let addr = defined_in(start, nelmts).then(|| {
+            let a = *cursor;
+            let size = dblk_size(nelmts) as u64;
+            *cursor += size;
+            ndata_blks += 1;
+            data_blk_size += size;
+            realized += nelmts as u64;
+            a
+        });
+        DataBlock {
+            start,
+            nelmts,
+            addr,
         }
-    }
-    let n_direct_dblks = dblk_sizes.len();
-
-    // Super block addresses (for super blocks sblk_min..nsblks-1)
-    let n_sblk_addrs = nsblks.saturating_sub(sblk_min);
-
-    // EAIB size
-    let aeib_size = 4
-        + 1
-        + 1
-        + os
-        + idx_blk_elmts as usize * elem_size
-        + n_direct_dblks * os
-        + n_sblk_addrs * os
-        + 4;
-
-    // Build AEHD
-    let mut aehd = Vec::with_capacity(aehd_size);
-    aehd.extend_from_slice(b"EAHD");
-    aehd.push(0); // version
-    aehd.push(client_id);
-    aehd.push(elem_size as u8);
-    aehd.push(max_nelmts_bits);
-    aehd.push(idx_blk_elmts);
-    aehd.push(min_dblk_nelmts);
-    aehd.push(super_blk_min_nelmts);
-    aehd.push(max_dblk_nelmts_bits);
-
-    // Count data blocks that will have chunks
-    let n_active_dblks: u64 = if remaining_after_inline > 0 {
-        let mut count = 0u64;
-        let mut ci = n_inline;
-        for &sz in &dblk_sizes {
-            if ci < num_elements {
-                count += 1;
-                ci += sz;
-            }
-        }
-        count
-    } else {
-        0
-    };
-    let blk_off_size = (max_nelmts_bits as usize).div_ceil(8);
-    let aedb_header_overhead = 4 + 1 + 1 + os + blk_off_size + 4;
-    let data_blk_total_size: u64 = if remaining_after_inline > 0 {
-        let mut total = 0u64;
-        let mut ci = n_inline;
-        for &sz in &dblk_sizes {
-            if ci < num_elements {
-                total += (aedb_header_overhead + sz * elem_size) as u64;
-                ci += sz;
-            }
-        }
-        total
-    } else {
-        0
-    };
-    let max_idx_set: u64 = if remaining_after_inline > 0 {
-        let mut max_set = idx_blk_elmts as u64;
-        let mut ci = n_inline;
-        for &sz in &dblk_sizes {
-            if ci < num_elements {
-                max_set += sz as u64;
-                ci += sz;
-            }
-        }
-        max_set
-    } else {
-        idx_blk_elmts as u64
     };
 
+    let mut direct: Vec<DataBlock> = Vec::with_capacity(ndblk_addrs);
+    for &(ndblks, nelmts, first) in &levels[..direct_levels] {
+        for k in 0..ndblks {
+            direct.push(plan_dblk(&mut cursor, first + k * nelmts, nelmts));
+        }
+    }
+    // (super block address, level, its data blocks)
+    let mut supers: Vec<(Option<u64>, usize, Vec<DataBlock>)> = Vec::with_capacity(nsblk_addrs);
+    for (u, &(ndblks, nelmts, first)) in levels.iter().enumerate().skip(direct_levels) {
+        if !defined_in(first, ndblks.saturating_mul(nelmts)) {
+            supers.push((None, u, Vec::new()));
+            continue;
+        }
+        let sb_size =
+            4 + 1 + 1 + os + arr_off_size + sblk_bitmap_len(ndblks, nelmts) + ndblks * os + 4;
+        let sb_addr = cursor;
+        cursor += sb_size as u64;
+        nsuper_blks += 1;
+        super_blk_size += sb_size as u64;
+        let dblks = (0..ndblks)
+            .map(|k| plan_dblk(&mut cursor, first + k * nelmts, nelmts))
+            .collect();
+        supers.push((Some(sb_addr), u, dblks));
+    }
+
+    let slot = |i: usize| slots.get(i).and_then(Option::as_ref);
     let write_length = |buf: &mut Vec<u8>, val: u64| match length_size {
         4 => buf.extend_from_slice(&(val as u32).to_le_bytes()),
         _ => buf.extend_from_slice(&val.to_le_bytes()),
     };
-    let write_addr = |buf: &mut Vec<u8>, val: u64| match offset_size {
-        4 => buf.extend_from_slice(&(val as u32).to_le_bytes()),
-        _ => buf.extend_from_slice(&val.to_le_bytes()),
+    let write_addr_opt = |buf: &mut Vec<u8>, addr: Option<u64>| match addr {
+        Some(a) => push_addr(buf, a, offset_size),
+        None => buf.extend(core::iter::repeat_n(0xFF, os)),
+    };
+    let block_prefix = |buf: &mut Vec<u8>, sig: &[u8; 4], block_off: usize| {
+        buf.extend_from_slice(sig);
+        buf.push(0); // version
+        buf.push(client_id);
+        push_addr(buf, ea_base_address, offset_size);
+        buf.extend_from_slice(&(block_off as u64).to_le_bytes()[..arr_off_size]);
+    };
+    // Serialise one data block (paged or not) onto `out`.
+    let write_dblk = |out: &mut Vec<u8>, db: &DataBlock| {
+        let at = out.len();
+        block_prefix(out, b"EADB", db.start);
+        let first = idx_blk + db.start;
+        if db.nelmts > page_nelmts {
+            // Paged: the prefix carries only its own checksum; each page
+            // follows with one of its own.
+            let sum = jenkins_lookup3(&out[at..]);
+            out.extend_from_slice(&sum.to_le_bytes());
+            for p in 0..db.nelmts / page_nelmts {
+                let page_at = out.len();
+                for e in 0..page_nelmts {
+                    let i = first + p * page_nelmts + e;
+                    push_index_element(out, slot(i), offset_size, chunk_size_bytes);
+                }
+                let sum = jenkins_lookup3(&out[page_at..]);
+                out.extend_from_slice(&sum.to_le_bytes());
+            }
+        } else {
+            for i in first..first + db.nelmts {
+                push_index_element(out, slot(i), offset_size, chunk_size_bytes);
+            }
+            let sum = jenkins_lookup3(&out[at..]);
+            out.extend_from_slice(&sum.to_le_bytes());
+        }
+        debug_assert_eq!(out.len() - at, dblk_size(db.nelmts));
     };
 
-    write_length(&mut aehd, 0);
-    write_length(&mut aehd, 0);
-    write_length(&mut aehd, n_active_dblks);
-    write_length(&mut aehd, data_blk_total_size);
-    write_length(&mut aehd, num_elements as u64);
-    write_length(&mut aehd, max_idx_set);
+    // Header (EAHD). The six statistics are, in order: super blocks, their
+    // bytes, data blocks, their bytes, max index set, elements realised.
+    let mut out = Vec::with_capacity((cursor - ea_base_address) as usize);
+    out.extend_from_slice(b"EAHD");
+    out.push(0); // version
+    out.push(client_id);
+    out.push(elem_size as u8);
+    out.push(MAX_NELMTS_BITS);
+    out.push(IDX_BLK_ELMTS);
+    out.push(DATA_BLK_MIN_ELMTS);
+    out.push(SUP_BLK_MIN_DATA_PTRS);
+    out.push(MAX_DBLK_PAGE_NELMTS_BITS);
+    write_length(&mut out, nsuper_blks);
+    write_length(&mut out, super_blk_size);
+    write_length(&mut out, ndata_blks);
+    write_length(&mut out, data_blk_size);
+    write_length(&mut out, max_idx_set as u64);
+    write_length(&mut out, realized);
+    push_addr(&mut out, aeib_address, offset_size);
+    let sum = jenkins_lookup3(&out);
+    out.extend_from_slice(&sum.to_le_bytes());
+    debug_assert_eq!(out.len(), aehd_size);
 
-    write_addr(&mut aehd, aeib_address);
-
-    let aehd_checksum = jenkins_lookup3(&aehd);
-    aehd.extend_from_slice(&aehd_checksum.to_le_bytes());
-    debug_assert_eq!(aehd.len(), aehd_size);
-
-    // Build AEIB
-    let mut aeib = Vec::with_capacity(aeib_size);
-    aeib.extend_from_slice(b"EAIB");
-    aeib.push(0);
-    aeib.push(client_id);
-
-    match offset_size {
-        4 => aeib.extend_from_slice(&(ea_base_address as u32).to_le_bytes()),
-        8 => aeib.extend_from_slice(&ea_base_address.to_le_bytes()),
-        _ => aeib.extend_from_slice(&ea_base_address.to_le_bytes()),
+    // Index block (EAIB): inline elements, data block and super block
+    // addresses.
+    let ib_start = out.len();
+    out.extend_from_slice(b"EAIB");
+    out.push(0);
+    out.push(client_id);
+    push_addr(&mut out, ea_base_address, offset_size);
+    for i in 0..idx_blk {
+        push_index_element(&mut out, slot(i), offset_size, chunk_size_bytes);
     }
-
-    // Inline elements
-    #[allow(clippy::needless_range_loop)]
-    for i in 0..idx_blk_elmts as usize {
-        if i < n_inline {
-            write_chunk_element(
-                &mut aeib,
-                &chunks[i],
-                offset_size,
-                has_filters,
-                chunk_size_bytes,
-            );
-        } else {
-            write_undefined_element(&mut aeib, offset_size, has_filters, chunk_size_bytes);
+    for db in &direct {
+        write_addr_opt(&mut out, db.addr);
     }
+    for (sb_addr, _, _) in &supers {
+        write_addr_opt(&mut out, *sb_addr);
     }
+    let sum = jenkins_lookup3(&out[ib_start..]);
+    out.extend_from_slice(&sum.to_le_bytes());
+    debug_assert_eq!(out.len() - ib_start, aeib_size);
 
-    // Data block addresses + build data blocks
-    let mut data_blocks_buf = Vec::new();
-    let dblks_base = aeib_address + aeib_size as u64;
-    let mut dblk_cursor = dblks_base;
-    let mut chunk_idx = n_inline;
-
-    for &nelmts in &dblk_sizes {
-        if chunk_idx >= num_elements {
-            match offset_size {
-                4 => aeib.extend_from_slice(&u32::MAX.to_le_bytes()),
-                8 => aeib.extend_from_slice(&u64::MAX.to_le_bytes()),
-                _ => aeib.extend_from_slice(&u64::MAX.to_le_bytes()),
+    for db in direct.iter().filter(|d| d.addr.is_some()) {
+        write_dblk(&mut out, db);
     }
+    for (sb_addr, u, dblks) in &supers {
+        if sb_addr.is_none() {
             continue;
         }
-
-        match offset_size {
-            4 => aeib.extend_from_slice(&(dblk_cursor as u32).to_le_bytes()),
-            8 => aeib.extend_from_slice(&dblk_cursor.to_le_bytes()),
-            _ => aeib.extend_from_slice(&dblk_cursor.to_le_bytes()),
-        }
-
-        // Build EADB
-        let mut aedb = Vec::new();
-        aedb.extend_from_slice(b"EADB");
-        aedb.push(0);
-        aedb.push(client_id);
-        match offset_size {
-            4 => aedb.extend_from_slice(&(ea_base_address as u32).to_le_bytes()),
-            8 => aedb.extend_from_slice(&ea_base_address.to_le_bytes()),
-            _ => aedb.extend_from_slice(&ea_base_address.to_le_bytes()),
-        }
-
-        let blk_off_size = (max_nelmts_bits as usize).div_ceil(8);
-        let blk_off_val = (chunk_idx - n_inline) as u64;
-        aedb.extend_from_slice(&blk_off_val.to_le_bytes()[..blk_off_size]);
-
-        for slot in 0..nelmts {
-            if chunk_idx + slot < num_elements {
-                write_chunk_element(
-                    &mut aedb,
-                    &chunks[chunk_idx + slot],
-                    offset_size,
-                    has_filters,
-                    chunk_size_bytes,
-                );
-            } else {
-                write_undefined_element(&mut aedb, offset_size, has_filters, chunk_size_bytes);
+        let (ndblks, nelmts, first) = levels[*u];
+        let sb_start = out.len();
+        block_prefix(&mut out, b"EASB", first);
+        if nelmts > page_nelmts {
+            // Page-init bits, `npages` per data block, packed MSB-first
+            // (`H5VM_bit_set`): every page of an allocated data block is
+            // written.
+            let npages = nelmts / page_nelmts;
+            let mut bitmap = vec![0u8; sblk_bitmap_len(ndblks, nelmts)];
+            for (k, db) in dblks.iter().enumerate() {
+                if db.addr.is_some() {
+                    for p in 0..npages {
+                        let bit = k * npages + p;
+                        bitmap[bit / 8] |= 0x80 >> (bit % 8);
                     }
                 }
-
-        let aedb_checksum = jenkins_lookup3(&aedb);
-        aedb.extend_from_slice(&aedb_checksum.to_le_bytes());
-
-        dblk_cursor += aedb.len() as u64;
-        data_blocks_buf.extend_from_slice(&aedb);
-        chunk_idx += nelmts;
             }
-
-    // Super block addresses (all undefined)
-    for _ in 0..n_sblk_addrs {
-        match offset_size {
-            4 => aeib.extend_from_slice(&u32::MAX.to_le_bytes()),
-            8 => aeib.extend_from_slice(&u64::MAX.to_le_bytes()),
-            _ => aeib.extend_from_slice(&u64::MAX.to_le_bytes()),
+            out.extend_from_slice(&bitmap);
+        }
+        for db in dblks {
+            write_addr_opt(&mut out, db.addr);
+        }
+        let sum = jenkins_lookup3(&out[sb_start..]);
+        out.extend_from_slice(&sum.to_le_bytes());
+        for db in dblks.iter().filter(|d| d.addr.is_some()) {
+            write_dblk(&mut out, db);
         }
     }
-
-    let aeib_checksum = jenkins_lookup3(&aeib);
-    aeib.extend_from_slice(&aeib_checksum.to_le_bytes());
-    debug_assert_eq!(aeib.len(), aeib_size);
-
-    let mut combined = aehd;
-    combined.extend_from_slice(&aeib);
-    combined.extend_from_slice(&data_blocks_buf);
-    combined
-}
-
-fn write_chunk_element(
-    buf: &mut Vec<u8>,
-    chunk: &WrittenChunk,
-    offset_size: u8,
-    has_filters: bool,
-    chunk_size_bytes: usize,
-) {
-    match offset_size {
-        4 => buf.extend_from_slice(&(chunk.address as u32).to_le_bytes()),
-        8 => buf.extend_from_slice(&chunk.address.to_le_bytes()),
-        _ => buf.extend_from_slice(&chunk.address.to_le_bytes()),
-    }
-    if has_filters {
-        let cs_bytes = chunk.compressed_size.to_le_bytes();
-        buf.extend_from_slice(&cs_bytes[..chunk_size_bytes]);
-        buf.extend_from_slice(&chunk.filter_mask.to_le_bytes());
-    }
-}
-
-fn write_undefined_element(
-    buf: &mut Vec<u8>,
-    offset_size: u8,
-    has_filters: bool,
-    chunk_size_bytes: usize,
-) {
-    let os = offset_size as usize;
-    // Use extend with repeat to avoid heap-allocating a temporary Vec on each call.
-    buf.extend(core::iter::repeat_n(0xFF, os));
-    if has_filters {
-        buf.extend(core::iter::repeat_n(0x00, chunk_size_bytes));
-        buf.extend_from_slice(&0u32.to_le_bytes());
-    }
+    debug_assert_eq!(out.len() as u64, cursor - ea_base_address);
+    out
 }

crates/clawhdf5-format/src/error.rs

@@ -80,6 +80,9 @@ pub enum FormatError {
     InvalidLocalHeapSignature,
     /// Invalid local heap version.
     InvalidLocalHeapVersion(u8),
+    /// A local heap's free list points outside its data segment (libhdf5:
+    /// "bad heap free list").
+    InvalidLocalHeapFreeList,
     /// Invalid B-tree v1 signature.
     InvalidBTreeSignature,
     /// Invalid B-tree node type.
@@ -117,6 +120,14 @@ pub enum FormatError {
     /// A message is marked shared but was parsed without access to the file,
     /// so the reference to the real message could not be followed.
     UnresolvedSharedMessage,
+    /// A shared-message reference points at an object header that holds no
+    /// (unshared) message of the referenced type (raw message type id).
+    SharedMessageTargetMissing(u16),
+    /// A superblock was parsed at a non-zero offset of the buffer (the file
+    /// has a user block of this many bytes). HDF5 addresses are relative to
+    /// the superblock, so the buffer must start there: see
+    /// `signature::split_user_block`.
+    UserBlockNotStripped(u64),
     /// A selection does not fit the dataset it was applied to (wrong rank, or
     /// it reaches past a dimension's extent).
     SelectionOutOfBounds(String),
@@ -190,6 +201,28 @@ pub enum FormatError {
     DuplicateDatasetName(String),
     /// Integer overflow in size computation (malformed data protection).
     Overflow(String),
+    /// An object header that libhdf5 refuses to load (the reason is
+    /// libhdf5's own error text): a misaligned or overrunning message, a
+    /// wrong message count, contradictory message flags, a message of a
+    /// class that cannot be shared flagged shareable, …
+    InvalidObjectHeader(&'static str),
+    /// A datatype message libhdf5 refuses to decode (the reason is
+    /// libhdf5's own error text): size 0, bit fields outside the type,
+    /// an empty enum name, a compound member outside its compound, …
+    InvalidDatatype(String),
+    /// A chunked layout whose chunk dimensions libhdf5 refuses: a zero
+    /// dimension, a rank that does not match the dataspace, an element size
+    /// that is not the datatype's, or a chunk of 4 GiB or more indexed by a
+    /// version-1 B-tree.
+    InvalidChunkDimensions(String),
+    /// The superblock's end-of-file address lies past the end of the file:
+    /// the file was truncated (libhdf5 refuses to open it).
+    TruncatedFile {
+        /// End of file recorded in the superblock (relative to byte 0).
+        stored_eof: u64,
+        /// The file's actual length in bytes.
+        actual_len: u64,
+    },
 }
 
 impl fmt::Display for FormatError {
@@ -270,6 +303,9 @@ impl fmt::Display for FormatError {
             FormatError::InvalidLocalHeapSignature => {
                 write!(f, "invalid local heap signature")
             }
+            FormatError::InvalidLocalHeapFreeList => {
+                write!(f, "bad local heap free list")
+            }
             FormatError::InvalidLocalHeapVersion(v) => {
                 write!(f, "invalid local heap version: {v}")
             }
@@ -339,6 +375,16 @@ impl fmt::Display for FormatError {
             FormatError::SelectionOutOfBounds(msg) => {
                 write!(f, "selection out of bounds: {msg}")
             }
+            FormatError::UserBlockNotStripped(n) => write!(
+                f,
+                "file has a {n}-byte user block: parse the bytes from the superblock on \
+                 (signature::split_user_block)"
+            ),
+            FormatError::SharedMessageTargetMissing(t) => write!(
+                f,
+                "shared message reference points at an object header with no message of type \
+                 {t:#06x}"
+            ),
             FormatError::UnresolvedSharedMessage => write!(
                 f,
                 "message is shared but no file data was available to resolve it"
@@ -382,9 +428,17 @@ impl fmt::Display for FormatError {
             FormatError::InvalidFilterPipelineVersion(v) => {
                 write!(f, "invalid filter pipeline version: {v}")
             }
-            FormatError::UnsupportedFilter(id) => {
-                write!(f, "unsupported filter: {id}")
-            }
+            FormatError::UnsupportedFilter(id) => match crate::filter_registry::known_filter(*id) {
+                Some((name, Some(feature))) => write!(
+                    f,
+                    "unsupported filter: {id} ({name}; this build lacks the `{feature}` feature)"
+                ),
+                Some((name, None)) => write!(
+                    f,
+                    "unsupported filter: {id} ({name}, not implemented by clawhdf5)"
+                ),
+                None => write!(f, "unsupported filter: {id}"),
+            },
             FormatError::FilterError(msg) => {
                 write!(f, "filter error: {msg}")
             }
@@ -421,6 +475,25 @@ impl fmt::Display for FormatError {
             FormatError::Overflow(msg) => {
                 write!(f, "integer overflow: {msg}")
             }
+            FormatError::InvalidObjectHeader(why) => {
+                write!(f, "corrupt object header: {why}")
+            }
+            FormatError::InvalidDatatype(why) => {
+                write!(f, "invalid datatype: {why}")
+            }
+            FormatError::InvalidChunkDimensions(why) => {
+                write!(f, "invalid chunk dimensions: {why}")
+            }
+            FormatError::TruncatedFile {
+                stored_eof,
+                actual_len,
+            } => {
+                write!(
+                    f,
+                    "truncated file: the superblock records end of file {stored_eof}, \
+                     but the file is {actual_len} bytes"
+                )
+            }
         }
     }
 }

crates/clawhdf5-format/src/extensible_array.rs

@@ -9,6 +9,7 @@ extern crate alloc;
 #[cfg(not(feature = "std"))]
 use alloc::{format, vec, vec::Vec};
 
+use crate::chunk_grid::ChunkGrid;
 use crate::chunked_read::ChunkInfo;
 use crate::error::FormatError;
 
@@ -203,8 +204,7 @@ fn read_element(
     offset_size: u8,
     chunk_byte_size: u64,
     linear_index: usize,
-    num_chunks_per_dim: &[u64],
-    chunk_dimensions: &[u32],
+    grid: &ChunkGrid,
 ) -> Result<(Option<ChunkInfo>, usize), FormatError> {
     let os = offset_size as usize;
 
@@ -220,7 +220,10 @@ fn read_element(
             return Ok((None, os));
         }
         let address = read_offset(data, pos, offset_size)?;
-        let offsets = index_to_chunk_offsets(linear_index, num_chunks_per_dim, chunk_dimensions);
+        // A slot beyond the current extent is ignored, as the library does.
+        let Some(offsets) = grid.offsets(linear_index as u64) else {
+            return Ok((None, os));
+        };
         Ok((
             Some(ChunkInfo {
                 chunk_size: chunk_byte_size as u32,
@@ -261,7 +264,9 @@ fn read_element(
             data[fm_off + 2],
             data[fm_off + 3],
         ]);
-        let offsets = index_to_chunk_offsets(linear_index, num_chunks_per_dim, chunk_dimensions);
+        let Some(offsets) = grid.offsets(linear_index as u64) else {
+            return Ok((None, elem_total));
+        };
         Ok((
             Some(ChunkInfo {
                 chunk_size: chunk_size as u32,
@@ -274,27 +279,6 @@ fn read_element(
     }
 }
 
-/// Convert a linear chunk index to N-dimensional chunk offsets in dataset space.
-fn index_to_chunk_offsets(
-    index: usize,
-    num_chunks_per_dim: &[u64],
-    chunk_dimensions: &[u32],
-) -> Vec<u64> {
-    let rank = num_chunks_per_dim.len();
-    let mut offsets = vec![0u64; rank];
-    let mut remaining = index as u64;
-    for d in (0..rank).rev() {
-        let nchunks = num_chunks_per_dim[d];
-        if nchunks == 0 {
-            continue;
-        }
-        let chunk_idx = remaining % nchunks;
-        remaining /= nchunks;
-        offsets[d] = chunk_idx * chunk_dimensions[d] as u64;
-    }
-    offsets
-}
-
 /// Collect elements from a data block at the given offset.
 #[allow(clippy::too_many_arguments)]
 /// Layout of super block `u`, per the HDF5 spec: the number of data blocks it
@@ -339,8 +323,7 @@ fn read_data_block_elements(
     offset_size: u8,
     chunk_byte_size: u64,
     start_index: usize,
-    num_chunks_per_dim: &[u64],
-    chunk_dimensions: &[u32],
+    grid: &ChunkGrid,
     page_init: &[u8],
     first_page: usize,
 ) -> Result<Vec<ChunkInfo>, FormatError> {
@@ -376,8 +359,7 @@ fn read_data_block_elements(
                 offset_size,
                 chunk_byte_size,
                 first_index + i,
-                num_chunks_per_dim,
-                chunk_dimensions,
+                grid,
             )?;
             if let Some(ci) = info {
                 chunks.push(ci);
@@ -449,25 +431,19 @@ pub fn read_extensible_array_chunks(
     file_data: &[u8],
     header: &ExtensibleArrayHeader,
     dataset_dims: &[u64],
+    max_dims: Option<&[u64]>,
     chunk_dimensions: &[u32],
     element_size: u32,
     offset_size: u8,
     _length_size: u8,
 ) -> Result<Vec<ChunkInfo>, FormatError> {
-    let rank = chunk_dimensions.len();
     let os = offset_size as usize;
 
-    let mut num_chunks_per_dim = Vec::with_capacity(rank);
-    for d in 0..rank {
-        let ch_dim = chunk_dimensions[d] as u64;
-        if ch_dim == 0 {
-            return Err(FormatError::ChunkedReadError(
-                "chunk dimension is zero".into(),
-            ));
-        }
-        let ds_dim = dataset_dims[d];
-        num_chunks_per_dim.push(ds_dim.div_ceil(ch_dim));
-    }
+    // Linear indexes follow the maximum dimensions, with the unlimited
+    // dimension swizzled to the slowest position (see `chunk_grid`).
+    let dims_u64: Vec<u64> = chunk_dimensions.iter().map(|&d| d as u64).collect();
+    let grid = ChunkGrid::extensible_array(dataset_dims, max_dims, &dims_u64)?;
+    let grid = &grid;
 
     let chunk_byte_size: u64 =
         chunk_dimensions.iter().map(|&d| d as u64).product::<u64>() * element_size as u64;
@@ -557,8 +533,7 @@ pub fn read_extensible_array_chunks(
             offset_size,
             chunk_byte_size,
             i,
-            &num_chunks_per_dim,
-            chunk_dimensions,
+            grid,
         )?;
         if let Some(ci) = info {
             chunks.push(ci);
@@ -594,8 +569,7 @@ pub fn read_extensible_array_chunks(
                 offset_size,
                 chunk_byte_size,
                 global_index,
-                &num_chunks_per_dim,
-                chunk_dimensions,
+                grid,
                 &[],
                 0,
             )?);
@@ -625,8 +599,7 @@ pub fn read_extensible_array_chunks(
                 offset_size,
                 chunk_byte_size,
                 global_index,
-                &num_chunks_per_dim,
-                chunk_dimensions,
+                grid,
             )?);
         }
         global_index =
@@ -653,8 +626,7 @@ fn read_super_block(
     offset_size: u8,
     chunk_byte_size: u64,
     start_index: usize,
-    num_chunks_per_dim: &[u64],
-    chunk_dimensions: &[u32],
+    grid: &ChunkGrid,
 ) -> Result<Vec<ChunkInfo>, FormatError> {
     let os = offset_size as usize;
     let sb_header_size = 4 + 1 + 1 + os + arr_off_size(header);
@@ -710,8 +682,7 @@ fn read_super_block(
                 offset_size,
                 chunk_byte_size,
                 global_idx,
-                num_chunks_per_dim,
-                chunk_dimensions,
+                grid,
                 bitmap,
                 i * npages,
             )?);
@@ -735,35 +706,18 @@ mod tests {
     }
     #[test]
     fn index_to_offsets_1d() {
-        let num_chunks = vec![5u64];
-        let chunk_dims = vec![20u32];
-        assert_eq!(index_to_chunk_offsets(0, &num_chunks, &chunk_dims), vec![0]);
-        assert_eq!(
-            index_to_chunk_offsets(1, &num_chunks, &chunk_dims),
-            vec![20]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(4, &num_chunks, &chunk_dims),
-            vec![80]
-        );
+        let g = ChunkGrid::fixed_array(&[100], None, &[20]).unwrap();
+        assert_eq!(g.offsets(0).unwrap(), vec![0]);
+        assert_eq!(g.offsets(1).unwrap(), vec![20]);
+        assert_eq!(g.offsets(4).unwrap(), vec![80]);
     }
 
     #[test]
     fn index_to_offsets_2d() {
-        let num_chunks = vec![3u64, 2];
-        let chunk_dims = vec![4u32, 3];
-        assert_eq!(
-            index_to_chunk_offsets(0, &num_chunks, &chunk_dims),
-            vec![0, 0]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(1, &num_chunks, &chunk_dims),
-            vec![0, 3]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(2, &num_chunks, &chunk_dims),
-            vec![4, 0]
-        );
+        let g = ChunkGrid::fixed_array(&[10, 6], None, &[4, 3]).unwrap();
+        assert_eq!(g.offsets(0).unwrap(), vec![0, 0]);
+        assert_eq!(g.offsets(1).unwrap(), vec![0, 3]);
+        assert_eq!(g.offsets(2).unwrap(), vec![4, 0]);
     }
 
     #[test]
@@ -830,7 +784,7 @@ mod tests {
             index_block_address: (usize::MAX - 4) as u64,
         };
         let buf = vec![0u8; 64];
-        let r = read_extensible_array_chunks(&buf, &header, &[100], &[20], 8, 8, 8);
+        let r = read_extensible_array_chunks(&buf, &header, &[100], None, &[20], 8, 8, 8);
         assert!(r.is_err());
     }
 
@@ -913,8 +867,16 @@ mod tests {
         let header = ExtensibleArrayHeader::parse(&file_data, aehd_offset, os, ls).unwrap();
         let ds_dims = vec![40u64]; // 2 chunks × 20 elements
         let chunk_dims = vec![20u32];
-        let chunks =
-            read_extensible_array_chunks(&file_data, &header, &ds_dims, &chunk_dims, 8, os, ls)
+        let chunks = read_extensible_array_chunks(
+            &file_data,
+            &header,
+            &ds_dims,
+            None,
+            &chunk_dims,
+            8,
+            os,
+            ls,
+        )
         .unwrap();
 
         assert_eq!(chunks.len(), 2);
@@ -1023,8 +985,16 @@ mod tests {
         let header = ExtensibleArrayHeader::parse(&file_data, aehd_offset, os, ls).unwrap();
         let ds_dims = vec![40u64];
         let chunk_dims = vec![10u32];
-        let chunks =
-            read_extensible_array_chunks(&file_data, &header, &ds_dims, &chunk_dims, 8, os, ls)
+        let chunks = read_extensible_array_chunks(
+            &file_data,
+            &header,
+            &ds_dims,
+            None,
+            &chunk_dims,
+            8,
+            os,
+            ls,
+        )
         .unwrap();
 
         assert_eq!(chunks.len(), 4);
@@ -1047,10 +1017,8 @@ mod tests {
     #[test]
     fn read_element_unallocated() {
         let data = vec![0xFFu8; 16];
-        let num_chunks = vec![5u64];
-        let chunk_dims = vec![10u32];
-        let (info, consumed) =
-            read_element(&data, 0, 0, 8, 8, 80, 0, &num_chunks, &chunk_dims).unwrap();
+        let grid = ChunkGrid::fixed_array(&[50], None, &[10]).unwrap();
+        let (info, consumed) = read_element(&data, 0, 0, 8, 8, 80, 0, &grid).unwrap();
         assert!(info.is_none());
         assert_eq!(consumed, 8);
     }
@@ -1069,20 +1037,9 @@ mod tests {
         // Filter mask
         data[12..16].copy_from_slice(&0u32.to_le_bytes());
 
-        let num_chunks = vec![5u64];
-        let chunk_dims = vec![10u32];
-        let (info, consumed) = read_element(
-            &data,
-            0,
-            1,
-            elem_size as u8,
-            os,
-            80,
-            2,
-            &num_chunks,
-            &chunk_dims,
-        )
-        .unwrap();
+        let grid = ChunkGrid::fixed_array(&[50], None, &[10]).unwrap();
+        let (info, consumed) =
+            read_element(&data, 0, 1, elem_size as u8, os, 80, 2, &grid).unwrap();
         let ci = info.unwrap();
         assert_eq!(ci.address, 0x2000);
         assert_eq!(ci.chunk_size, 120);

crates/clawhdf5-format/src/fill_value.rs

@@ -98,15 +98,50 @@ pub fn parse_fill_value(msg: &HeaderMessage) -> Result<Option<Vec<u8>>, FormatEr
 
 /// The fill value that applies to a dataset given its header messages. The new
 /// message wins over the old one when both are present.
+///
+/// A *shared* fill value message holds only a reference to the real message,
+/// which cannot be followed without the file: this returns
+/// [`FormatError::UnresolvedSharedMessage`] for one (it used to answer "zeros").
+/// Use [`dataset_fill_value_in`] when the file bytes are at hand.
 pub fn dataset_fill_value(messages: &[HeaderMessage]) -> Result<Option<Vec<u8>>, FormatError> {
+    fill_value_from(messages, |_| Err(FormatError::UnresolvedSharedMessage))
+}
+
+/// [`dataset_fill_value`] for a dataset in `file_data`, following a shared
+/// fill value message to where it lives: another object header, or the
+/// file's shared-message (SOHM) heap, as libhdf5 writes it when the file has
+/// a SOHM index for fill values.
+pub fn dataset_fill_value_in(
+    file_data: &[u8],
+    messages: &[HeaderMessage],
+    offset_size: u8,
+    length_size: u8,
+) -> Result<Option<Vec<u8>>, FormatError> {
+    fill_value_from(messages, |msg| {
+        crate::shared_message::message_data_with_sohm(file_data, msg, offset_size, length_size)
+            .map(|data| data.into_owned())
+    })
+}
+
+fn fill_value_from(
+    messages: &[HeaderMessage],
+    resolve_shared: impl Fn(&HeaderMessage) -> Result<Vec<u8>, FormatError>,
+) -> Result<Option<Vec<u8>>, FormatError> {
     for wanted in [MessageType::FillValue, MessageType::FillValueOld] {
         if let Some(msg) = messages.iter().find(|m| m.msg_type == wanted) {
-            if crate::shared_message::is_shared(msg.flags) {
-                // A shared fill value is legal but vanishingly rare; treat it
-                // as the default rather than misparsing the reference.
-                return Ok(None);
-            }
-            if let Some(value) = parse_fill_value(msg)? {
+            let value = if crate::shared_message::is_shared(msg.flags) {
+                let data = resolve_shared(msg)?;
+                parse_fill_value(&HeaderMessage {
+                    msg_type: msg.msg_type,
+                    size: data.len(),
+                    flags: msg.flags & !0x02,
+                    creation_order: msg.creation_order,
+                    data,
+                })?
+            } else {
+                parse_fill_value(msg)?
+            };
+            if let Some(value) = value {
                 return Ok(Some(value));
             }
         }
@@ -174,7 +209,7 @@ pub fn read_full_with_fill<E: From<FormatError>>(
     {
         return Err(FormatError::ExternalDataFilesUnsupported.into());
     }
-    let fill = dataset_fill_value(messages)?;
+    let fill = dataset_fill_value_in(file_data, messages, offset_size, length_size)?;
     if !has_storage(layout) {
         return Ok(filled_dataset(dataspace, elem_size, fill.as_deref())?);
     }

crates/clawhdf5-format/src/filter_pipeline.rs

@@ -19,8 +19,35 @@ pub const FILTER_SCALEOFFSET: u16 = 6;
 pub const FILTER_LZ4: u16 = 32004;
 /// Zstandard compression.
 pub const FILTER_ZSTD: u16 = 32015;
-/// Pcodec lossless numerical codec (clawhdf5 internal; not yet HDF5-registered).
-pub const FILTER_PCODEC: u16 = 32023;
+/// bzip2 (registered by PyTables; hdf5plugin's `BZip2`).
+pub const FILTER_BZIP2: u16 = 307;
+/// LZF — h5py's built-in `compression="lzf"`.
+pub const FILTER_LZF: u16 = 32000;
+/// Blosc 1 (hdf5-blosc; hdf5plugin's `Blosc`).
+pub const FILTER_BLOSC: u16 = 32001;
+/// Bitshuffle, optionally with LZ4 or Zstandard (hdf5plugin's `Bitshuffle`).
+pub const FILTER_BITSHUFFLE: u16 = 32008;
+/// ZFP lossy floating-point compression (hdf5plugin's `Zfp`). Not supported.
+pub const FILTER_ZFP: u16 = 32013;
+/// Blosc 2 (hdf5plugin's `Blosc2`).
+pub const FILTER_BLOSC2: u16 = 32026;
+/// Pcodec lossless numerical codec — a **private, unregistered** clawhdf5
+/// filter. Pcodec has no ID in the HDF Group's filter registry (checked
+/// 2026-09-25, `hdf5_plugins/docs/RegisteredFilterPlugins.md`), so it uses an
+/// ID from the registry's testing/private range (256–511). No libhdf5 plugin
+/// decodes it: h5py/libhdf5 report the filter as unavailable. Only clawhdf5
+/// (with the `pcodec` feature) reads these datasets.
+pub const FILTER_PCODEC: u16 = 480;
+/// Filter name written with [`FILTER_PCODEC`].
+pub const FILTER_PCODEC_NAME: &str = "pcodec (clawhdf5 private)";
+/// The ID clawhdf5 up to 2.7.0 wrote pcodec under. It is registered to
+/// Granular BitRound (GBR), whose decode is a pass-through, so libhdf5 with
+/// that plugin would have returned the compressed bytes as data. Read as
+/// pcodec only when the filter is named exactly [`FILTER_PCODEC_LEGACY_NAME`],
+/// the name those versions wrote; never written.
+pub const FILTER_PCODEC_LEGACY: u16 = 32023;
+/// The filter name clawhdf5 up to 2.7.0 wrote with [`FILTER_PCODEC_LEGACY`].
+pub const FILTER_PCODEC_LEGACY_NAME: &str = "pcodec";
 
 /// Description of a single filter in a pipeline.
 #[derive(Debug, Clone, PartialEq)]

crates/clawhdf5-format/src/filter_registry.rs

@@ -0,0 +1,477 @@
+//! Filter registry: every filter is looked up here by its HDF5 filter ID.
+//!
+//! Two tiers:
+//!
+//! * **Built-in filters** — a static table of the filters compiled into this
+//!   build: the HDF5 standard filters (deflate, shuffle, Fletcher32, szip,
+//!   N-Bit, scale-offset) and the plugin filters whose cargo features are
+//!   enabled (LZ4, Zstandard, pcodec, LZF, bitshuffle, bzip2, blosc).
+//!   [`builtin_filters`] lists them.
+//! * **Registered filters** (`std` only) — codecs the application supplies
+//!   for any other ID with [`register_filter`] (a [`FilterCodec`], or just a
+//!   decoding closure). A registered codec cannot shadow a built-in one,
+//!   except under 32023: that ID belongs to Granular BitRound, and the
+//!   built-in entry there only reads the pcodec chunks clawhdf5 <= 2.7.0
+//!   wrote (filter name `"pcodec"`), so a codec registered for 32023 handles
+//!   every other chunk with that ID, and writes.
+//!
+//! An ID in neither tier fails with [`FormatError::UnsupportedFilter`], as it
+//! always has.
+//!
+//! ```
+//! # #[cfg(feature = "std")] {
+//! use clawhdf5_format::filter_registry::{self, FilterContext};
+//! use clawhdf5_format::error::FormatError;
+//!
+//! // A toy filter in the private-use range: every byte XORed with 0x5A.
+//! filter_registry::register_filter(300, |input: &[u8], _ctx: &FilterContext<'_>| {
+//!     Ok::<_, FormatError>(input.iter().map(|b| b ^ 0x5A).collect())
+//! })
+//! .unwrap();
+//! assert!(filter_registry::is_filter_available(300));
+//! filter_registry::unregister_filter(300);
+//! # }
+//! ```
+
+#[cfg(not(feature = "std"))]
+extern crate alloc;
+
+#[cfg(not(feature = "std"))]
+use alloc::vec::Vec;
+
+use crate::error::FormatError;
+use crate::filter_pipeline::FilterDescription;
+
+/// What a codec is told about the filter it is applying.
+#[derive(Debug, Clone, Copy)]
+pub struct FilterContext<'a> {
+    /// The filter as recorded in the dataset's filter pipeline: its ID, name,
+    /// flags and client data (`cd_values`).
+    pub filter: &'a FilterDescription,
+    /// Size in bytes of one dataset element (the datatype's size).
+    pub element_size: usize,
+    /// Decoding only: the most bytes this stage may produce — what entered
+    /// the filter when the chunk was written. 0 means unknown; a decoder then
+    /// falls back to a fixed ceiling. Always 0 when encoding.
+    pub max_output: usize,
+}
+
+impl FilterContext<'_> {
+    /// The filter's client data (`cd_values`).
+    pub fn client_data(&self) -> &[u32] {
+        &self.filter.client_data
+    }
+
+    /// The largest output a decoder should allow: [`Self::max_output`], or
+    /// 256 MiB when that is unknown.
+    pub fn output_limit(&self) -> usize {
+        if self.max_output != 0 {
+            self.max_output
+        } else {
+            crate::filters::MAX_DECOMPRESS_SIZE
+        }
+    }
+}
+
+/// A filter implementation.
+///
+/// `decode` undoes the filter (the read direction). `encode` applies it (the
+/// write direction); the default refuses with
+/// [`FormatError::UnsupportedFilter`], which is right for a read-only codec.
+pub trait FilterCodec: Send + Sync {
+    /// Undo the filter on one chunk. The output must not exceed
+    /// [`FilterContext::output_limit`]; the pipeline rejects a larger one.
+    fn decode(&self, input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError>;
+
+    /// Apply the filter to one chunk.
+    fn encode(&self, input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+        let _ = input;
+        Err(FormatError::UnsupportedFilter(ctx.filter.filter_id))
+    }
+}
+
+/// Any `Fn(&[u8], &FilterContext) -> Result<Vec<u8>, FormatError>` is a
+/// decode-only codec.
+impl<F> FilterCodec for F
+where
+    F: Fn(&[u8], &FilterContext<'_>) -> Result<Vec<u8>, FormatError> + Send + Sync,
+{
+    fn decode(&self, input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+        self(input, ctx)
+    }
+}
+
+/// Signature of a built-in filter's decoder or encoder.
+pub type BuiltinFn = fn(&[u8], &FilterContext<'_>) -> Result<Vec<u8>, FormatError>;
+
+/// A filter compiled into this build.
+#[derive(Debug, Clone, Copy)]
+pub struct BuiltinFilter {
+    /// HDF5 filter ID.
+    pub id: u16,
+    /// Human-readable name.
+    pub name: &'static str,
+    /// Decoder.
+    pub(crate) decode: BuiltinFn,
+    /// Encoder, if this build can write the filter.
+    pub(crate) encode: Option<BuiltinFn>,
+}
+
+impl BuiltinFilter {
+    /// Whether this build can write the filter as well as read it.
+    pub fn can_encode(&self) -> bool {
+        self.encode.is_some()
+    }
+
+    /// Whether the built-in entry only borrows its ID for some chunks, so a
+    /// registered codec may take the rest: the legacy pcodec entry under
+    /// Granular BitRound's 32023, which claims only chunks named `"pcodec"`.
+    fn is_shared(&self) -> bool {
+        self.id == crate::filter_pipeline::FILTER_PCODEC_LEGACY
+    }
+
+    /// Whether this entry decodes chunks written with `filter`.
+    fn claims(&self, filter: &crate::filter_pipeline::FilterDescription) -> bool {
+        !self.is_shared()
+            || filter.name.as_deref() == Some(crate::filter_pipeline::FILTER_PCODEC_LEGACY_NAME)
+    }
+}
+
+/// The filters compiled into this build, in ID order.
+pub fn builtin_filters() -> &'static [BuiltinFilter] {
+    crate::filters::BUILTIN_FILTERS
+}
+
+/// The built-in filter with this ID, if it is compiled in.
+pub fn builtin_filter(id: u16) -> Option<&'static BuiltinFilter> {
+    builtin_filters().iter().find(|f| f.id == id)
+}
+
+/// Why a filter ID may be missing from this build: the filter's name, and
+/// the cargo feature that provides it (`None`: clawhdf5 does not implement
+/// it — register a codec for it with [`register_filter`]). `None` for an ID
+/// clawhdf5 knows nothing about.
+pub fn known_filter(id: u16) -> Option<(&'static str, Option<&'static str>)> {
+    Some(match id {
+        1 => ("deflate", Some("deflate")),
+        4 => ("SZIP", Some("szip")),
+        307 => ("bzip2", Some("bzip2")),
+        480 => ("pcodec", Some("pcodec")),
+        32000 => ("LZF", Some("lzf")),
+        32001 => ("Blosc", Some("blosc")),
+        32004 => ("LZ4", Some("lz4")),
+        32008 => ("bitshuffle", Some("bitshuffle")),
+        32013 => ("ZFP", None),
+        32015 => ("Zstandard", Some("zstd")),
+        32019 => ("JPEG", None),
+        32022 => ("BitGroom", None),
+        32023 => ("Granular BitRound", None),
+        32026 => ("Blosc2", None),
+        _ => return None,
+    })
+}
+
+/// Whether a chunk filtered with `id` can be decoded: a built-in filter or a
+/// registered one.
+pub fn is_filter_available(id: u16) -> bool {
+    if builtin_filter(id).is_some() {
+        return true;
+    }
+    #[cfg(feature = "std")]
+    {
+        registered(id).is_some()
+    }
+    #[cfg(not(feature = "std"))]
+    {
+        false
+    }
+}
+
+#[cfg(feature = "std")]
+mod custom {
+    use super::FilterCodec;
+    use std::collections::BTreeMap;
+    use std::sync::{Arc, PoisonError, RwLock};
+
+    pub(super) type Registry = BTreeMap<u16, Arc<dyn FilterCodec>>;
+
+    static REGISTRY: RwLock<Registry> = RwLock::new(BTreeMap::new());
+
+    pub(super) fn with_read<R>(f: impl FnOnce(&Registry) -> R) -> R {
+        // A panic while holding the lock cannot leave the map half-updated
+        // (every update is a single insert/remove), so poisoning is ignored.
+        f(&REGISTRY.read().unwrap_or_else(PoisonError::into_inner))
+    }
+
+    pub(super) fn with_write<R>(f: impl FnOnce(&mut Registry) -> R) -> R {
+        f(&mut REGISTRY.write().unwrap_or_else(PoisonError::into_inner))
+    }
+}
+
+/// Register a codec for filter `id`, process-wide. It is used for every
+/// chunk read (and, if it implements [`FilterCodec::encode`], written) with
+/// that filter ID, by every file.
+///
+/// A plain closure `Fn(&[u8], &FilterContext) -> Result<Vec<u8>, FormatError>`
+/// registers a decoder. Replaces (and returns) an earlier registration for
+/// the same ID. Fails with [`FormatError::FilterError`] if `id` is a built-in
+/// filter of this build: those cannot be overridden. The exception is 32023
+/// (Granular BitRound): with the `pcodec` feature the built-in entry there
+/// reads only chunks whose filter is named `"pcodec"` (clawhdf5 <= 2.7.0's
+/// files); a codec registered for 32023 decodes every other chunk with that
+/// ID and does all the writing.
+#[cfg(feature = "std")]
+pub fn register_filter<C>(
+    id: u16,
+    codec: C,
+) -> Result<Option<std::sync::Arc<dyn FilterCodec>>, FormatError>
+where
+    C: FilterCodec + 'static,
+{
+    if let Some(builtin) = builtin_filter(id).filter(|b| !b.is_shared()) {
+        return Err(FormatError::FilterError(format!(
+            "filter {id} ({}) is built in and cannot be re-registered",
+            builtin.name
+        )));
+    }
+    let codec: std::sync::Arc<dyn FilterCodec> = std::sync::Arc::new(codec);
+    Ok(custom::with_write(|r| r.insert(id, codec)))
+}
+
+/// Remove the codec registered for `id`. Returns whether one was registered.
+#[cfg(feature = "std")]
+pub fn unregister_filter(id: u16) -> bool {
+    custom::with_write(|r| r.remove(&id).is_some())
+}
+
+/// The codec registered for `id`, if any.
+#[cfg(feature = "std")]
+pub fn registered(id: u16) -> Option<std::sync::Arc<dyn FilterCodec>> {
+    custom::with_read(|r| r.get(&id).cloned())
+}
+
+/// Undo filter `ctx.filter` on `input`: the built-in decoder if there is one
+/// that claims the chunk, else a registered one, else the built-in decoder's
+/// own refusal or [`FormatError::UnsupportedFilter`].
+pub(crate) fn decode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    let id = ctx.filter.filter_id;
+    let builtin = builtin_filter(id);
+    if let Some(builtin) = builtin.filter(|b| b.claims(ctx.filter)) {
+        return (builtin.decode)(input, ctx);
+    }
+    #[cfg(feature = "std")]
+    if let Some(codec) = registered(id) {
+        let out = codec.decode(input, ctx)?;
+        // A registered codec is outside our control: hold it to the same
+        // bound the built-in decoders enforce.
+        if out.len() > ctx.output_limit() {
+            return Err(FormatError::DecompressionError(format!(
+                "filter {id}: decoded {} bytes, more than the {} the chunk can hold",
+                out.len(),
+                ctx.output_limit()
+            )));
+        }
+        return Ok(out);
+    }
+    match builtin {
+        Some(builtin) => (builtin.decode)(input, ctx),
+        None => Err(FormatError::UnsupportedFilter(id)),
+    }
+}
+
+/// Apply filter `ctx.filter` to `input`.
+pub(crate) fn encode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    let id = ctx.filter.filter_id;
+    #[cfg(feature = "std")]
+    if builtin_filter(id).is_some_and(|b| b.is_shared())
+        && let Some(codec) = registered(id)
+    {
+        return codec.encode(input, ctx);
+    }
+    if let Some(builtin) = builtin_filter(id) {
+        return match builtin.encode {
+            Some(encode) => encode(input, ctx),
+            None => Err(FormatError::UnsupportedFilter(id)),
+        };
+    }
+    #[cfg(feature = "std")]
+    if let Some(codec) = registered(id) {
+        return codec.encode(input, ctx);
+    }
+    Err(FormatError::UnsupportedFilter(id))
+}
+
+#[cfg(all(test, feature = "std"))]
+pub(crate) mod tests {
+    use super::*;
+    use crate::filter_pipeline::{FILTER_FLETCHER32, FILTER_SHUFFLE, FilterPipeline};
+    use crate::filters::{compress_chunk, decompress_chunk};
+
+    fn pipeline(id: u16) -> FilterPipeline {
+        FilterPipeline {
+            version: 2,
+            filters: vec![FilterDescription {
+                filter_id: id,
+                name: Some("test".into()),
+                flags: 0,
+                client_data: vec![7],
+            }],
+        }
+    }
+
+    struct Xor;
+    impl FilterCodec for Xor {
+        fn decode(&self, input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+            let k = ctx.client_data()[0] as u8;
+            Ok(input.iter().map(|b| b ^ k).collect())
+        }
+        fn encode(&self, input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+            self.decode(input, ctx)
+        }
+    }
+
+    // Each test uses its own ID: the registry is process-wide and tests run
+    // in parallel.
+
+    #[test]
+    fn unknown_filter_keeps_its_error() {
+        let err = decompress_chunk(b"abc", &pipeline(311), 3, 1).unwrap_err();
+        assert_eq!(err, FormatError::UnsupportedFilter(311));
+        let err = compress_chunk(b"abc", &pipeline(311), 1).unwrap_err();
+        assert_eq!(err, FormatError::UnsupportedFilter(311));
+    }
+
+    #[test]
+    fn registered_codec_round_trips_through_the_pipeline() {
+        assert!(!is_filter_available(312));
+        assert!(register_filter(312, Xor).unwrap().is_none());
+        assert!(is_filter_available(312));
+        let data = b"hello, registry".to_vec();
+        let enc = compress_chunk(&data, &pipeline(312), 1).unwrap();
+        assert_ne!(enc, data);
+        assert_eq!(
+            decompress_chunk(&enc, &pipeline(312), data.len(), 1).unwrap(),
+            data
+        );
+        assert!(unregister_filter(312));
+        assert!(!unregister_filter(312));
+        assert_eq!(
+            decompress_chunk(&enc, &pipeline(312), data.len(), 1).unwrap_err(),
+            FormatError::UnsupportedFilter(312)
+        );
+    }
+
+    #[test]
+    fn closure_registers_a_decoder_only() {
+        register_filter(313, |input: &[u8], _ctx: &FilterContext<'_>| {
+            Ok(input.iter().rev().copied().collect())
+        })
+        .unwrap();
+        assert_eq!(
+            decompress_chunk(b"abc", &pipeline(313), 3, 1).unwrap(),
+            b"cba"
+        );
+        assert_eq!(
+            compress_chunk(b"abc", &pipeline(313), 1).unwrap_err(),
+            FormatError::UnsupportedFilter(313)
+        );
+        unregister_filter(313);
+    }
+
+    #[test]
+    fn registered_decoder_output_is_bounded() {
+        register_filter(314, |_input: &[u8], _ctx: &FilterContext<'_>| {
+            Ok(vec![0u8; 1000])
+        })
+        .unwrap();
+        let err = decompress_chunk(b"abc", &pipeline(314), 10, 1).unwrap_err();
+        assert!(matches!(err, FormatError::DecompressionError(_)), "{err:?}");
+        unregister_filter(314);
+    }
+
+    #[test]
+    fn builtins_cannot_be_overridden() {
+        for id in [FILTER_SHUFFLE, FILTER_FLETCHER32] {
+            let Err(err) = register_filter(id, Xor) else {
+                panic!("built-in filter {id} was re-registered");
+            };
+            assert!(matches!(err, FormatError::FilterError(_)), "{err:?}");
+        }
+        assert!(builtin_filter(FILTER_SHUFFLE).is_some());
+    }
+
+    /// Serialises the tests that register or read filter 32023 (the
+    /// registry is process-wide).
+    pub(crate) static ID_32023: std::sync::Mutex<()> = std::sync::Mutex::new(());
+
+    /// 32023 is Granular BitRound's ID; the `pcodec` build's built-in entry
+    /// there reads only clawhdf5 <= 2.7.0's pcodec chunks (named "pcodec"),
+    /// so a codec can be registered for the rest, and writes with it.
+    #[test]
+    fn a_codec_can_be_registered_for_granular_bitround() {
+        let _guard = ID_32023
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        let named = |name: Option<&str>| FilterPipeline {
+            version: 2,
+            filters: vec![FilterDescription {
+                filter_id: 32023,
+                name: name.map(Into::into),
+                flags: 0,
+                client_data: vec![7],
+            }],
+        };
+        let prev = register_filter(32023, Xor).expect("32023 must be registrable");
+        assert!(prev.is_none());
+        let data = b"granular bitround".to_vec();
+        for name in [None, Some("granular_bitround"), Some("test")] {
+            let pl = named(name);
+            let enc = compress_chunk(&data, &pl, 1).unwrap();
+            assert_ne!(enc, data);
+            assert_eq!(decompress_chunk(&enc, &pl, data.len(), 1).unwrap(), data);
+        }
+        // clawhdf5 <= 2.7.0's pcodec chunks still go to the built-in reader.
+        #[cfg(feature = "pcodec")]
+        {
+            let raw: Vec<u8> = (0..64)
+                .flat_map(|i| (f64::from(i) * 0.5).to_le_bytes())
+                .collect();
+            let comp = crate::filters::pcodec_compress(&raw, 8).unwrap();
+            let mut pl = named(Some("pcodec"));
+            pl.filters[0].client_data = vec![8];
+            assert_eq!(decompress_chunk(&comp, &pl, raw.len(), 8).unwrap(), raw);
+        }
+        assert!(unregister_filter(32023));
+        let pl = named(None);
+        assert!(matches!(
+            decompress_chunk(&data, &pl, data.len(), 1),
+            Err(FormatError::UnsupportedFilter(32023))
+        ));
+    }
+
+    #[test]
+    fn unsupported_filter_error_names_the_filter() {
+        let msg = FormatError::UnsupportedFilter(32026).to_string();
+        assert!(
+            msg.contains("Blosc2") && msg.contains("not implemented"),
+            "{msg}"
+        );
+        let msg = FormatError::UnsupportedFilter(32013).to_string();
+        assert!(msg.contains("ZFP"), "{msg}");
+        let msg = FormatError::UnsupportedFilter(32000).to_string();
+        assert!(msg.contains("LZF") && msg.contains("`lzf`"), "{msg}");
+        assert_eq!(
+            FormatError::UnsupportedFilter(399).to_string(),
+            "unsupported filter: 399"
+        );
+    }
+
+    #[test]
+    fn builtin_table_is_sorted_and_unique() {
+        let ids: Vec<u16> = builtin_filters().iter().map(|f| f.id).collect();
+        let mut sorted = ids.clone();
+        sorted.sort_unstable();
+        sorted.dedup();
+        assert_eq!(ids, sorted);
+    }
+}

crates/clawhdf5-format/src/filters_bitshuffle.rs

@@ -0,0 +1,438 @@
+//! Bitshuffle (HDF5 filter 32008) and the bit transpose it shares with blosc.
+//!
+//! **The transform.** A block of `n` elements (`n` a multiple of 8) of
+//! `es` bytes each is viewed as an `n × 8·es` bit matrix — row *i* is
+//! element *i*, column `8·j + k` is bit *k* (LSB first) of its byte *j* — and
+//! transposed: the output is `8·es` rows of `n` bits, row `8·j + k` holding
+//! bit *k* of byte *j* of every element in order, packed LSB first. That is
+//! what `bshuf_trans_bit_elem` produces (checked against hdf5plugin's
+//! library bit for bit).
+//!
+//! **The filter** (`bshuf_h5filter.c`). `cd_values`: `[0..2]` bitshuffle
+//! version, `[2]` element size, `[3]` block size in elements (0 = default:
+//! 8192 bytes' worth, rounded down to a multiple of 8, at least 128),
+//! `[4]` compression (0 none, 2 LZ4, 3 Zstandard), `[5]` Zstandard level.
+//! The chunk is cut into blocks of `block size` elements; the tail shorter
+//! than a block is transposed as one block rounded down to a multiple of 8
+//! elements, and the last `n mod 8` elements are stored as they are.
+//! Uncompressed, that is the whole chunk. Compressed, the chunk starts with a
+//! 12-byte header — the decoded size (u64 big-endian) and the block size in
+//! bytes (u32 big-endian) — and each transposed block is stored as a u32
+//! big-endian length and an LZ4 block / Zstandard frame; the untransposed
+//! tail follows the last block.
+
+#[cfg(not(feature = "std"))]
+extern crate alloc;
+#[cfg(not(feature = "std"))]
+use alloc::{format, vec, vec::Vec};
+
+use crate::error::FormatError;
+#[cfg(feature = "bitshuffle")]
+use crate::filter_registry::FilterContext;
+
+/// Transpose an 8×8 bit matrix packed in a u64 (byte *r* = row *r*, bit *c*
+/// of that byte = column *c*). An involution.
+#[inline]
+fn transpose8(mut x: u64) -> u64 {
+    let t = (x ^ (x >> 7)) & 0x00AA_00AA_00AA_00AA;
+    x = x ^ t ^ (t << 7);
+    let t = (x ^ (x >> 14)) & 0x0000_CCCC_0000_CCCC;
+    x = x ^ t ^ (t << 14);
+    let t = (x ^ (x >> 28)) & 0x0000_0000_F0F0_F0F0;
+    x ^ t ^ (t << 28)
+}
+
+/// Bit-transpose one block: `input` and `out` are `n * es` bytes, `n` a
+/// multiple of 8.
+pub(crate) fn bitshuffle_block(input: &[u8], out: &mut [u8], n: usize, es: usize) {
+    debug_assert!(n.is_multiple_of(8) && input.len() == n * es && out.len() == n * es);
+    let row = n / 8;
+    for j in 0..es {
+        for g in 0..row {
+            let mut x = 0u64;
+            for t in 0..8 {
+                x |= u64::from(input[(8 * g + t) * es + j]) << (8 * t);
+            }
+            let y = transpose8(x);
+            for k in 0..8 {
+                out[(8 * j + k) * row + g] = (y >> (8 * k)) as u8;
+            }
+        }
+    }
+}
+
+/// Undo [`bitshuffle_block`].
+pub(crate) fn bitunshuffle_block(input: &[u8], out: &mut [u8], n: usize, es: usize) {
+    debug_assert!(n.is_multiple_of(8) && input.len() == n * es && out.len() == n * es);
+    let row = n / 8;
+    for j in 0..es {
+        for g in 0..row {
+            let mut y = 0u64;
+            for k in 0..8 {
+                y |= u64::from(input[(8 * j + k) * row + g]) << (8 * k);
+            }
+            let x = transpose8(y);
+            for t in 0..8 {
+                out[(8 * g + t) * es + j] = (x >> (8 * t)) as u8;
+            }
+        }
+    }
+}
+
+/// `bshuf_default_block_size`: 8 KiB of elements, a multiple of 8, >= 128.
+#[cfg(feature = "bitshuffle")]
+fn default_block_size(es: usize) -> usize {
+    ((8192 / es) / 8 * 8).max(128)
+}
+
+#[cfg(feature = "bitshuffle")]
+fn err(msg: &str) -> FormatError {
+    FormatError::DecompressionError(format!("bitshuffle: {msg}"))
+}
+
+/// `cd_values[4]`: the compression bitshuffle applies after the transpose.
+#[cfg(feature = "bitshuffle")]
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum Codec {
+    None,
+    Lz4,
+    Zstd,
+}
+
+#[cfg(feature = "bitshuffle")]
+fn codec(cd: &[u32]) -> Result<Codec, FormatError> {
+    match cd.get(4).copied().unwrap_or(0) {
+        0 => Ok(Codec::None),
+        2 => Ok(Codec::Lz4),
+        3 => Ok(Codec::Zstd),
+        other => Err(FormatError::FilterError(format!(
+            "bitshuffle: unknown compression {other}"
+        ))),
+    }
+}
+
+/// The element counts of the transposed blocks for `size` elements.
+#[cfg(feature = "bitshuffle")]
+fn blocks(size: usize, block: usize) -> impl Iterator<Item = usize> {
+    let full = size / block;
+    let last = (size % block) / 8 * 8;
+    core::iter::repeat_n(block, full).chain((last > 0).then_some(last))
+}
+
+/// Decode a bitshuffle-filtered chunk.
+#[cfg(feature = "bitshuffle")]
+pub(crate) fn bitshuffle_decode(
+    input: &[u8],
+    ctx: &FilterContext<'_>,
+) -> Result<Vec<u8>, FormatError> {
+    let cd = ctx.client_data();
+    let es = match cd.get(2) {
+        Some(&e) if e != 0 => e as usize,
+        _ => return Err(err("missing element size")),
+    };
+    let codec = codec(cd)?;
+    let limit = ctx.output_limit();
+    if codec == Codec::None {
+        if input.len() > limit {
+            return Err(err("output exceeds the chunk size"));
+        }
+        let block = match cd.get(3) {
+            Some(&b) if b != 0 => b as usize,
+            _ => default_block_size(es),
+        };
+        if !block.is_multiple_of(8) {
+            return Err(err("block size is not a multiple of 8"));
+        }
+        if !input.len().is_multiple_of(es) {
+            return Err(err("chunk is not a whole number of elements"));
+        }
+        let size = input.len() / es;
+        let mut out = vec![0u8; input.len()];
+        let mut pos = 0;
+        for n in blocks(size, block) {
+            let bytes = n * es;
+            bitunshuffle_block(&input[pos..pos + bytes], &mut out[pos..pos + bytes], n, es);
+            pos += bytes;
+        }
+        out[pos..].copy_from_slice(&input[pos..]);
+        return Ok(out);
+    }
+
+    let header = input.get(..12).ok_or_else(|| err("truncated header"))?;
+    let total = u64::from_be_bytes(header[..8].try_into().unwrap());
+    let block_bytes = u32::from_be_bytes(header[8..12].try_into().unwrap()) as usize;
+    let total = usize::try_from(total)
+        .ok()
+        .filter(|&t| t <= limit)
+        .ok_or_else(|| err("decoded size exceeds the chunk size"))?;
+    if !total.is_multiple_of(es) {
+        return Err(err("chunk is not a whole number of elements"));
+    }
+    if block_bytes == 0 || !block_bytes.is_multiple_of(es) {
+        return Err(err("bad block size"));
+    }
+    let block = block_bytes / es;
+    if !block.is_multiple_of(8) {
+        return Err(err("block size is not a multiple of 8"));
+    }
+    let size = total / es;
+    let mut out = vec![0u8; total];
+    let mut tmp = vec![0u8; block_bytes.min(total)];
+    let mut ip = 12usize;
+    let mut op = 0usize;
+    let mut zstd = None;
+    for n in blocks(size, block) {
+        let bytes = n * es;
+        let len = input
+            .get(ip..ip + 4)
+            .map(|b| u32::from_be_bytes(b.try_into().unwrap()) as usize)
+            .ok_or_else(|| err("truncated block header"))?;
+        ip += 4;
+        let comp = input
+            .get(ip..ip.saturating_add(len))
+            .ok_or_else(|| err("truncated block"))?;
+        ip += len;
+        let dst = &mut tmp[..bytes];
+        let got = match codec {
+            Codec::Lz4 => lz4_flex::block::decompress_into(comp, dst)
+                .map_err(|e| err(&format!("lz4: {e}")))?,
+            Codec::Zstd => zstd_decode_into(
+                zstd.get_or_insert_with(ruzstd::decoding::FrameDecoder::new),
+                comp,
+                dst,
+            )?,
+            Codec::None => unreachable!(),
+        };
+        if got != bytes {
+            return Err(err("block decoded to the wrong size"));
+        }
+        bitunshuffle_block(dst, &mut out[op..op + bytes], n, es);
+        op += bytes;
+    }
+    let tail = total - op;
+    let rest = input
+        .get(ip..ip + tail)
+        .ok_or_else(|| err("truncated trailing elements"))?;
+    out[op..].copy_from_slice(rest);
+    Ok(out)
+}
+
+/// Decode Zstandard frames into exactly `dst`, failing if they hold more.
+#[cfg(any(feature = "bitshuffle", feature = "blosc"))]
+pub(crate) fn zstd_decode_into(
+    decoder: &mut ruzstd::decoding::FrameDecoder,
+    frames: &[u8],
+    dst: &mut [u8],
+) -> Result<usize, FormatError> {
+    decoder
+        .decode_all(frames, dst)
+        .map_err(|e| FormatError::DecompressionError(format!("zstd: {e}")))
+}
+
+/// Compress with ruzstd. It implements one level (roughly zstd's level 1),
+/// so the requested level only matters to other encoders.
+#[cfg(any(feature = "bitshuffle", feature = "blosc"))]
+pub(crate) fn zstd_encode(data: &[u8]) -> Vec<u8> {
+    ruzstd::encoding::compress_to_vec(data, ruzstd::encoding::CompressionLevel::Fastest)
+}
+
+/// Encode a chunk with the bitshuffle filter.
+#[cfg(feature = "bitshuffle")]
+pub(crate) fn bitshuffle_encode(
+    input: &[u8],
+    ctx: &FilterContext<'_>,
+) -> Result<Vec<u8>, FormatError> {
+    let cd = ctx.client_data();
+    let es = match cd.get(2) {
+        Some(&e) if e != 0 => e as usize,
+        _ => ctx.element_size.max(1),
+    };
+    let codec = codec(cd)?;
+    let block = match cd.get(3) {
+        Some(&b) if b != 0 => b as usize,
+        _ => default_block_size(es),
+    };
+    let cerr = |m: &str| FormatError::CompressionError(format!("bitshuffle: {m}"));
+    if !block.is_multiple_of(8) {
+        return Err(cerr("block size is not a multiple of 8"));
+    }
+    if !input.len().is_multiple_of(es) {
+        return Err(cerr("chunk is not a whole number of elements"));
+    }
+    let size = input.len() / es;
+    let mut out = Vec::with_capacity(input.len() + 12 + input.len() / 64);
+    if codec != Codec::None {
+        out.extend_from_slice(&(input.len() as u64).to_be_bytes());
+        let block_bytes =
+            u32::try_from(block * es).map_err(|_| cerr("block size does not fit in 32 bits"))?;
+        out.extend_from_slice(&block_bytes.to_be_bytes());
+    }
+    let mut tmp = vec![0u8; (block * es).min(input.len())];
+    let mut pos = 0;
+    for n in blocks(size, block) {
+        let bytes = n * es;
+        let dst = &mut tmp[..bytes];
+        bitshuffle_block(&input[pos..pos + bytes], dst, n, es);
+        match codec {
+            Codec::None => out.extend_from_slice(dst),
+            Codec::Lz4 | Codec::Zstd => {
+                let comp = if codec == Codec::Lz4 {
+                    lz4_flex::block::compress(dst)
+                } else {
+                    zstd_encode(dst)
+                };
+                out.extend_from_slice(&(comp.len() as u32).to_be_bytes());
+                out.extend_from_slice(&comp);
+            }
+        }
+        pos += bytes;
+    }
+    out.extend_from_slice(&input[pos..]);
+    Ok(out)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// The definition, one bit at a time.
+    fn naive(input: &[u8], n: usize, es: usize) -> Vec<u8> {
+        let mut out = vec![0u8; n * es];
+        for i in 0..n {
+            for j in 0..es {
+                for k in 0..8 {
+                    if input[i * es + j] >> k & 1 == 1 {
+                        let p = (8 * j + k) * n + i;
+                        out[p / 8] |= 1 << (p % 8);
+                    }
+                }
+            }
+        }
+        out
+    }
+
+    #[test]
+    fn transpose_matches_the_definition_and_inverts() {
+        for (n, es) in [(8, 1), (16, 2), (24, 4), (128, 8), (64, 3), (8, 16)] {
+            let input: Vec<u8> = (0..n * es)
+                .map(|i| (i as u32).wrapping_mul(2_654_435_761).rotate_left(7) as u8)
+                .collect();
+            let mut out = vec![0u8; n * es];
+            bitshuffle_block(&input, &mut out, n, es);
+            assert_eq!(out, naive(&input, n, es), "n={n} es={es}");
+            let mut back = vec![0u8; n * es];
+            bitunshuffle_block(&out, &mut back, n, es);
+            assert_eq!(back, input);
+        }
+    }
+
+    #[cfg(feature = "bitshuffle")]
+    fn ctx_for(cd: Vec<u32>) -> crate::filter_pipeline::FilterDescription {
+        crate::filter_pipeline::FilterDescription {
+            filter_id: crate::filter_pipeline::FILTER_BITSHUFFLE,
+            name: None,
+            flags: 0,
+            client_data: cd,
+        }
+    }
+
+    #[cfg(feature = "bitshuffle")]
+    #[test]
+    fn filter_round_trips_every_mode() {
+        for es in [1usize, 2, 4, 8] {
+            for n in [0usize, 1, 7, 8, 100, 1000, 5003] {
+                let data: Vec<u8> = (0..n * es)
+                    .map(|i| (i % 97) as u8 ^ (i / 300) as u8)
+                    .collect();
+                for (comp, block) in [(0, 0), (0, 16), (2, 0), (2, 64), (3, 0), (3, 1024)] {
+                    let f = ctx_for(vec![0, 4, es as u32, block, comp]);
+                    let ctx = FilterContext {
+                        filter: &f,
+                        element_size: es,
+                        max_output: data.len(),
+                    };
+                    let enc = bitshuffle_encode(&data, &ctx).unwrap();
+                    let dec = bitshuffle_decode(&enc, &ctx).unwrap();
+                    assert_eq!(dec, data, "es={es} n={n} comp={comp} block={block}");
+                }
+            }
+        }
+    }
+
+    #[cfg(feature = "bitshuffle")]
+    #[test]
+    fn rejects_oversized_and_truncated_chunks() {
+        let data = vec![5u8; 4096];
+        let f = ctx_for(vec![0, 4, 4, 0, 2]);
+        let mut ctx = FilterContext {
+            filter: &f,
+            element_size: 4,
+            max_output: data.len(),
+        };
+        let enc = bitshuffle_encode(&data, &ctx).unwrap();
+        assert!(bitshuffle_decode(&enc[..enc.len() - 1], &ctx).is_err());
+        ctx.max_output = 100;
+        assert!(bitshuffle_decode(&enc, &ctx).is_err());
+    }
+
+    /// Random and mutated chunks, in every mode, and hostile `cd_values`:
+    /// errors are fine, panics are not.
+    #[cfg(feature = "bitshuffle")]
+    #[test]
+    fn fuzzed_chunks_never_panic() {
+        use crate::test_fuzz::{Rng, fuzz_decoder};
+        let data: Vec<u8> = (0..3001u32)
+            .flat_map(|i| ((i / 7) as u16).to_le_bytes())
+            .collect();
+        for (comp, block) in [(0, 0), (0, 16), (2, 0), (2, 64), (3, 0), (3, 1024)] {
+            let f = ctx_for(vec![0, 4, 2, block, comp]);
+            let ctx = FilterContext {
+                filter: &f,
+                element_size: 2,
+                max_output: data.len(),
+            };
+            let seeds = vec![
+                bitshuffle_encode(&data, &ctx).unwrap(),
+                bitshuffle_encode(&data[..34], &ctx).unwrap(),
+                bitshuffle_encode(&data[..512], &ctx).unwrap(),
+            ];
+            fuzz_decoder(
+                0xb5 + comp as u64 * 7 + block as u64,
+                &seeds,
+                4_000,
+                data.len(),
+                |s| bitshuffle_decode(s, &ctx),
+            );
+        }
+        // Hostile filter parameters on a valid chunk.
+        let mut rng = Rng::new(0xcd);
+        let good = ctx_for(vec![0, 4, 2, 0, 2]);
+        let enc = bitshuffle_encode(
+            &data,
+            &FilterContext {
+                filter: &good,
+                element_size: 2,
+                max_output: data.len(),
+            },
+        )
+        .unwrap();
+        for _ in 0..3_000 {
+            let cd: Vec<u32> = (0..rng.below(7))
+                .map(|_| match rng.below(4) {
+                    0 => rng.below(5) as u32,
+                    1 => u32::MAX - rng.below(4) as u32,
+                    2 => 1 << rng.below(32),
+                    _ => rng.next_u64() as u32,
+                })
+                .collect();
+            let f = ctx_for(cd);
+            let ctx = FilterContext {
+                filter: &f,
+                element_size: 2,
+                max_output: data.len(),
+            };
+            let _ = bitshuffle_decode(&enc, &ctx);
+            let _ = bitshuffle_decode(&data, &ctx);
+        }
+    }
+}

crates/clawhdf5-format/src/filters_blosc.rs

@@ -0,0 +1,711 @@
+//! Blosc 1 (HDF5 filter 32001, `hdf5-blosc`, hdf5plugin's `Blosc`), in pure
+//! Rust: the Blosc 1 frame, its byte shuffle and bit shuffle, and the
+//! BloscLZ, LZ4/LZ4HC, Snappy, Zlib and Zstandard codecs inside it.
+//!
+//! **Frame** (c-blosc 1.x, format version 2). A 16-byte header — version
+//! (2), codec format version (1), flags, type size, then little-endian `u32`
+//! decoded size, block size and frame size. Flags: bit 0 byte shuffle, bit
+//! 1 stored raw ("memcpyed": the data follows the header), bit 2 bit
+//! shuffle, bit 4 "do not split", bits 5-7 the codec (0 BloscLZ, 1 LZ4 and
+//! LZ4HC, 2 Snappy, 3 Zlib, 4 Zstandard). Unless stored raw, a table of
+//! `u32` block offsets follows, one per block of `block size` bytes (the
+//! last one may be shorter). A block is one stream, or — when the "do not
+//! split" flag is clear, the type size is at most 16, the block holds at
+//! least 128 elements, and it is not the short last block — `type size`
+//! streams, one per byte plane. Each stream is a `u32` length and the
+//! codec's output; a length equal to the stream's decoded size means the
+//! bytes are stored raw. The decoded block is then unshuffled (byte shuffle
+//! for type size > 1; bit shuffle when the block holds a multiple of 8
+//! elements, the trailing partial element copied as is).
+//!
+//! **Filter** (`blosc_filter.c`) `cd_values`: `[0]` filter revision, `[1]`
+//! Blosc format version, `[2]` type size, `[3]` chunk size in bytes, `[4]`
+//! compression level, `[5]` shuffle (0 none, 1 byte, 2 bit), `[6]`
+//! compressor (0 blosclz, 1 lz4, 2 lz4hc, 3 snappy, 4 zlib, 5 zstd). The
+//! decoder needs only the frame.
+
+use crate::error::FormatError;
+use crate::filter_registry::FilterContext;
+use crate::filters_bitshuffle::{bitshuffle_block, bitunshuffle_block};
+
+const HEADER: usize = 16;
+const FLAG_SHUFFLE: u8 = 0x01;
+const FLAG_MEMCPYED: u8 = 0x02;
+const FLAG_BITSHUFFLE: u8 = 0x04;
+const FLAG_FUTURE: u8 = 0x08;
+const FLAG_DONT_SPLIT: u8 = 0x10;
+const MAX_SPLITS: usize = 16;
+const MIN_BUFFERSIZE: usize = 128;
+
+fn err(msg: &str) -> FormatError {
+    FormatError::DecompressionError(format!("blosc: {msg}"))
+}
+
+fn le32(b: &[u8], at: usize) -> Result<usize, FormatError> {
+    b.get(at..at + 4)
+        .map(|s| u32::from_le_bytes(s.try_into().unwrap()) as usize)
+        .ok_or_else(|| err("truncated frame"))
+}
+
+/// The codec inside a Blosc frame (flags bits 5-7).
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum Codec {
+    BloscLz,
+    Lz4,
+    Snappy,
+    Zlib,
+    Zstd,
+}
+
+impl Codec {
+    fn from_flags(flags: u8) -> Result<Codec, FormatError> {
+        match flags >> 5 {
+            0 => Ok(Codec::BloscLz),
+            1 => Ok(Codec::Lz4),
+            2 => Ok(Codec::Snappy),
+            3 => Ok(Codec::Zlib),
+            4 => Ok(Codec::Zstd),
+            other => Err(err(&format!("unknown codec {other}"))),
+        }
+    }
+}
+
+/// Decode one codec stream into exactly `dst`.
+fn decode_stream(
+    codec: Codec,
+    src: &[u8],
+    dst: &mut [u8],
+    zstd: &mut Option<ruzstd::decoding::FrameDecoder>,
+) -> Result<(), FormatError> {
+    let n = match codec {
+        Codec::BloscLz => blosclz_decompress(src, dst),
+        Codec::Lz4 => {
+            lz4_flex::block::decompress_into(src, dst).map_err(|e| err(&format!("lz4: {e}")))?
+        }
+        Codec::Snappy => {
+            let len = snap::raw::decompress_len(src).map_err(|e| err(&format!("snappy: {e}")))?;
+            if len != dst.len() {
+                return Err(err("snappy stream has the wrong size"));
+            }
+            snap::raw::Decoder::new()
+                .decompress(src, dst)
+                .map_err(|e| err(&format!("snappy: {e}")))?
+        }
+        Codec::Zlib => {
+            let out = crate::filters::inflate_bounded(src, dst.len(), dst.len())
+                .map_err(|e| err(&format!("zlib: {e}")))?;
+            let n = out.len();
+            if n == dst.len() {
+                dst.copy_from_slice(&out);
+            }
+            n
+        }
+        Codec::Zstd => crate::filters_bitshuffle::zstd_decode_into(
+            zstd.get_or_insert_with(ruzstd::decoding::FrameDecoder::new),
+            src,
+            dst,
+        )?,
+    };
+    if n != dst.len() {
+        return Err(err("stream decoded to the wrong size"));
+    }
+    Ok(())
+}
+
+/// Decode a Blosc-filtered chunk: one Blosc 1 frame.
+///
+/// An HDF5 chunk is never empty, so a frame that decodes to nothing where
+/// the chunk size is known is corrupt (libhdf5's filter fails it too).
+pub(crate) fn blosc_decode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    let out = blosc_decompress(input, ctx.output_limit())?;
+    if out.is_empty() && ctx.max_output != 0 {
+        return Err(err("empty frame for a non-empty chunk"));
+    }
+    Ok(out)
+}
+
+/// Decompress a Blosc 1 frame, refusing more than `limit` bytes of output.
+pub fn blosc_decompress(input: &[u8], limit: usize) -> Result<Vec<u8>, FormatError> {
+    if input.len() < HEADER {
+        return Err(err("truncated header"));
+    }
+    let version = input[0];
+    let codec_version = input[1];
+    let flags = input[2];
+    let typesize = input[3] as usize;
+    let nbytes = le32(input, 4)?;
+    let blocksize = le32(input, 8)?;
+    let cbytes = le32(input, 12)?;
+    if version != 1 && version != 2 {
+        return Err(err(&format!(
+            "frame format version {version} is not Blosc 1 (a Blosc 2 chunk?)"
+        )));
+    }
+    if flags & FLAG_FUTURE != 0 {
+        return Err(err("unknown header flags"));
+    }
+    if nbytes > limit {
+        return Err(err("decoded size exceeds the chunk size"));
+    }
+    if cbytes > input.len() {
+        return Err(err("frame is longer than the chunk"));
+    }
+    if cbytes < HEADER {
+        return Err(err("truncated frame"));
+    }
+    let src = &input[..cbytes];
+    if nbytes == 0 {
+        return Ok(Vec::new());
+    }
+    if blocksize == 0 || typesize == 0 {
+        return Err(err("bad block or type size"));
+    }
+    let mut out = vec![0u8; nbytes];
+    if flags & FLAG_MEMCPYED != 0 {
+        if cbytes != nbytes + HEADER {
+            return Err(err("stored frame has the wrong size"));
+        }
+        out.copy_from_slice(&src[HEADER..]);
+        return Ok(out);
+    }
+    let codec = Codec::from_flags(flags)?;
+    if codec_version != 1 {
+        return Err(err(&format!(
+            "unsupported {codec:?} format version {codec_version}"
+        )));
+    }
+    let nblocks = nbytes.div_ceil(blocksize);
+    let leftover = nbytes % blocksize;
+    if nblocks > (cbytes - HEADER) / 4 {
+        return Err(err("block table is truncated"));
+    }
+    let block_len = blocksize.min(nbytes);
+    let mut tmp = vec![0u8; block_len];
+    let mut zstd = None;
+    let dont_split = flags & FLAG_DONT_SPLIT != 0;
+    for j in 0..nblocks {
+        let is_leftover = j == nblocks - 1 && leftover > 0;
+        let bsize = if is_leftover { leftover } else { blocksize };
+        let nsplits = if !dont_split
+            && typesize <= MAX_SPLITS
+            && bsize / typesize >= MIN_BUFFERSIZE
+            && !is_leftover
+        {
+            typesize
+        } else {
+            1
+        };
+        let neblock = bsize / nsplits;
+        let mut pos = le32(src, HEADER + 4 * j)?;
+        let tmp = &mut tmp[..bsize];
+        for s in 0..nsplits {
+            let clen = src
+                .get(pos..)
+                .and_then(|rest| rest.get(..4))
+                .map(|b| u32::from_le_bytes(b.try_into().unwrap()) as usize)
+                .ok_or_else(|| err("block offset out of range"))?;
+            pos += 4;
+            let stream = src
+                .get(pos..pos.saturating_add(clen))
+                .ok_or_else(|| err("stream runs past the frame"))?;
+            let dst = &mut tmp[s * neblock..(s + 1) * neblock];
+            if clen == neblock {
+                dst.copy_from_slice(stream);
+            } else {
+                decode_stream(codec, stream, dst, &mut zstd)?;
+            }
+            pos += clen;
+        }
+        // `bsize` is a whole number of splits by construction (`nsplits` > 1
+        // only for full blocks, and c-blosc sizes those in whole elements).
+        if nsplits * neblock != bsize {
+            return Err(err("block is not a whole number of streams"));
+        }
+        let dest = &mut out[j * blocksize..j * blocksize + bsize];
+        unshuffle_block(flags, typesize, tmp, dest);
+    }
+    Ok(out)
+}
+
+/// Undo the frame's shuffle on one decoded block.
+fn unshuffle_block(flags: u8, typesize: usize, src: &[u8], dest: &mut [u8]) {
+    let bsize = src.len();
+    if flags & FLAG_SHUFFLE != 0 && typesize > 1 {
+        let n = bsize / typesize;
+        for i in 0..n {
+            for b in 0..typesize {
+                dest[i * typesize + b] = src[b * n + i];
+            }
+        }
+        dest[n * typesize..].copy_from_slice(&src[n * typesize..]);
+    } else if flags & FLAG_BITSHUFFLE != 0 && bsize >= typesize {
+        let n = bsize / typesize;
+        if n.is_multiple_of(8) {
+            let body = n * typesize;
+            bitunshuffle_block(&src[..body], &mut dest[..body], n, typesize);
+            dest[body..].copy_from_slice(&src[body..]);
+        } else {
+            dest.copy_from_slice(src);
+        }
+    } else {
+        dest.copy_from_slice(src);
+    }
+}
+
+/// BloscLZ decompression (c-blosc 1.21 `blosclz_decompress`): returns the
+/// number of bytes written, or 0 on malformed input — exactly as the C
+/// decoder, including stopping before a match that ends the stream, so a
+/// stream libblosc rejects is rejected here too.
+///
+/// Instructions: a control byte `ctrl`. Below 32, a literal run of
+/// `ctrl + 1` bytes. Otherwise a match: length `(ctrl >> 5) + 2`, extended
+/// by following bytes while they are 255 when the top three bits are all
+/// set; distance `((ctrl & 31) << 8) + next byte + 1`, or — when that byte
+/// is 255 and the high bits are 31 — a 16-bit big-endian distance plus 8192.
+/// The first instruction is always a literal.
+pub(crate) fn blosclz_decompress(input: &[u8], out: &mut [u8]) -> usize {
+    const MAX_DISTANCE: usize = 8191;
+    let limit = input.len();
+    if limit == 0 {
+        return 0;
+    }
+    let mut ip = 1usize;
+    let mut op = 0usize;
+    let mut ctrl = (input[0] & 31) as usize;
+    loop {
+        if ctrl >= 32 {
+            let mut len = (ctrl >> 5) - 1;
+            let ofs = (ctrl & 31) << 8;
+            if len == 6 {
+                loop {
+                    if ip + 1 >= limit {
+                        return 0;
+                    }
+                    let code = input[ip] as usize;
+                    ip += 1;
+                    len += code;
+                    if code != 255 {
+                        break;
+                    }
+                }
+            } else if ip + 1 >= limit {
+                return 0;
+            }
+            let code = input[ip] as usize;
+            ip += 1;
+            len += 3;
+            // The copy source is `distance` bytes back.
+            let mut distance = ofs + code + 1;
+            if code == 255 && ofs == 31 << 8 {
+                if ip + 1 >= limit {
+                    return 0;
+                }
+                let far = ((input[ip] as usize) << 8) + input[ip + 1] as usize;
+                ip += 2;
+                distance = far + MAX_DISTANCE + 1;
+            }
+            if op + len > out.len() {
+                return 0;
+            }
+            if distance > op {
+                return 0;
+            }
+            if ip >= limit {
+                break;
+            }
+            ctrl = input[ip] as usize;
+            ip += 1;
+            let start = op - distance;
+            if distance >= len {
+                out.copy_within(start..start + len, op);
+            } else {
+                for k in 0..len {
+                    out[op + k] = out[start + k];
+                }
+            }
+            op += len;
+        } else {
+            let run = ctrl + 1;
+            if op + run > out.len() || ip + run > limit {
+                return 0;
+            }
+            out[op..op + run].copy_from_slice(&input[ip..ip + run]);
+            op += run;
+            ip += run;
+            if ip >= limit {
+                break;
+            }
+            ctrl = input[ip] as usize;
+            ip += 1;
+        }
+    }
+    op
+}
+
+/// The codec our encoder puts inside the frame.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum EncodeCodec {
+    Lz4,
+    Snappy,
+    Zlib,
+    Zstd,
+}
+
+impl EncodeCodec {
+    /// From the filter's `cd_values[6]` compressor code.
+    fn from_cd(code: u32) -> Result<EncodeCodec, FormatError> {
+        match code {
+            1 | 2 => Ok(EncodeCodec::Lz4),
+            3 => Ok(EncodeCodec::Snappy),
+            4 => Ok(EncodeCodec::Zlib),
+            5 => Ok(EncodeCodec::Zstd),
+            0 => Err(FormatError::CompressionError(
+                "blosc: clawhdf5 cannot write BloscLZ; choose lz4, snappy, zlib or zstd".into(),
+            )),
+            other => Err(FormatError::CompressionError(format!(
+                "blosc: unknown compressor {other}"
+            ))),
+        }
+    }
+
+    fn flags(self) -> u8 {
+        (match self {
+            EncodeCodec::Lz4 => 1,
+            EncodeCodec::Snappy => 2,
+            EncodeCodec::Zlib => 3,
+            EncodeCodec::Zstd => 4,
+        }) << 5
+    }
+
+    fn encode(self, data: &[u8], level: u32) -> Result<Vec<u8>, FormatError> {
+        match self {
+            EncodeCodec::Lz4 => Ok(lz4_flex::block::compress(data)),
+            EncodeCodec::Snappy => snap::raw::Encoder::new()
+                .compress_vec(data)
+                .map_err(|e| FormatError::CompressionError(format!("blosc: snappy: {e}"))),
+            EncodeCodec::Zlib => crate::filters::deflate_bounded(data, level.min(9))
+                .map_err(|e| FormatError::CompressionError(format!("blosc: zlib: {e}"))),
+            EncodeCodec::Zstd => Ok(crate::filters_bitshuffle::zstd_encode(data)),
+        }
+    }
+}
+
+/// Block size our encoder uses: at most 256 KiB, a whole number of
+/// elements (and, for bit shuffle, of 8-element groups).
+fn encode_block_size(nbytes: usize, typesize: usize, bitshuffle: bool) -> usize {
+    let unit = if bitshuffle { 8 * typesize } else { typesize };
+    let target = (256 * 1024).min(nbytes);
+    if target < unit {
+        return nbytes.max(1);
+    }
+    target / unit * unit
+}
+
+/// Encode a chunk as one Blosc 1 frame. `cd_values` as hdf5-blosc:
+/// `[2]` type size, `[4]` level (0 = store), `[5]` shuffle, `[6]` codec.
+pub(crate) fn blosc_encode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    let cd = ctx.client_data();
+    let cerr = |m: &str| FormatError::CompressionError(format!("blosc: {m}"));
+    let typesize = match cd.get(2) {
+        Some(&t) if t != 0 => t as usize,
+        _ => ctx.element_size.max(1),
+    };
+    // Blosc records the type size in one byte; c-blosc treats larger types
+    // as bytes.
+    let typesize = if typesize > 255 { 1 } else { typesize };
+    let level = cd.get(4).copied().unwrap_or(5);
+    let shuffle = cd.get(5).copied().unwrap_or(1);
+    let codec = EncodeCodec::from_cd(cd.get(6).copied().unwrap_or(1))?;
+    let nbytes = input.len();
+    if nbytes > i32::MAX as usize - HEADER {
+        return Err(cerr("chunk too large for a Blosc frame"));
+    }
+    let mut flags = codec.flags();
+    match shuffle {
+        0 => {}
+        1 => flags |= FLAG_SHUFFLE,
+        2 => flags |= FLAG_BITSHUFFLE,
+        other => return Err(cerr(&format!("unknown shuffle mode {other}"))),
+    }
+    let blocksize = encode_block_size(nbytes, typesize, shuffle == 2);
+    let header = |flags: u8, blocksize: usize, cbytes: usize| {
+        let mut h = Vec::with_capacity(HEADER);
+        h.extend_from_slice(&[2, 1, flags, typesize as u8]);
+        h.extend_from_slice(&(nbytes as u32).to_le_bytes());
+        h.extend_from_slice(&(blocksize as u32).to_le_bytes());
+        h.extend_from_slice(&(cbytes as u32).to_le_bytes());
+        h
+    };
+    let stored = || {
+        let mut out = header(
+            FLAG_MEMCPYED | (flags & !(FLAG_SHUFFLE | FLAG_BITSHUFFLE)),
+            blocksize,
+            nbytes + HEADER,
+        );
+        out.extend_from_slice(input);
+        out
+    };
+    if level == 0 || nbytes == 0 {
+        return Ok(stored());
+    }
+
+    let nblocks = nbytes.div_ceil(blocksize);
+    let leftover = nbytes % blocksize;
+    let mut body = Vec::with_capacity(nbytes / 2);
+    let mut starts = Vec::with_capacity(nblocks);
+    let table_end = HEADER + 4 * nblocks;
+    let mut shuffled = vec![0u8; blocksize];
+    for j in 0..nblocks {
+        let is_leftover = j == nblocks - 1 && leftover > 0;
+        let bsize = if is_leftover { leftover } else { blocksize };
+        let block = &input[j * blocksize..j * blocksize + bsize];
+        let sh = &mut shuffled[..bsize];
+        shuffle_block(flags, typesize, block, sh);
+        starts.push(table_end + body.len());
+        let nsplits = if typesize <= MAX_SPLITS
+            && bsize / typesize >= MIN_BUFFERSIZE
+            && !is_leftover
+            && bsize.is_multiple_of(typesize)
+        {
+            typesize
+        } else {
+            1
+        };
+        let neblock = bsize / nsplits;
+        for s in 0..nsplits {
+            let part = &sh[s * neblock..(s + 1) * neblock];
+            let comp = codec.encode(part, level)?;
+            if comp.len() < neblock {
+                body.extend_from_slice(&(comp.len() as u32).to_le_bytes());
+                body.extend_from_slice(&comp);
+            } else {
+                body.extend_from_slice(&(neblock as u32).to_le_bytes());
+                body.extend_from_slice(part);
+            }
+        }
+        if table_end + body.len() >= nbytes + HEADER {
+            // Incompressible: store instead, as c-blosc does.
+            return Ok(stored());
+        }
+    }
+    // A split block must decode as split: the decoder infers splitting from
+    // the same rule, which requires a whole number of elements per block.
+    let cbytes = table_end + body.len();
+    let mut out = header(flags, blocksize, cbytes);
+    for s in starts {
+        out.extend_from_slice(&(s as u32).to_le_bytes());
+    }
+    out.extend_from_slice(&body);
+    Ok(out)
+}
+
+/// Apply the frame's shuffle to one block (the inverse of
+/// [`unshuffle_block`]).
+fn shuffle_block(flags: u8, typesize: usize, src: &[u8], dest: &mut [u8]) {
+    let bsize = src.len();
+    if flags & FLAG_SHUFFLE != 0 && typesize > 1 {
+        let n = bsize / typesize;
+        for i in 0..n {
+            for b in 0..typesize {
+                dest[b * n + i] = src[i * typesize + b];
+            }
+        }
+        dest[n * typesize..].copy_from_slice(&src[n * typesize..]);
+    } else if flags & FLAG_BITSHUFFLE != 0 && bsize >= typesize {
+        let n = bsize / typesize;
+        if n.is_multiple_of(8) {
+            let body = n * typesize;
+            bitshuffle_block(&src[..body], &mut dest[..body], n, typesize);
+            dest[body..].copy_from_slice(&src[body..]);
+        } else {
+            dest.copy_from_slice(src);
+        }
+    } else {
+        dest.copy_from_slice(src);
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::filter_pipeline::{FILTER_BLOSC, FilterDescription};
+
+    /// A blosclz stream: literal "abc", then a 9-byte match 3 back (a run
+    /// of "abc"), then literal "Z".
+    #[test]
+    fn blosclz_decodes_literals_and_overlapping_matches() {
+        // Match: length (ctrl >> 5) + 2 = 8, distance ofs + code + 1 = 3.
+        let stream = [2, b'a', b'b', b'c', (6 << 5), 2, 0, b'Z'];
+        let mut out = [0u8; 12];
+        assert_eq!(blosclz_decompress(&stream, &mut out), 12);
+        assert_eq!(&out, b"abcabcabcabZ");
+        // A stream cut inside a match is malformed.
+        let mut out = [0u8; 11];
+        assert_eq!(blosclz_decompress(&stream[..6], &mut out), 0);
+        // A match before the start of the output is malformed.
+        assert_eq!(blosclz_decompress(&[0, b'a', 32, 5, 0, b'x'], &mut out), 0);
+    }
+
+    fn desc(cd: Vec<u32>) -> FilterDescription {
+        FilterDescription {
+            filter_id: FILTER_BLOSC,
+            name: None,
+            flags: 0,
+            client_data: cd,
+        }
+    }
+
+    #[test]
+    fn frame_round_trips_every_codec_and_shuffle() {
+        for ts in [1usize, 2, 4, 8, 3, 32] {
+            for n in [0usize, 5, 100, 1000, 70_000, 300_001] {
+                if n * ts > 1 << 20 && ts > 1 {
+                    continue;
+                }
+                let data: Vec<u8> = (0..n * ts)
+                    .map(|i| ((i / ts) % 200) as u8 ^ (i % ts) as u8)
+                    .collect();
+                for codec in [1u32, 3, 4, 5] {
+                    for shuffle in [0u32, 1, 2] {
+                        for level in [0u32, 5] {
+                            let f = desc(vec![2, 2, ts as u32, 0, level, shuffle, codec]);
+                            let ctx = FilterContext {
+                                filter: &f,
+                                element_size: ts,
+                                max_output: data.len(),
+                            };
+                            let enc = blosc_encode(&data, &ctx).unwrap();
+                            let dec = blosc_decode(&enc, &ctx).unwrap_or_else(|e| {
+                                panic!("ts={ts} n={n} codec={codec} shuffle={shuffle}: {e}")
+                            });
+                            assert!(
+                                dec == data,
+                                "ts={ts} n={n} codec={codec} shuffle={shuffle} level={level}"
+                            );
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn rejects_bad_frames() {
+        let data = vec![9u8; 50_000];
+        let f = desc(vec![2, 2, 4, 0, 5, 1, 1]);
+        let ctx = FilterContext {
+            filter: &f,
+            element_size: 4,
+            max_output: data.len(),
+        };
+        let enc = blosc_encode(&data, &ctx).unwrap();
+        assert!(blosc_decode(&enc[..enc.len() - 3], &ctx).is_err());
+        let small = FilterContext {
+            max_output: 49_999,
+            ..ctx
+        };
+        assert!(blosc_decode(&enc, &small).is_err());
+        let mut v3 = enc.clone();
+        v3[0] = 3;
+        assert!(blosc_decode(&v3, &ctx).is_err());
+        let f0 = desc(vec![2, 2, 4, 0, 5, 1, 0]);
+        let ctx0 = FilterContext { filter: &f0, ..ctx };
+        assert!(blosc_encode(&data, &ctx0).is_err());
+    }
+
+    /// A frame that declares no data, for a chunk that has some.
+    #[test]
+    fn empty_frame_for_a_non_empty_chunk_is_an_error() {
+        let mut frame = vec![2u8, 1, 0x20, 4];
+        for v in [0u32, 64, 16] {
+            frame.extend_from_slice(&v.to_le_bytes());
+        }
+        assert_eq!(blosc_decompress(&frame, 64).unwrap(), b"");
+        let f = desc(vec![2, 2, 4, 64, 5, 1, 1]);
+        let ctx = FilterContext {
+            filter: &f,
+            element_size: 4,
+            max_output: 64,
+        };
+        assert!(blosc_decode(&frame, &ctx).is_err());
+    }
+
+    /// A frame whose header claims a compressed size smaller than the
+    /// header itself, not stored raw: an error, not an arithmetic overflow
+    /// (it panicked in debug builds).
+    #[test]
+    fn frame_size_below_the_header_is_an_error() {
+        let mut frame = vec![2u8, 1, 1 << 5, 4];
+        for v in [64u32, 64, 8] {
+            frame.extend_from_slice(&v.to_le_bytes());
+        }
+        frame.extend_from_slice(&[0; 40]);
+        assert!(blosc_decompress(&frame, 1000).is_err());
+        for cbytes in 0..16u32 {
+            frame[12..16].copy_from_slice(&cbytes.to_le_bytes());
+            assert!(blosc_decompress(&frame, 1000).is_err(), "cbytes={cbytes}");
+        }
+    }
+
+    /// A BloscLZ frame (our encoder cannot write one): a single block,
+    /// one stream, no shuffle.
+    fn blosclz_frame() -> Vec<u8> {
+        let stream = [2, b'a', b'b', b'c', (6 << 5), 2, 0, b'Z'];
+        let mut f = vec![2u8, 1, 0, 1];
+        for v in [12u32, 12, (HEADER + 4 + 4 + stream.len()) as u32] {
+            f.extend_from_slice(&v.to_le_bytes());
+        }
+        f.extend_from_slice(&((HEADER + 4) as u32).to_le_bytes());
+        f.extend_from_slice(&(stream.len() as u32).to_le_bytes());
+        f.extend_from_slice(&stream);
+        f
+    }
+
+    /// Random and mutated frames, every codec and shuffle: errors are fine,
+    /// panics are not.
+    #[test]
+    fn fuzzed_frames_never_panic() {
+        let limit = 6000;
+        let data: Vec<u8> = (0..1500u32).flat_map(|i| (i / 5).to_le_bytes()).collect();
+        let mut seeds = vec![blosclz_frame()];
+        for codec in [1u32, 3, 4, 5] {
+            for shuffle in [0u32, 1, 2] {
+                for (ts, n) in [(4usize, data.len()), (4, 520), (1, 300), (2, 4)] {
+                    let f = desc(vec![2, 2, ts as u32, 0, 5, shuffle, codec]);
+                    let ctx = FilterContext {
+                        filter: &f,
+                        element_size: ts,
+                        max_output: n,
+                    };
+                    seeds.push(blosc_encode(&data[..n], &ctx).unwrap());
+                }
+            }
+        }
+        // Stored raw.
+        let f = desc(vec![2, 2, 4, 0, 0, 1, 1]);
+        let ctx = FilterContext {
+            filter: &f,
+            element_size: 4,
+            max_output: 64,
+        };
+        seeds.push(blosc_encode(&data[..64], &ctx).unwrap());
+        crate::test_fuzz::fuzz_decoder(0xb10, &seeds, 30_000, limit, |s| {
+            blosc_decompress(s, limit)
+        });
+    }
+
+    /// BloscLZ streams on their own, random and mutated.
+    #[test]
+    fn fuzzed_blosclz_streams_never_panic() {
+        let seed = blosclz_frame()[HEADER + 8..].to_vec();
+        let mut out = [0u8; 64];
+        crate::test_fuzz::fuzz_decoder(0xb11, &[seed], 30_000, 64, |s| {
+            let n = blosclz_decompress(s, &mut out);
+            if n == 0 {
+                Err(err("malformed"))
+            } else {
+                Ok(out[..n].to_vec())
+            }
+        });
+    }
+}

crates/clawhdf5-format/src/filters_bzip2.rs

@@ -0,0 +1,132 @@
+//! bzip2 (HDF5 filter 307, PyTables' `H5Zbzip2.c`, hdf5plugin's `BZip2`).
+//!
+//! The chunk is one bzip2 stream; `cd_values[0]` is the block size (1-9,
+//! the compression level). Decoded with the `bzip2` crate's default backend,
+//! `libbz2-rs-sys`, a pure-Rust port of libbzip2.
+
+use crate::error::FormatError;
+use crate::filter_registry::FilterContext;
+
+fn err(msg: &str) -> FormatError {
+    FormatError::DecompressionError(format!("bzip2: {msg}"))
+}
+
+/// Decode a bzip2-filtered chunk, refusing output beyond the chunk size.
+pub(crate) fn bzip2_decode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    use bzip2::{Decompress, Status};
+    let limit = ctx.output_limit();
+    let max_capacity = limit.saturating_add(1);
+    let hint = if ctx.max_output != 0 {
+        ctx.max_output
+    } else {
+        input.len().saturating_mul(4)
+    };
+    let mut out = Vec::new();
+    out.try_reserve_exact(hint.clamp(1, max_capacity))
+        .map_err(|_| err("cannot allocate the output buffer"))?;
+    let mut dec = Decompress::new(false);
+    loop {
+        let (in_before, out_before) = (dec.total_in(), dec.total_out());
+        let status = dec
+            .decompress_vec(&input[in_before as usize..], &mut out)
+            .map_err(|e| err(&e.to_string()))?;
+        if out.len() > limit {
+            return Err(err("output exceeds the chunk size"));
+        }
+        if status == Status::StreamEnd {
+            return Ok(out);
+        }
+        if out.len() == out.capacity() {
+            let grow = out
+                .capacity()
+                .min(max_capacity.saturating_sub(out.capacity()))
+                .max(1);
+            out.try_reserve_exact(grow)
+                .map_err(|_| err("cannot allocate the output buffer"))?;
+        } else if dec.total_in() as usize >= input.len()
+            || (dec.total_in(), dec.total_out()) == (in_before, out_before)
+        {
+            return Err(err("truncated stream"));
+        }
+    }
+}
+
+/// Encode a chunk as one bzip2 stream at block size `cd_values[0]`
+/// (default 9, as hdf5plugin).
+pub(crate) fn bzip2_encode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    use bzip2::{Action, Compress, Compression, Status};
+    let level = ctx.client_data().first().copied().unwrap_or(9).clamp(1, 9);
+    let cerr = |m: String| FormatError::CompressionError(format!("bzip2: {m}"));
+    let mut enc = Compress::new(Compression::new(level), 0);
+    // bzip2's worst case is about 1% + 600 bytes over the input.
+    let mut out = Vec::with_capacity(input.len() + input.len() / 100 + 600);
+    loop {
+        let consumed = enc.total_in() as usize;
+        let status = enc
+            .compress_vec(&input[consumed..], &mut out, Action::Finish)
+            .map_err(|e| cerr(e.to_string()))?;
+        if status == Status::StreamEnd {
+            return Ok(out);
+        }
+        if out.len() == out.capacity() {
+            out.reserve(out.capacity().max(4096));
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::filter_pipeline::{FILTER_BZIP2, FilterDescription};
+
+    fn desc(level: u32) -> FilterDescription {
+        FilterDescription {
+            filter_id: FILTER_BZIP2,
+            name: None,
+            flags: 0,
+            client_data: vec![level],
+        }
+    }
+
+    #[test]
+    fn round_trips_and_bounds() {
+        let data: Vec<u8> = (0..100_000u32)
+            .flat_map(|i| (i % 777).to_le_bytes())
+            .collect();
+        for level in [1, 5, 9] {
+            let f = desc(level);
+            let ctx = FilterContext {
+                filter: &f,
+                element_size: 4,
+                max_output: data.len(),
+            };
+            let enc = bzip2_encode(&data, &ctx).unwrap();
+            assert!(enc.len() < data.len() / 4);
+            assert_eq!(bzip2_decode(&enc, &ctx).unwrap(), data);
+            // Truncated, and larger than the chunk: errors, not data.
+            assert!(bzip2_decode(&enc[..enc.len() / 2], &ctx).is_err());
+            let small = FilterContext {
+                max_output: data.len() - 1,
+                ..ctx
+            };
+            assert!(bzip2_decode(&enc, &small).is_err());
+        }
+    }
+
+    /// Random and mutated streams: errors are fine, panics are not.
+    #[test]
+    fn fuzzed_streams_never_panic() {
+        let f = desc(9);
+        let data: Vec<u8> = (0..4000u32).flat_map(|i| (i % 91).to_le_bytes()).collect();
+        let ctx = FilterContext {
+            filter: &f,
+            element_size: 4,
+            max_output: data.len(),
+        };
+        let seeds = vec![
+            bzip2_encode(&data, &ctx).unwrap(),
+            bzip2_encode(&data[..40], &ctx).unwrap(),
+        ];
+        crate::test_fuzz::fuzz_decoder(0xb2, &seeds, 3_000, data.len(), |s| bzip2_decode(s, &ctx));
+    }
+}

crates/clawhdf5-format/src/filters_lzf.rs

@@ -0,0 +1,260 @@
+//! LZF (HDF5 filter 32000) — h5py's built-in compression filter
+//! (`compression="lzf"`), in pure Rust.
+//!
+//! The chunk is one raw LZF stream (liblzf 3.x format, no header). The
+//! stream is a sequence of instructions, each starting with a control byte:
+//!
+//! * `000LLLLL` — a literal run: the next `L + 1` bytes (1..=32) are copied.
+//! * `LLLOOOOO [E] OOOOOOOO` — a back reference: copy `len + 2` bytes from
+//!   `distance` bytes back, where `len` is the top three bits (1..=6), or
+//!   `7 + E` when they are all ones, and `distance` is the 13-bit offset
+//!   (high five bits in the control byte, low eight in the last byte) plus 1.
+//!
+//! h5py's filter (`lzf_filter.c`) records the chunk's size in bytes in
+//! `cd_values[2]` (slots 0 and 1 hold the filter and liblzf versions) and
+//! sizes its output buffer from it.
+
+#[cfg(not(feature = "std"))]
+extern crate alloc;
+#[cfg(not(feature = "std"))]
+use alloc::{format, vec, vec::Vec};
+
+use crate::error::FormatError;
+use crate::filter_registry::FilterContext;
+
+/// `H5PY_FILTER_LZF_VERSION`, written to `cd_values[0]`.
+pub const LZF_FILTER_VERSION: u32 = 4;
+/// `LZF_VERSION` (liblzf 1.5), written to `cd_values[1]`.
+pub const LZF_API_VERSION: u32 = 0x0105;
+
+const MAX_LITERAL: usize = 32;
+const MAX_OFFSET: usize = 1 << 13;
+const MAX_REF: usize = (1 << 8) + (1 << 3);
+const HASH_LOG: u32 = 14;
+
+fn err(msg: &str) -> FormatError {
+    FormatError::DecompressionError(format!("lzf: {msg}"))
+}
+
+/// Decode an LZF-filtered chunk.
+pub(crate) fn lzf_decode(input: &[u8], ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    let limit = ctx.output_limit();
+    let hint = match ctx.client_data().get(2) {
+        Some(&n) if n != 0 => n as usize,
+        _ => input.len().saturating_mul(2),
+    };
+    lzf_decompress(input, hint.min(limit), limit)
+}
+
+/// Decompress a raw LZF stream, refusing to produce more than `limit` bytes.
+pub fn lzf_decompress(
+    input: &[u8],
+    size_hint: usize,
+    limit: usize,
+) -> Result<Vec<u8>, FormatError> {
+    let mut out: Vec<u8> = Vec::new();
+    out.try_reserve(size_hint)
+        .map_err(|_| err("cannot allocate the output buffer"))?;
+    let mut ip = 0usize;
+    while ip < input.len() {
+        let ctrl = input[ip] as usize;
+        ip += 1;
+        if ctrl < 32 {
+            let run = ctrl + 1;
+            let lit = input
+                .get(ip..ip + run)
+                .ok_or_else(|| err("literal run past the end of the input"))?;
+            if out.len() + run > limit {
+                return Err(err("output exceeds the chunk size"));
+            }
+            out.extend_from_slice(lit);
+            ip += run;
+        } else {
+            let mut len = ctrl >> 5;
+            if len == 7 {
+                len += *input
+                    .get(ip)
+                    .ok_or_else(|| err("truncated back reference"))?
+                    as usize;
+                ip += 1;
+            }
+            let low = *input
+                .get(ip)
+                .ok_or_else(|| err("truncated back reference"))? as usize;
+            ip += 1;
+            let distance = ((ctrl & 0x1f) << 8) + low + 1;
+            let len = len + 2;
+            if distance > out.len() {
+                return Err(err("back reference before the start of the output"));
+            }
+            if out.len() + len > limit {
+                return Err(err("output exceeds the chunk size"));
+            }
+            let start = out.len() - distance;
+            if distance >= len {
+                out.extend_from_within(start..start + len);
+            } else {
+                // Overlapping copy: repeats the last `distance` bytes.
+                for k in 0..len {
+                    let b = out[start + k];
+                    out.push(b);
+                }
+            }
+        }
+    }
+    Ok(out)
+}
+
+/// Encode a chunk with the LZF filter.
+pub(crate) fn lzf_encode(input: &[u8], _ctx: &FilterContext<'_>) -> Result<Vec<u8>, FormatError> {
+    Ok(lzf_compress(input))
+}
+
+fn hash3(b: &[u8]) -> usize {
+    let v = (u32::from(b[0]) << 16) | (u32::from(b[1]) << 8) | u32::from(b[2]);
+    (v.wrapping_mul(2_654_435_761) >> (32 - HASH_LOG)) as usize
+}
+
+fn flush_literals(out: &mut Vec<u8>, lit: &[u8]) {
+    for run in lit.chunks(MAX_LITERAL) {
+        out.push((run.len() - 1) as u8);
+        out.extend_from_slice(run);
+    }
+}
+
+/// Compress `input` into a raw LZF stream any liblzf decoder reads.
+///
+/// Incompressible input grows by one byte per 32. (h5py's own filter gives
+/// up on such a chunk and stores it unfiltered; storing the slightly larger
+/// stream is equally readable.)
+pub fn lzf_compress(input: &[u8]) -> Vec<u8> {
+    let n = input.len();
+    let mut out = Vec::with_capacity(n + n / MAX_LITERAL + 1);
+    let mut table = vec![0u32; 1 << HASH_LOG];
+    let mut lit_start = 0usize;
+    let mut i = 0usize;
+    while i + 2 < n {
+        let h = hash3(&input[i..]);
+        let cand = table[h] as usize;
+        table[h] = (i + 1) as u32;
+        if cand != 0 {
+            let r = cand - 1;
+            let distance = i - r;
+            if distance <= MAX_OFFSET && input[r..r + 3] == input[i..i + 3] {
+                let max_len = (n - i).min(MAX_REF);
+                let mut len = 3;
+                while len < max_len && input[r + len] == input[i + len] {
+                    len += 1;
+                }
+                flush_literals(&mut out, &input[lit_start..i]);
+                let code = len - 2;
+                let off = distance - 1;
+                if code < 7 {
+                    out.push(((code << 5) | (off >> 8)) as u8);
+                } else {
+                    out.push(((7 << 5) | (off >> 8)) as u8);
+                    out.push((code - 7) as u8);
+                }
+                out.push((off & 0xff) as u8);
+                // Index the positions the match covered so later data can
+                // refer back into it.
+                let end = i + len;
+                let mut j = i + 1;
+                while j < end && j + 2 < n {
+                    table[hash3(&input[j..])] = (j + 1) as u32;
+                    j += 1;
+                }
+                i = end;
+                lit_start = i;
+                continue;
+            }
+        }
+        i += 1;
+    }
+    flush_literals(&mut out, &input[lit_start..]);
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn round_trip(data: &[u8]) {
+        let c = lzf_compress(data);
+        assert_eq!(lzf_decompress(&c, data.len(), data.len()).unwrap(), data);
+    }
+
+    #[test]
+    fn round_trips() {
+        round_trip(b"");
+        round_trip(b"a");
+        round_trip(b"abcabcabcabcabcabcabcabcabcabcabcabc");
+        round_trip(&[7u8; 10_000]);
+        let noise: Vec<u8> = (0..70_000u32)
+            .map(|i| (i.wrapping_mul(2_654_435_761) >> 13) as u8)
+            .collect();
+        round_trip(&noise);
+        let ramp: Vec<u8> = (0..100_000u32)
+            .flat_map(|i| (i % 1000).to_le_bytes())
+            .collect();
+        round_trip(&ramp);
+    }
+
+    #[test]
+    fn compresses_repetitive_data() {
+        let data = [42u8; 4096];
+        assert!(lzf_compress(&data).len() < 100);
+    }
+
+    /// The chunk h5py 3.16's bundled liblzf writes for
+    /// `b"hello hello hello hello"` (read back with `read_direct_chunk`): a
+    /// 7-byte literal, a 14-byte back reference 6 bytes back (extended
+    /// length), and a 2-byte literal.
+    #[test]
+    fn decodes_liblzf_output() {
+        let stream = b"\x06hello h\xe0\x05\x05\x01lo";
+        assert_eq!(
+            lzf_decompress(stream, 23, 23).unwrap(),
+            b"hello hello hello hello"
+        );
+    }
+
+    #[test]
+    fn rejects_corrupt_streams() {
+        // Back reference before the start.
+        assert!(lzf_decompress(&[0x20, 0x00], 10, 10).is_err());
+        // Literal run past the end.
+        assert!(lzf_decompress(&[0x05, 1, 2], 10, 10).is_err());
+        // Output over the limit.
+        let c = lzf_compress(&[1u8; 100]);
+        assert!(lzf_decompress(&c, 10, 99).is_err());
+    }
+
+    /// Random and mutated streams: errors are fine, panics are not.
+    #[test]
+    fn fuzzed_streams_never_panic() {
+        let seeds: Vec<Vec<u8>> = [
+            b"hello hello hello hello".to_vec(),
+            vec![7u8; 3000],
+            (0..2000u32).flat_map(|i| (i % 37).to_le_bytes()).collect(),
+            (0..500u32)
+                .map(|i| (i.wrapping_mul(2_654_435_761) >> 13) as u8)
+                .collect(),
+        ]
+        .iter()
+        .map(|d| lzf_compress(d))
+        .collect();
+        for limit in [0usize, 23, 4096, 8000] {
+            crate::test_fuzz::fuzz_decoder(
+                0x1f2 + limit as u64,
+                &seeds[..1],
+                5_000,
+                limit.max(23),
+                |s| lzf_decompress(s, limit, limit.max(23)),
+            );
+        }
+        crate::test_fuzz::fuzz_decoder(0x1f3, &seeds, 20_000, 8000, |s| {
+            lzf_decompress(s, 8000, 8000)
+        });
+    }
+}

crates/clawhdf5-format/src/filters_szip.rs

@@ -1,19 +1,40 @@
 //! SZIP (libaec Adaptive Entropy Coding) decompression.
 //!
 //! Gated by the `szip` feature which links against the system libaec library.
+//!
+//! libhdf5's SZIP filter (`H5Zszip.c`) prefixes each chunk with its
+//! uncompressed size and hands the rest to szlib's `SZ_BufftoBuffDecompress`.
+//! libaec implements that call (`sz_compat.c`) on top of `aec_buffer_decode`
+//! with some reshaping — 32/64-bit samples are coded as byte planes of 8-bit
+//! samples, and scanlines that are not a whole number of blocks are padded —
+//! which [`szip_decompress`] reproduces so its output matches libhdf5's.
 
 #[cfg(not(feature = "std"))]
 use alloc::vec::Vec;
 
 use crate::error::FormatError;
 
-/// Decompress SZIP-compressed data using libaec.
+/// `SZ_MSB_OPTION_MASK`: samples are big-endian.
+#[cfg(feature = "szip")]
+const SZ_MSB_OPTION_MASK: u32 = 16;
+/// `SZ_NN_OPTION_MASK`: nearest-neighbour preprocessing.
+#[cfg(feature = "szip")]
+const SZ_NN_OPTION_MASK: u32 = 32;
+
+/// Decompress one SZIP-filtered chunk.
 ///
-/// `cd` is the HDF5 SZIP filter client data (matches `H5Z_SZIP_PARM_*` indices):
-///   cd[0] = options mask  (`H5_SZIP_NN_OPTION_MASK = 0x20` enables NN preprocessing)
-///   cd[1] = pixels per block (H5Z_SZIP_PARM_PPB; 8, 10, 16, or 32)
-///   cd[2] = bits per sample  (H5Z_SZIP_PARM_BPP; element bit width)
-///   cd[3] = pixels per scan line (H5Z_SZIP_PARM_PPS; informational only)
+/// `cd` is the HDF5 SZIP filter client data (`H5Z_SZIP_PARM_*` indices):
+///   cd[0] = options mask (`SZ_*_OPTION_MASK`: 16 = MSB byte order,
+///           32 = nearest-neighbour preprocessing; K13/EC/LSB/RAW bits carry
+///           no decoding information for libaec)
+///   cd[1] = pixels per block
+///   cd[2] = bits per pixel (sample precision, rounded up to 32 or 64 above
+///           24 by libhdf5)
+///   cd[3] = pixels per scanline
+///
+/// The chunk is a 4-byte little-endian uncompressed size followed by the
+/// szlib stream.
+#[cfg_attr(not(feature = "szip"), allow(dead_code))]
 pub(crate) fn szip_decompress(
     _data: &[u8],
     _cd: &[u32],
@@ -33,62 +54,174 @@ pub(crate) fn szip_decompress(
 
 #[cfg(feature = "szip")]
 fn szip_decode_impl(data: &[u8], cd: &[u32], chunk_size: usize) -> Result<Vec<u8>, FormatError> {
-    if cd.len() < 3 {
-        return Err(FormatError::ChunkedReadError(
-            "szip: missing client data".into(),
-        ));
+    let err = |m: &str| FormatError::ChunkedReadError(format!("szip: {m}"));
+    if cd.len() < 4 {
+        return Err(err("missing client data"));
     }
     let options = cd[0];
-    let pixels_per_block = cd[1];
-    let bits_per_sample = cd[2]; // H5Z_SZIP_PARM_BPP
-    if bits_per_sample == 0 || bits_per_sample > 32 {
-        return Err(FormatError::ChunkedReadError(
-            "szip: invalid bits per sample".into(),
-        ));
+    let pixels_per_block = cd[1] as usize;
+    let bits_per_pixel = cd[2];
+    let pixels_per_scanline = cd[3] as usize;
+    if !(1..=32).contains(&bits_per_pixel) && bits_per_pixel != 64 {
+        return Err(err("invalid bits per sample"));
     }
-    if chunk_size == 0 {
-        return Err(FormatError::ChunkedReadError(
-            "szip: unknown output size".into(),
-        ));
+    if pixels_per_block == 0 || pixels_per_scanline == 0 {
+        return Err(err("invalid block or scanline size"));
     }
-    if data.is_empty() {
-        return Err(FormatError::ChunkedReadError("szip: empty input".into()));
+    if data.len() < 4 {
+        return Err(err("chunk too short"));
     }
+    // H5Zszip.c: UINT32DECODE of the uncompressed size, then the stream.
+    let dest_len = u32::from_le_bytes([data[0], data[1], data[2], data[3]]) as usize;
+    let limit = if chunk_size != 0 {
+        chunk_size
+    } else {
+        crate::filters::MAX_DECOMPRESS_SIZE
+    };
+    if dest_len > limit {
+        return Err(err("declared size exceeds chunk size"));
+    }
+    let stream = &data[4..];
 
-    // Map HDF5 option mask to libaec flags.
-    // HDF5 always stores SZIP data in MSB order, so AEC_DATA_MSB is unconditional.
-    // H5_SZIP_NN_OPTION_MASK (0x20): NN differential preprocessing.
-    let mut flags: u32 = libaec_sys::AEC_DATA_MSB;
-    if options & 0x20 != 0 {
+    // --- libaec sz_compat.c: SZ_BufftoBuffDecompress ---
+    let rsi = pixels_per_scanline.div_ceil(pixels_per_block);
+    let mut flags = 0;
+    if options & SZ_MSB_OPTION_MASK != 0 {
+        flags |= libaec_sys::AEC_DATA_MSB;
+    }
+    if options & SZ_NN_OPTION_MASK != 0 {
         flags |= libaec_sys::AEC_DATA_PREPROCESS;
     }
+    let pad_scanline = !pixels_per_scanline.is_multiple_of(pixels_per_block);
+    let deinterleave = bits_per_pixel == 32 || bits_per_pixel == 64;
+    let bits_per_sample = if deinterleave { 8 } else { bits_per_pixel };
+    let pixel_size = match bits_per_sample {
+        17.. => 4,
+        9.. => 2,
+        _ => 1,
+    };
+    let scanlines = (dest_len / pixel_size).div_ceil(pixels_per_scanline);
+    let buf_size = if pad_scanline {
+        rsi.checked_mul(pixels_per_block)
+            .and_then(|n| n.checked_mul(pixel_size))
+            .and_then(|n| n.checked_mul(scanlines))
+            .filter(|&n| n <= crate::filters::MAX_DECOMPRESS_SIZE.max(limit))
+            .ok_or_else(|| err("scanline padding too large"))?
+    } else {
+        dest_len
+    };
 
-    let mut out = vec![0u8; chunk_size];
+    let mut buf = vec![0u8; buf_size];
     let mut strm = libaec_sys::AecStream::zeroed();
-    strm.next_in = data.as_ptr();
-    strm.avail_in = data.len();
-    strm.next_out = out.as_mut_ptr();
-    strm.avail_out = chunk_size;
+    strm.next_in = stream.as_ptr();
+    strm.avail_in = stream.len();
+    strm.next_out = buf.as_mut_ptr();
+    strm.avail_out = buf_size;
     strm.bits_per_sample = bits_per_sample;
-    strm.block_size = pixels_per_block;
-    strm.rsi = 128; // HDF5 default: 128 blocks per reference sample interval
+    strm.block_size = pixels_per_block as u32;
+    strm.rsi = rsi as u32;
     strm.flags = flags;
-
+    // SAFETY: next_in/avail_in and next_out/avail_out describe live buffers
+    // (`stream` and `buf`) that outlive the call.
     let result = unsafe { libaec_sys::aec_buffer_decode(&mut strm) };
     if result != 0 {
         return Err(FormatError::DecompressionError(format!(
             "szip: libaec error {result}"
         )));
     }
-    let decoded_len = chunk_size - strm.avail_out;
-    out.truncate(decoded_len);
+    let mut total_out = strm.total_out;
+    if pad_scanline {
+        let line = pixels_per_scanline * pixel_size;
+        let padded_line = rsi * pixels_per_block * pixel_size;
+        // remove_padding: compact each padded line down to `line` bytes.
+        let mut i = line;
+        let mut j = padded_line;
+        while j < total_out {
+            let end = (j + line).min(buf.len());
+            buf.copy_within(j..end, i);
+            i += line;
+            j += padded_line;
+        }
+        total_out = scanlines * line;
+    }
+    if total_out < dest_len {
+        return Err(err("stream decoded to fewer bytes than declared"));
+    }
+    buf.truncate(dest_len);
+    if deinterleave {
+        // deinterleave_buffer: byte planes back into words.
+        let w = (bits_per_pixel / 8) as usize;
+        let n = dest_len / w;
+        let mut out = vec![0u8; dest_len];
+        for i in 0..n {
+            for j in 0..w {
+                out[i * w + j] = buf[j * n + i];
+            }
+        }
         Ok(out)
+    } else {
+        Ok(buf)
+    }
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    #[cfg(feature = "szip")]
+    fn unhex(s: &str) -> Vec<u8> {
+        (0..s.len())
+            .step_by(2)
+            .map(|i| u8::from_str_radix(&s[i..i + 2], 16).unwrap())
+            .collect()
+    }
+
+    /// SZIP chunks written by libhdf5, decoded exactly as libhdf5 decodes
+    /// them. Each case: fixture, chunk byte offset and size (from h5py's
+    /// `get_chunk_info`), the filter's cd_values, and the chunk's values as
+    /// h5py reads them (file byte order, hex). Before the fix every one of
+    /// these came back as garbage or zeros (or "invalid bits per sample" for
+    /// 64-bit): the 4-byte size prefix was fed to libaec, 32/64-bit samples
+    /// were not de-interleaved from byte planes, the reference sample
+    /// interval was fixed at 128 instead of derived from the scanline, padded
+    /// scanlines were not unpadded, and LE data was decoded as MSB.
+    #[cfg(feature = "szip")]
+    #[test]
+    fn szip_decodes_libhdf5_chunks_exactly() {
+        /// (name, file, chunk offset, chunk size, cd_values, decoded hex)
+        type Case<'a> = (&'a str, &'a [u8], usize, usize, [u32; 4], &'a str);
+        let noencoder: &[u8] = include_bytes!("../tests/fixtures/filters/noencoder.h5");
+        let le_data: &[u8] = include_bytes!("../tests/fixtures/filters/le_data.h5");
+        let h5py: &[u8] = include_bytes!("../tests/fixtures/filters/szip_h5py.h5");
+        #[rustfmt::skip]
+        let cases: &[Case] = &[
+            // <i4, 10 px/scanline over 4 px/block: padded scanlines + byte planes.
+            ("noencoder /noencoder_szip_dset.h5", noencoder, 6040, 16, [168, 4, 32, 10],
+             "00000000010000000200000003000000040000000500000006000000070000000800000009000000"),
+            // <f4, LSB + NN.
+            ("le_data /Szip_float_data_le", le_data, 55224, 48, [169, 4, 32, 12],
+             "abaaaa3eabaa2a3f0000803fabaa2a3f0000803fabaaaa3f0000803fabaaaa3f5555d53fabaaaa3f5555d53f00000040"),
+            // >f4, MSB + NN.
+            ("le_data /Szip_float_data_be", le_data, 55396, 48, [177, 4, 32, 12],
+             "3eaaaaab3f2aaaab3f8000003f2aaaab3f8000003faaaaab3f8000003faaaaab3fd555553faaaaab3fd5555540000000"),
+            // <f8 (64-bit), NN.
+            ("szip_h5py /f8", h5py, 4016, 100, [169, 8, 64, 10],
+             "00000000000008c000000000000008c000000000000008c000000000000008c000000000000004c000000000000004c000000000000004c000000000000004c000000000000000c000000000000000c000000000000000c000000000000000c0000000000000f8bf000000000000f8bf000000000000f8bf000000000000f8bf000000000000f0bf000000000000f0bf000000000000f0bf000000000000f0bf000000000000e0bf000000000000e0bf000000000000e0bf000000000000e0bf0000000000000000000000000000000000000000000000000000000000000000000000000000e03f000000000000e03f000000000000e03f000000000000e03f000000000000f03f000000000000f03f000000000000f03f000000000000f03f000000000000f83f000000000000f83f000000000000f83f000000000000f83f"),
+            // <i8 (64-bit), entropy coding without NN.
+            ("szip_h5py /i8", h5py, 4188, 53, [141, 4, 64, 10],
+             "000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000300000000000000030000000000000003000000000000000300000000000000030000000000000003000000000000000300000000000000030000000000000006000000000000000600000000000000060000000000000006000000000000000600000000000000060000000000000006000000000000000600000000000000090000000000000009000000000000000900000000000000090000000000000009000000000000000900000000000000090000000000000009000000000000000c000000000000000c000000000000000c000000000000000c000000000000000c000000000000000c000000000000000c000000000000000c00000000000000"),
+            // <u2, 35 px/scanline over 8 px/block: padded scanlines, 16-bit samples.
+            ("szip_h5py /u2", h5py, 4308, 43, [169, 8, 16, 35],
+             "00000000000000006100610061006100c200c200c200c20023012301230123018401840184018401e501e501e501e5014602460246024602a702a702a702a702080308030803"),
+        ];
+        for (name, file, off, len, cd, want) in cases {
+            let want = unhex(want);
+            let got = szip_decompress(&file[*off..off + len], cd, want.len())
+                .unwrap_or_else(|e| panic!("{name}: {e:?}"));
+            assert_eq!(got, want, "{name}");
+        }
+    }
+
     #[test]
     fn szip_disabled_returns_unsupported() {
         #[cfg(not(feature = "szip"))]
@@ -132,6 +265,8 @@ mod tests {
         assert_eq!(rc, 0, "aec_buffer_encode failed: {rc}");
         let enc_len = encoded.len() - enc.avail_out;
         encoded.truncate(enc_len);
+        // H5Zszip.c prefixes the stream with the uncompressed size.
+        encoded.splice(0..0, (original.len() as u32).to_le_bytes());
 
         // Decode through our public interface.
         // cd[0]=0 (no NN bit 0x20), cd[1]=8 (ppb), cd[2]=8 (bpp), cd[3]=1024 (pps).
@@ -163,6 +298,8 @@ mod tests {
         assert_eq!(rc, 0, "aec_buffer_encode with NN failed: {rc}");
         let enc_len = encoded.len() - enc.avail_out;
         encoded.truncate(enc_len);
+        // H5Zszip.c prefixes the stream with the uncompressed size.
+        encoded.splice(0..0, (original.len() as u32).to_le_bytes());
 
         // cd[0] = 0x20 (H5_SZIP_NN_OPTION_MASK) → decoder must set AEC_DATA_PREPROCESS.
         let cd = [0x20u32, 8, 8, 1024];

crates/clawhdf5-format/src/fixed_array.rs

@@ -6,6 +6,7 @@ extern crate alloc;
 #[cfg(not(feature = "std"))]
 use alloc::{format, vec, vec::Vec};
 
+use crate::chunk_grid::ChunkGrid;
 use crate::chunked_read::ChunkInfo;
 use crate::error::FormatError;
 
@@ -151,13 +152,13 @@ pub fn read_fixed_array_chunks(
     file_data: &[u8],
     header: &FixedArrayHeader,
     dataset_dims: &[u64],
+    max_dims: Option<&[u64]>,
     chunk_dimensions: &[u32],
     element_size: u32,
     offset_size: u8,
     _length_size: u8,
 ) -> Result<Vec<ChunkInfo>, FormatError> {
     let db_offset = header.data_block_address as usize;
-    let rank = chunk_dimensions.len();
 
     // Parse data block header: FADB(4) + version(1) + client_id(1) + header_address(offset_size)
     let db_header_size = 4 + 1 + 1 + offset_size as usize;
@@ -198,19 +199,10 @@ pub fn read_fixed_array_chunks(
             ))
     };
 
-    // Compute chunk offsets based on index.
-    // Chunks are stored in row-major order within the dataset space.
-    let mut num_chunks_per_dim = Vec::with_capacity(rank);
-    for d_idx in 0..rank {
-        let ch_dim = chunk_dimensions[d_idx] as u64;
-        if ch_dim == 0 {
-            return Err(FormatError::ChunkedReadError(
-                "chunk dimension is zero".into(),
-            ));
-        }
-        let ds_dim = dataset_dims[d_idx];
-        num_chunks_per_dim.push(ds_dim.div_ceil(ch_dim));
-    }
+    // The index is laid out over the chunk grid of the *maximum* dimensions
+    // (row-major), so a dataset smaller than its maxshape has gaps.
+    let dims_u64: Vec<u64> = chunk_dimensions.iter().map(|&d| d as u64).collect();
+    let grid = ChunkGrid::fixed_array(dataset_dims, max_dims, &dims_u64)?;
 
     let chunk_byte_size: u64 =
         chunk_dimensions.iter().map(|&d| d as u64).product::<u64>() * element_size as u64;
@@ -226,7 +218,11 @@ pub fn read_fixed_array_chunks(
                 header.element_size,
                 chunk_byte_size,
             )? {
-                let offsets = index_to_chunk_offsets(i, &num_chunks_per_dim, chunk_dimensions);
+                // A slot beyond the current extent is ignored, as the
+                // library does.
+                let Some(offsets) = grid.offsets(i as u64) else {
+                    return Ok(());
+                };
                 chunks.push(ChunkInfo {
                     chunk_size,
                     filter_mask,
@@ -367,27 +363,6 @@ fn parse_fa_element(
     }
 }
 
-/// Convert a linear chunk index to N-dimensional chunk offsets in dataset space.
-fn index_to_chunk_offsets(
-    index: usize,
-    num_chunks_per_dim: &[u64],
-    chunk_dimensions: &[u32],
-) -> Vec<u64> {
-    let rank = num_chunks_per_dim.len();
-    let mut offsets = vec![0u64; rank];
-    let mut remaining = index as u64;
-    for d in (0..rank).rev() {
-        let nchunks = num_chunks_per_dim[d];
-        if nchunks == 0 {
-            continue;
-        }
-        let chunk_idx = remaining % nchunks;
-        remaining /= nchunks;
-        offsets[d] = chunk_idx * chunk_dimensions[d] as u64;
-    }
-    offsets
-}
-
 /// Read a variable-length little-endian unsigned integer.
 fn read_variable_length(data: &[u8], size: usize) -> Result<u64, FormatError> {
     if size > 8 || data.len() < size {
@@ -416,44 +391,21 @@ mod tests {
 
     #[test]
     fn index_to_offsets_1d() {
-        let num_chunks = vec![5u64];
-        let chunk_dims = vec![20u32];
-        assert_eq!(index_to_chunk_offsets(0, &num_chunks, &chunk_dims), vec![0]);
-        assert_eq!(
-            index_to_chunk_offsets(1, &num_chunks, &chunk_dims),
-            vec![20]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(4, &num_chunks, &chunk_dims),
-            vec![80]
-        );
+        let g = ChunkGrid::fixed_array(&[100], None, &[20]).unwrap();
+        assert_eq!(g.offsets(0).unwrap(), vec![0]);
+        assert_eq!(g.offsets(1).unwrap(), vec![20]);
+        assert_eq!(g.offsets(4).unwrap(), vec![80]);
     }
 
     #[test]
     fn index_to_offsets_2d() {
         // 10x6 dataset with 4x3 chunks => ceil(10/4)=3, ceil(6/3)=2 => 6 chunks
-        let num_chunks = vec![3u64, 2];
-        let chunk_dims = vec![4u32, 3];
-        assert_eq!(
-            index_to_chunk_offsets(0, &num_chunks, &chunk_dims),
-            vec![0, 0]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(1, &num_chunks, &chunk_dims),
-            vec![0, 3]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(2, &num_chunks, &chunk_dims),
-            vec![4, 0]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(3, &num_chunks, &chunk_dims),
-            vec![4, 3]
-        );
-        assert_eq!(
-            index_to_chunk_offsets(5, &num_chunks, &chunk_dims),
-            vec![8, 3]
-        );
+        let g = ChunkGrid::fixed_array(&[10, 6], None, &[4, 3]).unwrap();
+        assert_eq!(g.offsets(0).unwrap(), vec![0, 0]);
+        assert_eq!(g.offsets(1).unwrap(), vec![0, 3]);
+        assert_eq!(g.offsets(2).unwrap(), vec![4, 0]);
+        assert_eq!(g.offsets(3).unwrap(), vec![4, 3]);
+        assert_eq!(g.offsets(5).unwrap(), vec![8, 3]);
     }
 
     #[test]
@@ -517,7 +469,7 @@ mod tests {
 
         let read = |f: &[u8], fahd: usize| -> Result<Vec<ChunkInfo>, FormatError> {
             let h = FixedArrayHeader::parse(f, fahd, 8, 8)?;
-            read_fixed_array_chunks(f, &h, &[60], &[20], 8, 8, 8)
+            read_fixed_array_chunks(f, &h, &[60], None, &[20], 8, 8, 8)
         };
 
         let (clean, fahd) = build();
@@ -562,7 +514,7 @@ mod tests {
         let db = 0x100usize;
         buf[db..db + 4].copy_from_slice(b"FADB");
         let header = FixedArrayHeader::parse(&buf, fahd, 8, 8).unwrap();
-        let r = read_fixed_array_chunks(&buf, &header, &[100], &[20], 8, 8, 8);
+        let r = read_fixed_array_chunks(&buf, &header, &[100], None, &[20], 8, 8, 8);
         assert!(r.is_err());
     }
 
@@ -579,7 +531,7 @@ mod tests {
         stamp_checksum(&mut buf, fahd, fahd + 24);
         buf[0x80..0x84].copy_from_slice(b"FADB");
         let header = FixedArrayHeader::parse(&buf, fahd, 8, 8).unwrap();
-        let r = read_fixed_array_chunks(&buf, &header, &[100], &[20], 8, 8, 8);
+        let r = read_fixed_array_chunks(&buf, &header, &[100], None, &[20], 8, 8, 8);
         assert!(r.is_err());
     }
 
@@ -602,7 +554,7 @@ mod tests {
             data_block_address: (usize::MAX - 4) as u64,
         };
         let buf = vec![0u8; 64];
-        let r = read_fixed_array_chunks(&buf, &header, &[100], &[20], 8, 8, 8);
+        let r = read_fixed_array_chunks(&buf, &header, &[100], None, &[20], 8, 8, 8);
         assert!(r.is_err());
     }
 
@@ -664,6 +616,7 @@ mod tests {
             &file_data,
             &header,
             &ds_dims,
+            None,
             &chunk_dims,
             8,
             offset_size,
@@ -740,6 +693,7 @@ mod tests {
             &file_data,
             &header,
             &ds_dims,
+            None,
             &chunk_dims,
             8,
             offset_size,
@@ -840,6 +794,7 @@ mod tests {
             &file_data,
             &header,
             &ds_dims,
+            None,
             &chunk_dims,
             8,
             offset_size,

crates/clawhdf5-format/src/fractal_heap.rs

@@ -1,12 +1,14 @@
 //! HDF5 Fractal Heap parsing for v2 group link storage.
 
 #[cfg(not(feature = "std"))]
-use alloc::vec::Vec;
+use alloc::{format, vec::Vec};
 
 #[cfg(feature = "checksum")]
 use byteorder::{ByteOrder, LittleEndian};
 
+use crate::btree_v2::{BTreeV2Header, collect_btree_v2_records};
 use crate::error::FormatError;
+use crate::filter_pipeline::FilterPipeline;
 
 /// Parsed fractal heap header (signature "FRHP").
 #[derive(Debug, Clone)]
@@ -33,6 +35,23 @@ pub struct FractalHeapHeader {
     pub current_rows_in_root_indirect_block: u16,
     /// Total number of managed objects.
     pub managed_objects_count: u64,
+    /// Address of the v2 B-tree indexing "huge" objects (undefined address
+    /// when the heap has none). Huge objects are those larger than
+    /// `max_managed_object_size`; they live outside the heap's blocks.
+    pub huge_btree_address: u64,
+    /// The heap's I/O filter pipeline, if it has one. It applies to managed
+    /// direct blocks and to huge objects.
+    pub filter_pipeline: Option<FilterPipeline>,
+    /// Stored (filtered) size of the root direct block; meaningful only when
+    /// the heap is filtered and its root is a direct block.
+    pub root_direct_block_filtered_size: u64,
+    /// Filter mask of the root direct block (bit *i* set = filter *i*
+    /// skipped); meaningful only when the heap is filtered.
+    pub root_direct_block_filter_mask: u32,
+    /// Size of addresses in the file ("Size of Offsets").
+    pub offset_size: u8,
+    /// Size of lengths in the file ("Size of Lengths").
+    pub length_size: u8,
 }
 
 fn read_offset(data: &[u8], pos: usize, size: u8) -> Result<u64, FormatError> {
@@ -79,6 +98,38 @@ fn is_undefined(val: u64, offset_size: u8) -> bool {
     }
 }
 
+/// Little-endian unsigned integer of up to 8 bytes.
+fn le_uint(bytes: &[u8]) -> u64 {
+    bytes
+        .iter()
+        .take(8)
+        .enumerate()
+        .fold(0u64, |acc, (i, &b)| acc | (u64::from(b) << (i * 8)))
+}
+
+fn heap_error(msg: &str) -> FormatError {
+    FormatError::ChunkedReadError(format!("fractal heap: {msg}"))
+}
+
+/// Heap ID type, from bits 4-5 of an ID's first byte (libhdf5's
+/// `H5HF_ID_TYPE_MASK`, 0x30); bits 6-7 are the ID version, which must be 0.
+const HEAP_ID_MANAGED: u8 = 0;
+const HEAP_ID_HUGE: u8 = 1;
+const HEAP_ID_TINY: u8 = 2;
+
+/// The type (0 managed, 1 huge, 2 tiny) of a heap ID from its first byte,
+/// refusing an ID version other than 0.
+fn heap_id_type(first: u8) -> Result<u8, FormatError> {
+    if first >> 6 != 0 {
+        return Err(heap_error("unsupported heap ID version"));
+    }
+    Ok((first >> 4) & 0x03)
+}
+
+/// v2 B-tree record types indexing a heap's huge objects.
+const BTREE_HUGE_INDIRECT: u8 = 1;
+const BTREE_HUGE_INDIRECT_FILTERED: u8 = 2;
+
 impl FractalHeapHeader {
     /// Parse a fractal heap header at the given offset.
     pub fn parse(
@@ -122,11 +173,17 @@ impl FractalHeapHeader {
         ]);
         pos += 4;
 
-        // Skip several fixed fields: next_huge_object_id(ls), btree_huge_objects_address(os),
-        // free_space_managed_blocks(ls), managed_block_free_space_manager_address(os),
+        // next_huge_object_id (length_size)
+        ensure_len(file_data, pos, ls)?;
+        pos += ls;
+        // btree_huge_objects_address (offset_size)
+        let huge_btree_address = read_offset(file_data, pos, offset_size)?;
+        pos += os;
+
+        // Skip: free_space_managed_blocks(ls), managed_block_free_space_manager_address(os),
         // managed_space_in_heap(ls), allocated_managed_space_in_heap(ls),
         // direct_block_allocation_iterator_offset(ls)
-        let skip_size = 5 * ls + 2 * os;
+        let skip_size = 4 * ls + os;
         ensure_len(file_data, pos, skip_size)?;
         pos += skip_size;
 
@@ -134,14 +191,9 @@ impl FractalHeapHeader {
         let managed_objects_count = read_offset(file_data, pos, length_size)?;
         pos += ls;
 
-        // huge_objects_size (length_size)
-        pos += ls;
-        // huge_objects_count (length_size)
-        pos += ls;
-        // tiny_objects_size (length_size)
-        pos += ls;
-        // tiny_objects_count (length_size)
-        pos += ls;
+        // huge_objects_size, huge_objects_count, tiny_objects_size,
+        // tiny_objects_count (length_size each)
+        pos += 4 * ls;
 
         // table_width (2)
         ensure_len(file_data, pos, 2)?;
@@ -175,16 +227,28 @@ impl FractalHeapHeader {
         ensure_len(file_data, pos, 2)?;
         let current_rows_in_root_indirect_block =
             u16::from_le_bytes([file_data[pos], file_data[pos + 1]]);
-        #[allow(unused_variables, unused_mut, unused_assignments)]
-        let mut pos = pos + 2;
+        pos += 2;
 
-        // Skip IO filter encoded info if present
+        // With I/O filters: root direct block's filtered size (length_size),
+        // its filter mask (4), then the encoded filter pipeline message.
+        let mut filter_pipeline = None;
+        let mut root_direct_block_filtered_size = 0;
+        let mut root_direct_block_filter_mask = 0;
         if io_filter_encoded_length > 0 {
-            // root_block_filter_info_size (length_size) + filter_mask (4)
-            #[allow(unused_assignments)]
-            {
-                pos += ls + 4;
-            }
+            root_direct_block_filtered_size = read_offset(file_data, pos, length_size)?;
+            pos += ls;
+            ensure_len(file_data, pos, 4)?;
+            root_direct_block_filter_mask = u32::from_le_bytes([
+                file_data[pos],
+                file_data[pos + 1],
+                file_data[pos + 2],
+                file_data[pos + 3],
+            ]);
+            pos += 4;
+            let n = io_filter_encoded_length as usize;
+            ensure_len(file_data, pos, n)?;
+            filter_pipeline = Some(FilterPipeline::parse(&file_data[pos..pos + n])?);
+            pos += n;
         }
 
         // Validate header checksum
@@ -200,6 +264,8 @@ impl FractalHeapHeader {
                 });
             }
         }
+        #[cfg(not(feature = "checksum"))]
+        let _ = pos;
 
         Ok(FractalHeapHeader {
             heap_id_length,
@@ -213,13 +279,19 @@ impl FractalHeapHeader {
             root_block_address,
             current_rows_in_root_indirect_block,
             managed_objects_count,
+            huge_btree_address,
+            filter_pipeline,
+            root_direct_block_filtered_size,
+            root_direct_block_filter_mask,
+            offset_size,
+            length_size,
         })
     }
 
     /// Decode a managed heap ID into (offset_in_heap, object_length).
     ///
     /// The heap ID layout for managed objects (type 0):
-    /// - Byte 0: bits 6-7 = type (0), bits 4-5 = version (0), bits 0-3 = reserved
+    /// - Byte 0: bits 6-7 = version (0), bits 4-5 = type (0), bits 0-3 = reserved
     /// - Bytes 1+: offset (max_heap_size bits, LE) then length (remaining bits, LE)
     pub fn decode_managed_id(&self, id_bytes: &[u8]) -> Result<(u64, u64), FormatError> {
         if id_bytes.is_empty() {
@@ -229,8 +301,8 @@ impl FractalHeapHeader {
             });
         }
 
-        let id_type = (id_bytes[0] >> 6) & 0x03;
-        if id_type != 0 {
+        let id_type = heap_id_type(id_bytes[0])?;
+        if id_type != HEAP_ID_MANAGED {
             return Err(FormatError::InvalidHeapIdType(id_type));
         }
 
@@ -269,12 +341,183 @@ impl FractalHeapHeader {
         Ok((heap_offset, length_val))
     }
 
-    /// Read a managed object from the heap given its raw heap ID bytes.
+    /// Read any object from the heap given its raw heap ID bytes: managed
+    /// (stored in the heap's blocks), huge (stored outside them, found
+    /// directly from the ID or through the huge-object v2 B-tree, optionally
+    /// filtered) or tiny (stored in the ID itself).
+    ///
+    /// Despite its name this accepts every ID type; `offset_size` must match
+    /// the one the header was parsed with.
     pub fn read_managed_object(
         &self,
         file_data: &[u8],
         id_bytes: &[u8],
         offset_size: u8,
+    ) -> Result<Vec<u8>, FormatError> {
+        let Some(&first) = id_bytes.first() else {
+            return Err(FormatError::UnexpectedEof {
+                expected: 1,
+                available: 0,
+            });
+        };
+        match heap_id_type(first)? {
+            HEAP_ID_MANAGED => self.read_heap_managed(file_data, id_bytes, offset_size),
+            HEAP_ID_HUGE => self.read_huge_object(file_data, id_bytes),
+            HEAP_ID_TINY => self.read_tiny_object(id_bytes),
+            other => Err(FormatError::InvalidHeapIdType(other)),
+        }
+    }
+
+    /// Whether a huge object's ID holds its address and length directly
+    /// (libhdf5 does this when they fit in the ID), rather than a key into
+    /// the huge-object B-tree.
+    fn huge_ids_direct(&self) -> bool {
+        let room = usize::from(self.heap_id_length).saturating_sub(1);
+        let os = usize::from(self.offset_size);
+        let ls = usize::from(self.length_size);
+        if self.filter_pipeline.is_some() {
+            room >= os + ls + 4 + ls
+        } else {
+            room >= os + ls
+        }
+    }
+
+    /// Read a huge object (heap ID type 1).
+    fn read_huge_object(&self, file_data: &[u8], id: &[u8]) -> Result<Vec<u8>, FormatError> {
+        let os = usize::from(self.offset_size);
+        let ls = usize::from(self.length_size);
+        // (address, stored length, filter mask, decoded length); the last two
+        // only matter for a filtered heap.
+        let (addr, stored_len, mask, mem_len) = if self.huge_ids_direct() {
+            let body = &id[1..];
+            let need = if self.filter_pipeline.is_some() {
+                os + ls + 4 + ls
+            } else {
+                os + ls
+            };
+            ensure_len(body, 0, need)?;
+            let addr = le_uint(&body[..os]);
+            let len = le_uint(&body[os..os + ls]);
+            if self.filter_pipeline.is_some() {
+                let mask = u32::from_le_bytes([
+                    body[os + ls],
+                    body[os + ls + 1],
+                    body[os + ls + 2],
+                    body[os + ls + 3],
+                ]);
+                let mem = le_uint(&body[os + ls + 4..os + ls + 4 + ls]);
+                (addr, len, mask, mem)
+            } else {
+                (addr, len, 0, len)
+            }
+        } else {
+            let key_len = (usize::from(self.heap_id_length).saturating_sub(1)).min(8);
+            ensure_len(id, 1, key_len)?;
+            let key = le_uint(&id[1..1 + key_len]);
+            self.find_huge_record(file_data, key)?
+        };
+
+        let start = usize::try_from(addr).map_err(|_| heap_error("huge object address"))?;
+        let len = usize::try_from(stored_len).map_err(|_| heap_error("huge object length"))?;
+        ensure_len(file_data, start, len)?;
+        let stored = &file_data[start..start + len];
+        match &self.filter_pipeline {
+            None => Ok(stored.to_vec()),
+            Some(pipeline) => {
+                let mem = usize::try_from(mem_len).map_err(|_| heap_error("huge object size"))?;
+                let out = crate::filters::decompress_chunk_masked(stored, pipeline, mem, 1, mask)?;
+                if out.len() != mem {
+                    return Err(heap_error("filtered huge object decoded to the wrong size"));
+                }
+                Ok(out)
+            }
+        }
+    }
+
+    /// Look up huge object `key` in the huge-object v2 B-tree, returning
+    /// (address, stored length, filter mask, decoded length).
+    fn find_huge_record(
+        &self,
+        file_data: &[u8],
+        key: u64,
+    ) -> Result<(u64, u64, u32, u64), FormatError> {
+        if is_undefined(self.huge_btree_address, self.offset_size) {
+            return Err(heap_error(
+                "huge object ID but the heap has no huge-object index",
+            ));
+        }
+        let hdr = BTreeV2Header::parse(
+            file_data,
+            self.huge_btree_address as usize,
+            self.offset_size,
+            self.length_size,
+        )?;
+        let os = usize::from(self.offset_size);
+        let ls = usize::from(self.length_size);
+        let filtered = self.filter_pipeline.is_some();
+        let (expected_type, rec_len) = if filtered {
+            (BTREE_HUGE_INDIRECT_FILTERED, os + ls + 4 + ls + ls)
+        } else {
+            (BTREE_HUGE_INDIRECT, os + ls + ls)
+        };
+        if hdr.tree_type != expected_type || usize::from(hdr.record_size) < rec_len {
+            return Err(heap_error("unexpected huge-object B-tree record type"));
+        }
+        let records =
+            collect_btree_v2_records(file_data, &hdr, self.offset_size, self.length_size)?;
+        for rec in &records {
+            let d = &rec.data;
+            if d.len() < rec_len {
+                continue;
+            }
+            let addr = le_uint(&d[..os]);
+            let len = le_uint(&d[os..os + ls]);
+            if filtered {
+                let mask = u32::from_le_bytes([
+                    d[os + ls],
+                    d[os + ls + 1],
+                    d[os + ls + 2],
+                    d[os + ls + 3],
+                ]);
+                let mem = le_uint(&d[os + ls + 4..os + 2 * ls + 4]);
+                let id = le_uint(&d[os + 2 * ls + 4..os + 3 * ls + 4]);
+                if id == key {
+                    return Ok((addr, len, mask, mem));
+                }
+            } else {
+                let id = le_uint(&d[os + ls..os + 2 * ls]);
+                if id == key {
+                    return Ok((addr, len, 0, len));
+                }
+            }
+        }
+        Err(heap_error("huge object not found in its B-tree"))
+    }
+
+    /// Read a tiny object (heap ID type 2), stored in the ID itself.
+    fn read_tiny_object(&self, id: &[u8]) -> Result<Vec<u8>, FormatError> {
+        // libhdf5 uses a one-byte length (low 4 bits of byte 0) unless the ID
+        // is long enough to need 12 bits, which then borrow byte 1.
+        let extended = usize::from(self.heap_id_length).saturating_sub(1) > 17;
+        let (len, start) = if extended {
+            ensure_len(id, 0, 2)?;
+            (
+                ((usize::from(id[0] & 0x0F)) << 8 | usize::from(id[1])) + 1,
+                2,
+            )
+        } else {
+            (usize::from(id[0] & 0x0F) + 1, 1)
+        };
+        ensure_len(id, start, len)?;
+        Ok(id[start..start + len].to_vec())
+    }
+
+    /// Read a managed object (heap ID type 0).
+    fn read_heap_managed(
+        &self,
+        file_data: &[u8],
+        id_bytes: &[u8],
+        offset_size: u8,
     ) -> Result<Vec<u8>, FormatError> {
         let (heap_offset, obj_len) = self.decode_managed_id(id_bytes)?;
 
@@ -289,12 +532,15 @@ impl FractalHeapHeader {
             // Root is a direct block
             self.read_from_direct_block(
                 file_data,
-                self.root_block_address as usize,
-                self.starting_block_size,
-                0, // block offset in heap = 0 for root
+                DirectBlock {
+                    addr: self.root_block_address as usize,
+                    size: self.starting_block_size,
+                    heap_offset: 0,
+                    filtered_size: self.root_direct_block_filtered_size,
+                    filter_mask: self.root_direct_block_filter_mask,
+                },
                 heap_offset,
                 obj_len as usize,
-                offset_size,
             )
         } else {
             // Root is an indirect block — limit recursion to 64 levels
@@ -313,27 +559,41 @@ impl FractalHeapHeader {
 
     /// Read an object from a direct block.
     ///
-    /// The heap offset is relative to the start of the block (including its header),
-    /// so we just add it to the block address minus the block's heap offset.
-    #[allow(clippy::too_many_arguments)]
+    /// The heap offset is relative to the start of the block (including its
+    /// header), so we just add it to the block address minus the block's heap
+    /// offset. A filtered heap stores each direct block (header included)
+    /// through its filter pipeline, so the block is decoded first.
     fn read_from_direct_block(
         &self,
         file_data: &[u8],
-        block_addr: usize,
-        _block_size: u64,
-        block_heap_offset: u64,
+        block: DirectBlock,
         target_offset: u64,
         length: usize,
-        _offset_size: u8,
     ) -> Result<Vec<u8>, FormatError> {
-        if target_offset < block_heap_offset {
+        if target_offset < block.heap_offset {
             return Err(FormatError::UnexpectedEof {
-                expected: block_heap_offset as usize,
+                expected: block.heap_offset as usize,
                 available: target_offset as usize,
             });
         }
-        let local_offset = (target_offset - block_heap_offset) as usize;
-        let pos = block_addr
+        let local_offset = (target_offset - block.heap_offset) as usize;
+        if let Some(pipeline) = &self.filter_pipeline {
+            let stored_len = usize::try_from(block.filtered_size)
+                .map_err(|_| heap_error("direct block size"))?;
+            let size = usize::try_from(block.size).map_err(|_| heap_error("direct block size"))?;
+            ensure_len(file_data, block.addr, stored_len)?;
+            let decoded = crate::filters::decompress_chunk_masked(
+                &file_data[block.addr..block.addr + stored_len],
+                pipeline,
+                size,
+                1,
+                block.filter_mask,
+            )?;
+            ensure_len(&decoded, local_offset, length)?;
+            return Ok(decoded[local_offset..local_offset + length].to_vec());
+        }
+        let pos = block
+            .addr
             .checked_add(local_offset)
             .ok_or(FormatError::UnexpectedEof {
                 expected: usize::MAX,
@@ -371,19 +631,13 @@ impl FractalHeapHeader {
         let iblock_header = 5 + offset_size as usize + block_offset_bytes;
         let mut pos = iblock_addr + iblock_header;
 
-        // Compute block sizes for each row using the doubling table
         let tw = self.table_width as u64;
-
         let nrows_usize = nrows as usize;
-
-        // Build table of (block_size, heap_offset) for each child entry
         let mut current_heap_offset = iblock_heap_offset;
 
         // Rows below max_direct_rows hold direct blocks; rows at/above hold
         // child indirect blocks. (NOT the FRHP "starting rows" field.)
         let start_indirect = self.max_direct_rows();
-
-        // Read child addresses for direct block rows
         let max_direct_rows = nrows_usize.min(start_indirect);
 
         for row in 0..max_direct_rows {
@@ -393,48 +647,66 @@ impl FractalHeapHeader {
                 let child_addr = read_offset(file_data, pos, offset_size)?;
                 pos += offset_size as usize;
 
-                if self.io_filter_encoded_length > 0 {
-                    // filtered_size(length_size) + filter_mask(4)
-                    // Skip for now - we don't handle filtered direct blocks in fractal heaps
-                    pos += 4; // filter_mask - simplified
-                }
+                // A filtered heap stores each direct block's filtered size
+                // (length_size) and filter mask (4) after its address.
+                let (filtered_size, filter_mask) = if self.filter_pipeline.is_some() {
+                    let size = read_offset(file_data, pos, self.length_size)?;
+                    pos += usize::from(self.length_size);
+                    ensure_len(file_data, pos, 4)?;
+                    let mask = u32::from_le_bytes([
+                        file_data[pos],
+                        file_data[pos + 1],
+                        file_data[pos + 2],
+                        file_data[pos + 3],
+                    ]);
+                    pos += 4;
+                    (size, mask)
+                } else {
+                    (0, 0)
+                };
 
-                if !is_undefined(child_addr, offset_size) {
-                    let block_end = current_heap_offset + block_size;
-                    if target_offset >= current_heap_offset && target_offset < block_end {
+                let block_end = current_heap_offset.saturating_add(block_size);
+                if !is_undefined(child_addr, offset_size)
+                    && target_offset >= current_heap_offset
+                    && target_offset < block_end
+                {
                     return self.read_from_direct_block(
                         file_data,
-                            child_addr as usize,
-                            block_size,
-                            current_heap_offset,
+                        DirectBlock {
+                            addr: child_addr as usize,
+                            size: block_size,
+                            heap_offset: current_heap_offset,
+                            filtered_size,
+                            filter_mask,
+                        },
                         target_offset,
                         length,
-                            offset_size,
                     );
                 }
-                }
-                current_heap_offset += block_size;
+                current_heap_offset = block_end;
             }
         }
 
-        // If we have indirect block rows
+        // Rows at and above `start_indirect` hold child indirect blocks. A
+        // child in row r spans exactly that row's block size of heap space,
+        // so it has as many rows as a table of that total size needs.
         for row in start_indirect..nrows_usize {
-            let _block_size = self.block_size_for_row(row);
-            let child_nrows = row - start_indirect + 1;
+            let child_space = self.block_size_for_row(row);
+            let child_nrows = self.rows_for_size(child_space);
 
             for _col in 0..tw {
                 let child_addr = read_offset(file_data, pos, offset_size)?;
                 pos += offset_size as usize;
 
-                if !is_undefined(child_addr, offset_size) {
-                    // Calculate total heap space covered by this indirect block child
-                    let total_child_space = self.indirect_block_heap_size(child_nrows);
-                    let block_end = current_heap_offset + total_child_space;
-                    if target_offset >= current_heap_offset && target_offset < block_end {
+                let block_end = current_heap_offset.saturating_add(child_space);
+                if !is_undefined(child_addr, offset_size)
+                    && target_offset >= current_heap_offset
+                    && target_offset < block_end
+                {
                     return self.read_from_indirect_block(
                         file_data,
                         child_addr as usize,
-                            child_nrows as u16,
+                        child_nrows,
                         current_heap_offset,
                         target_offset,
                         length,
@@ -442,11 +714,7 @@ impl FractalHeapHeader {
                         depth_remaining - 1,
                     );
                 }
-                    current_heap_offset += total_child_space;
-                } else {
-                    let total_child_space = self.indirect_block_heap_size(child_nrows);
-                    current_heap_offset += total_child_space;
-                }
+                current_heap_offset = block_end;
             }
         }
 
@@ -475,25 +743,34 @@ impl FractalHeapHeader {
         log2 + 2
     }
 
+    /// Rows an indirect block needs to span `size` bytes of heap space:
+    /// `log2(size) - log2(starting_block_size * table_width) + 1`, as
+    /// libhdf5's `H5HF__dtable_size_to_rows`.
+    fn rows_for_size(&self, size: u64) -> u16 {
+        let log2 = |v: u64| 63u32.saturating_sub(v.max(1).leading_zeros());
+        let first_row_bits = log2(self.starting_block_size) + log2(u64::from(self.table_width));
+        (log2(size).saturating_sub(first_row_bits) + 1) as u16
+    }
+
     /// Get block size for a given row in the doubling table.
     fn block_size_for_row(&self, row: usize) -> u64 {
         let sbs = self.starting_block_size;
         if row <= 1 {
             sbs
         } else {
-            sbs * (1u64 << (row - 1))
+            sbs.saturating_mul(1u64.checked_shl((row - 1) as u32).unwrap_or(u64::MAX))
         }
     }
+}
 
-    /// Total heap space covered by an indirect block with the given number of rows.
-    fn indirect_block_heap_size(&self, nrows: usize) -> u64 {
-        let tw = self.table_width as u64;
-        let mut total = 0u64;
-        for row in 0..nrows {
-            total += self.block_size_for_row(row) * tw;
-        }
-        total
-    }
+/// A managed direct block's location, extent and (for a filtered heap) its
+/// stored size and filter mask.
+struct DirectBlock {
+    addr: usize,
+    size: u64,
+    heap_offset: u64,
+    filtered_size: u64,
+    filter_mask: u32,
 }
 
 #[cfg(test)]
@@ -641,7 +918,7 @@ mod tests {
         let hdr = FractalHeapHeader::parse(&file_data, 0, 8, 8).unwrap();
 
         // Build a managed heap ID:
-        // byte 0: type=0 (bits 6-7 = 00), version=0 (bits 4-5), reserved (bits 0-3)
+        // byte 0: version=0 (bits 6-7), type=0 (bits 4-5), reserved (bits 0-3)
         // bytes 1-6: offset (max_heap_size=16 bits) then length (remaining bits)
         // For offset=0, length=13:
         // payload = offset | (length << 16) = 0 | (13 << 16) = 0x000D0000
@@ -705,9 +982,46 @@ mod tests {
     fn invalid_heap_id_type() {
         let (file_data, _) = build_simple_heap(8, 8);
         let hdr = FractalHeapHeader::parse(&file_data, 0, 8, 8).unwrap();
-        // Type = 1 (tiny) in bits 6-7
-        let id = vec![0x40u8, 0, 0, 0, 0, 0, 0]; // bit 6 set = type 1
+        // Type = 1 (huge) in bits 4-5 is not a managed ID
+        let id = vec![0x10u8, 0, 0, 0, 0, 0, 0];
         let err = hdr.decode_managed_id(&id).unwrap_err();
         assert_eq!(err, FormatError::InvalidHeapIdType(1));
     }
+
+    #[test]
+    fn tiny_object_is_read_from_the_id() {
+        let (file_data, _) = build_simple_heap(8, 8);
+        let hdr = FractalHeapHeader::parse(&file_data, 0, 8, 8).unwrap();
+        // Type 2 (0x20), length - 1 in the low 4 bits, data after.
+        let id = [0x20 | 2, b'a', b'b', b'c', 0, 0, 0];
+        assert_eq!(hdr.read_managed_object(&file_data, &id, 8).unwrap(), b"abc");
+        // A length running past the ID is an error, not a short read.
+        let id = [0x20 | 9, b'a', b'b', b'c', 0, 0, 0];
+        assert!(hdr.read_managed_object(&file_data, &id, 8).is_err());
+    }
+
+    #[test]
+    fn huge_object_with_a_direct_id() {
+        // With IDs long enough for an address and a length, libhdf5 stores
+        // huge objects' location in the ID instead of the huge-object B-tree.
+        let (mut file_data, _) = build_simple_heap(8, 8);
+        let mut hdr = FractalHeapHeader::parse(&file_data, 0, 8, 8).unwrap();
+        hdr.heap_id_length = 17;
+        file_data[900..905].copy_from_slice(b"huge!");
+        let mut id = vec![0x10u8];
+        id.extend_from_slice(&900u64.to_le_bytes());
+        id.extend_from_slice(&5u64.to_le_bytes());
+        assert_eq!(
+            hdr.read_managed_object(&file_data, &id, 8).unwrap(),
+            b"huge!"
+        );
+    }
+
+    #[test]
+    fn unknown_heap_id_version_is_refused() {
+        let (file_data, _) = build_simple_heap(8, 8);
+        let hdr = FractalHeapHeader::parse(&file_data, 0, 8, 8).unwrap();
+        let id = [0x40u8, 0, 0, 0, 0, 0, 0];
+        assert!(hdr.read_managed_object(&file_data, &id, 8).is_err());
+    }
 }

crates/clawhdf5-format/src/gather.rs

@@ -0,0 +1,343 @@
+//! Copying a selection out of a row-major buffer one contiguous run at a time.
+//!
+//! A selection's elements, in output order, fall into runs that are adjacent
+//! in the source: a whole block along the last dimension, blocks that touch
+//! (`stride == block`), and whole rows when the inner dimensions are selected
+//! in full. Copying run by run turns a 256 x 256 hyperslab of a 1024-wide
+//! dataset into 256 `memcpy`s of 1 KiB, where the old extractor recursed and
+//! bounds-checked once per element.
+
+#[cfg(not(feature = "std"))]
+use alloc::{vec, vec::Vec};
+
+use crate::data_read::NativeElement;
+use crate::error::FormatError;
+use crate::selection::Selection;
+
+/// Row-major element strides of `dims` (the last dimension has stride 1).
+fn strides(dims: &[u64]) -> Vec<u64> {
+    let mut s = vec![1u64; dims.len()];
+    for d in (0..dims.len().saturating_sub(1)).rev() {
+        s[d] = s[d + 1].wrapping_mul(dims[d + 1]);
+    }
+    s
+}
+
+/// Merges adjacent runs before handing them on.
+struct Coalesce<F: FnMut(u64, u64)> {
+    start: u64,
+    len: u64,
+    emit: F,
+}
+
+impl<F: FnMut(u64, u64)> Coalesce<F> {
+    #[inline]
+    fn push(&mut self, start: u64, len: u64) {
+        if len == 0 {
+            return;
+        }
+        if self.len > 0 && self.start.wrapping_add(self.len) == start {
+            self.len += len;
+            return;
+        }
+        self.flush();
+        self.start = start;
+        self.len = len;
+    }
+
+    fn flush(&mut self) {
+        if self.len > 0 {
+            (self.emit)(self.start, self.len);
+            self.len = 0;
+        }
+    }
+}
+
+/// Call `emit(first_element, element_count)` for each run of a hyperslab's
+/// elements that is contiguous in a row-major dataset of shape `dims`, in
+/// the order the selection returns them. Adjacent runs are merged.
+///
+/// Coordinates at or past a dimension's extent are skipped, as the
+/// element-wise extractor always did; callers that want them to be an error
+/// validate the selection first. The four vectors must have `dims.len()`
+/// entries.
+pub(crate) fn hyperslab_runs(
+    dims: &[u64],
+    start: &[u64],
+    stride: &[u64],
+    count: &[u64],
+    block: &[u64],
+    emit: impl FnMut(u64, u64),
+) {
+    let rank = dims.len();
+    let mut out = Coalesce {
+        start: 0,
+        len: 0,
+        emit,
+    };
+    if rank == 0 {
+        out.push(0, 1);
+        out.flush();
+        return;
+    }
+    if (0..rank).any(|d| count[d] == 0 || block[d] == 0) {
+        return;
+    }
+    let strides = strides(dims);
+    let last = rank - 1;
+    // Odometer over the outer dimensions: (block index, offset in block).
+    let mut ci = vec![0u64; last];
+    let mut bi = vec![0u64; last];
+    'outer: loop {
+        // Base offset of this row, or skip it if a coordinate is out of range.
+        let mut base = 0u64;
+        let mut in_range = true;
+        for d in 0..last {
+            let coord = start[d]
+                .saturating_add(ci[d].saturating_mul(stride[d]))
+                .saturating_add(bi[d]);
+            if coord >= dims[d] {
+                in_range = false;
+                break;
+            }
+            base = base.wrapping_add(coord.wrapping_mul(strides[d]));
+        }
+        if in_range && (stride[last] == block[last] || count[last] == 1) {
+            // Blocks that touch (the common unit-stride case: block 1,
+            // stride 1) are one range; don't split it into per-element runs.
+            let s = start[last];
+            let e = s
+                .saturating_add(count[last].saturating_mul(block[last]))
+                .min(dims[last]);
+            if s < e {
+                out.push(base.wrapping_add(s), e - s);
+            }
+        } else if in_range {
+            for c in 0..count[last] {
+                let s = start[last].saturating_add(c.saturating_mul(stride[last]));
+                if s >= dims[last] {
+                    continue;
+                }
+                let e = s.saturating_add(block[last]).min(dims[last]);
+                out.push(base.wrapping_add(s), e - s);
+            }
+        }
+        // Advance the odometer, last outer dimension fastest.
+        let mut d = last;
+        loop {
+            if d == 0 {
+                break 'outer;
+            }
+            d -= 1;
+            bi[d] += 1;
+            if bi[d] < block[d] {
+                break;
+            }
+            bi[d] = 0;
+            ci[d] += 1;
+            if ci[d] < count[d] {
+                break;
+            }
+            ci[d] = 0;
+        }
+    }
+    out.flush();
+}
+
+/// The selected elements of `src` — a row-major dataset of shape `dims` and
+/// `elem_size`-byte elements — copied into a fresh `Vec<T>`, one `memcpy` per
+/// contiguous run, with no zero-filling of the output first.
+///
+/// For `T` other than `u8`, `elem_size` must equal `size_of::<T>()`. The
+/// selection must be a validated hyperslab, point list or `None` (`All` is the
+/// caller's to handle); `src` must hold exactly the dataset. Anything that
+/// would read outside `src` is an error, never a partial result.
+pub(crate) fn gather<T: NativeElement>(
+    src: &[u8],
+    dims: &[u64],
+    elem_size: usize,
+    selection: &Selection,
+) -> Result<Vec<T>, FormatError> {
+    let t_size = core::mem::size_of::<T>();
+    if elem_size == 0 || (t_size != 1 && t_size != elem_size) {
+        return Err(FormatError::DataSizeMismatch {
+            expected: t_size,
+            actual: elem_size,
+        });
+    }
+    let n_elements = match selection {
+        Selection::None => 0,
+        Selection::Hyperslab { count, block, .. } => count
+            .iter()
+            .zip(block)
+            .try_fold(1u64, |acc, (&c, &b)| acc.checked_mul(c.checked_mul(b)?))
+            .ok_or_else(|| FormatError::Overflow("hyperslab count x block overflows".into()))?,
+        Selection::Points(points) => points.len() as u64,
+        Selection::All => {
+            return Err(FormatError::SelectionOutOfBounds(
+                "gather does not take Selection::All".into(),
+            ));
+        }
+    };
+    let out_bytes = crate::chunked_read::checked_byte_len(n_elements, elem_size)?;
+    let out_len = out_bytes / t_size;
+    let mut out: Vec<T> = crate::bulk_alloc::vec_for_bulk(out_len);
+    let dst = out.as_mut_ptr().cast::<u8>();
+    let mut written = 0usize;
+    let mut failed = false;
+    let mut copy_run = |first: u64, n: u64| {
+        if failed {
+            return;
+        }
+        let range = usize::try_from(first)
+            .ok()
+            .and_then(|f| f.checked_mul(elem_size))
+            .zip(
+                usize::try_from(n)
+                    .ok()
+                    .and_then(|n| n.checked_mul(elem_size)),
+            )
+            .and_then(|(at, len)| Some((at, len, at.checked_add(len)?)));
+        match range {
+            Some((at, len, end)) if end <= src.len() && written + len <= out_bytes => {
+                // SAFETY: `src[at..end]` is in bounds (checked above), and
+                // `dst + written .. + len` lies within `out`'s capacity of
+                // `out_bytes` bytes (checked above); `out` is a fresh
+                // allocation, so the regions do not overlap.
+                unsafe {
+                    core::ptr::copy_nonoverlapping(src.as_ptr().add(at), dst.add(written), len)
+                };
+                written += len;
+            }
+            _ => failed = true,
+        }
+    };
+    let mut bad_point = false;
+    match selection {
+        Selection::Hyperslab {
+            start,
+            stride,
+            count,
+            block,
+        } => {
+            let rank = dims.len();
+            if [start.len(), stride.len(), count.len(), block.len()] != [rank; 4] {
+                return Err(FormatError::SelectionOutOfBounds(
+                    "hyperslab rank does not match dataset rank".into(),
+                ));
+            }
+            hyperslab_runs(dims, start, stride, count, block, &mut copy_run);
+        }
+        Selection::Points(points) => {
+            let strides = strides(dims);
+            let mut runs = Coalesce {
+                start: 0,
+                len: 0,
+                emit: &mut copy_run,
+            };
+            for p in points {
+                if p.len() != dims.len() || p.iter().zip(dims).any(|(c, n)| c >= n) {
+                    bad_point = true;
+                    break;
+                }
+                let at = p
+                    .iter()
+                    .zip(&strides)
+                    .fold(0u64, |acc, (c, s)| acc.wrapping_add(c.wrapping_mul(*s)));
+                runs.push(at, 1);
+            }
+            runs.flush();
+        }
+        Selection::None | Selection::All => {}
+    }
+    if failed || bad_point || written != out_bytes {
+        return Err(FormatError::SelectionOutOfBounds(
+            "selection addresses elements outside the dataset".into(),
+        ));
+    }
+    // SAFETY: all `out_bytes` bytes, i.e. `out_len` values of `T`, were
+    // written above, and every bit pattern is a valid `T` (`NativeElement`).
+    unsafe { out.set_len(out_len) };
+    Ok(out)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn runs(dims: &[u64], sel: [&[u64]; 4]) -> Vec<(u64, u64)> {
+        let mut v = Vec::new();
+        hyperslab_runs(dims, sel[0], sel[1], sel[2], sel[3], |s, n| v.push((s, n)));
+        v
+    }
+
+    #[test]
+    fn runs_merge_blocks_and_whole_rows() {
+        // A box: one run per row.
+        assert_eq!(
+            runs(&[4, 10], [&[1, 2], &[1, 1], &[2, 3], &[1, 1]]),
+            vec![(12, 3), (22, 3)]
+        );
+        // Whole rows: one run.
+        assert_eq!(
+            runs(&[4, 10], [&[1, 0], &[1, 1], &[3, 10], &[1, 1]]),
+            vec![(10, 30)]
+        );
+        // stride == block: blocks merge.
+        assert_eq!(
+            runs(&[1, 10], [&[0, 1], &[1, 2], &[1, 4], &[1, 2]]),
+            vec![(1, 8)]
+        );
+        // Strided with blocks along both dimensions.
+        assert_eq!(
+            runs(&[6, 10], [&[0, 1], &[3, 4], &[2, 2], &[2, 2]]),
+            vec![
+                (1, 2),
+                (5, 2),
+                (11, 2),
+                (15, 2),
+                (31, 2),
+                (35, 2),
+                (41, 2),
+                (45, 2)
+            ]
+        );
+        // Empty.
+        assert!(runs(&[4, 10], [&[0, 0], &[1, 1], &[0, 3], &[1, 1]]).is_empty());
+        // Scalar.
+        assert_eq!(runs(&[], [&[], &[], &[], &[]]), vec![(0, 1)]);
+    }
+
+    #[test]
+    fn gather_matches_element_order_and_rejects_out_of_range() {
+        let dims = [3u64, 4];
+        let src: Vec<u8> = (0..12u16).flat_map(|v| v.to_le_bytes()).collect();
+        let sel = Selection::Hyperslab {
+            start: vec![0, 1],
+            stride: vec![2, 2],
+            count: vec![2, 2],
+            block: vec![1, 1],
+        };
+        let got: Vec<u8> = gather(&src, &dims, 2, &sel).unwrap();
+        let want: Vec<u8> = [1u16, 3, 9, 11]
+            .iter()
+            .flat_map(|v| v.to_le_bytes())
+            .collect();
+        assert_eq!(got, want);
+        let pts = Selection::Points(vec![vec![2, 3], vec![0, 0], vec![0, 1]]);
+        let got: Vec<u8> = gather(&src, &dims, 2, &pts).unwrap();
+        let want: Vec<u8> = [11u16, 0, 1].iter().flat_map(|v| v.to_le_bytes()).collect();
+        assert_eq!(got, want);
+        // Past the extent, or a source shorter than the dataset: an error.
+        let bad = Selection::Points(vec![vec![3, 0]]);
+        assert!(gather::<u8>(&src, &dims, 2, &bad).is_err());
+        let past = Selection::Hyperslab {
+            start: vec![2, 0],
+            stride: vec![1, 1],
+            count: vec![2, 4],
+            block: vec![1, 1],
+        };
+        assert!(gather::<u8>(&src, &dims, 2, &past).is_err());
+        assert!(gather::<u8>(&src[..20], &dims, 2, &pts).is_err());
+    }
+}

crates/clawhdf5-format/src/global_heap.rs

@@ -1,7 +1,7 @@
 //! HDF5 Global Heap collection parsing.
 
 #[cfg(not(feature = "std"))]
-use alloc::vec::Vec;
+use alloc::{format, string::String, vec::Vec};
 
 use crate::error::FormatError;
 
@@ -52,11 +52,42 @@ fn read_length(data: &[u8], offset: usize, length_size: u8) -> Result<u64, Forma
     })
 }
 
+fn object_overrun_msg(index: u16, size: usize, collection_size: u64) -> String {
+    format!(
+        "global heap object {index} ({size} bytes) runs past the end of its \
+         {collection_size}-byte collection"
+    )
+}
+
 /// Round up to next multiple of 8.
 fn pad8(x: usize) -> usize {
     (x + 7) & !7
 }
 
+/// Where one object of a global heap collection lies in the file, without
+/// its data: see [`GlobalHeapCollection::parse_index`].
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct GlobalHeapObjectRef {
+    /// Object index (1-based; 0 is the free space marker).
+    pub index: u16,
+    /// Reference count.
+    pub reference_count: u16,
+    /// Offset of the object's data in the file data the collection was
+    /// parsed from.
+    pub offset: usize,
+    /// Size of the object's data in bytes.
+    pub size: usize,
+}
+
+/// A global heap collection's objects, located but not copied.
+#[derive(Debug, Clone)]
+pub struct GlobalHeapIndex {
+    /// Total size of this collection including header.
+    pub collection_size: u64,
+    /// The objects, in file order.
+    pub objects: Vec<GlobalHeapObjectRef>,
+}
+
 impl GlobalHeapCollection {
     /// Parse a global heap collection at the given offset in the file data.
     pub fn parse(
@@ -64,8 +95,38 @@ impl GlobalHeapCollection {
         offset: usize,
         length_size: u8,
     ) -> Result<GlobalHeapCollection, FormatError> {
-        // signature(4) + version(1) + reserved(3) + collection_size(length_size)
-        let header_size = 8 + length_size as usize;
+        let index = Self::parse_index(file_data, offset, length_size)?;
+        Ok(GlobalHeapCollection {
+            collection_size: index.collection_size,
+            objects: index
+                .objects
+                .iter()
+                .map(|o| GlobalHeapObject {
+                    index: o.index,
+                    reference_count: o.reference_count,
+                    data: file_data[o.offset..o.offset + o.size].to_vec(),
+                })
+                .collect(),
+        })
+    }
+
+    /// Locate the objects of the global heap collection at `offset` without
+    /// copying their data, so a caller can keep many collections indexed
+    /// for the cost of their object headers.
+    ///
+    /// The collection must lie inside `file_data`, and every object inside
+    /// the collection, as libhdf5 lays them out; an object that runs past
+    /// its collection is an error.
+    pub fn parse_index(
+        file_data: &[u8],
+        offset: usize,
+        length_size: u8,
+    ) -> Result<GlobalHeapIndex, FormatError> {
+        // signature(4) + version(1) + reserved(3) + collection_size(length_size),
+        // padded to a multiple of 8 as libhdf5 lays it out (`H5HG_SIZEOF_HDR`).
+        // With 8-byte lengths the padding is 0; with 4-byte lengths it is 4,
+        // and reading without it put every object 4 bytes early.
+        let header_size = pad8(8 + length_size as usize);
         ensure_len(file_data, offset, header_size)?;
 
         if file_data[offset..offset + 4] != GCOL_SIGNATURE {
@@ -78,25 +139,25 @@ impl GlobalHeapCollection {
         }
 
         let collection_size = read_length(file_data, offset + 8, length_size)?;
-        let collection_size_usize =
-            usize::try_from(collection_size).map_err(|_| FormatError::UnexpectedEof {
-                expected: u64::MAX as usize,
-                available: file_data.len(),
-            })?;
-        let collection_end =
-            offset
-                .checked_add(collection_size_usize)
+        let collection_end = usize::try_from(collection_size)
+            .ok()
+            .and_then(|size| offset.checked_add(size))
             .ok_or(FormatError::UnexpectedEof {
                 expected: usize::MAX,
                 available: file_data.len(),
             })?;
+        if collection_end > file_data.len() {
+            return Err(FormatError::UnexpectedEof {
+                expected: collection_end,
+                available: file_data.len(),
+            });
+        }
 
         let mut pos = offset + header_size;
         let mut objects = Vec::new();
 
         // Parse objects until we hit index 0 (free space) or run out of space
         while pos + 2 <= collection_end {
-            ensure_len(file_data, pos, 2)?;
             let object_index = u16::from_le_bytes([file_data[pos], file_data[pos + 1]]);
 
             if object_index == 0 {
@@ -104,28 +165,39 @@ impl GlobalHeapCollection {
                 break;
             }
 
-            // object_index(2) + reference_count(2) + reserved(4) + object_size(length_size)
-            let obj_header_size = 8 + length_size as usize;
-            ensure_len(file_data, pos, obj_header_size)?;
+            // object_index(2) + reference_count(2) + reserved(4) +
+            // object_size(length_size), padded to 8 (`H5HG_SIZEOF_OBJHDR`).
+            let obj_header_size = pad8(8 + length_size as usize);
+            ensure_len(&file_data[..collection_end], pos, obj_header_size)?;
 
             let reference_count = u16::from_le_bytes([file_data[pos + 2], file_data[pos + 3]]);
-            let object_size = read_length(file_data, pos + 8, length_size)? as usize;
+            let object_size = usize::try_from(read_length(file_data, pos + 8, length_size)?)
+                .map_err(|_| FormatError::Overflow("global heap object size".into()))?;
 
             pos += obj_header_size;
-            ensure_len(file_data, pos, object_size)?;
-            let data = file_data[pos..pos + object_size].to_vec();
+            if pos
+                .checked_add(object_size)
+                .is_none_or(|end| end > collection_end)
+            {
+                return Err(FormatError::VlDataError(object_overrun_msg(
+                    object_index,
+                    object_size,
+                    collection_size,
+                )));
+            }
 
-            objects.push(GlobalHeapObject {
+            objects.push(GlobalHeapObjectRef {
                 index: object_index,
                 reference_count,
-                data,
+                offset: pos,
+                size: object_size,
             });
 
             // Advance past data + padding to 8-byte boundary
-            pos += pad8(object_size);
+            pos = pos.saturating_add(pad8(object_size));
         }
 
-        Ok(GlobalHeapCollection {
+        Ok(GlobalHeapIndex {
             collection_size,
             objects,
         })
@@ -149,10 +221,11 @@ mod tests {
         let ls = length_size as usize;
 
         // Calculate total size
-        let header_size = 8 + ls;
+        // libhdf5 pads both headers to a multiple of 8.
+        let header_size = pad8(8 + ls);
         let mut obj_size_total = 0usize;
         for (_, _, data) in objects {
-            let obj_header = 8 + ls;
+            let obj_header = pad8(8 + ls);
             obj_size_total += obj_header + pad8(data.len());
         }
         // Free space marker (2 bytes for index 0)
@@ -170,6 +243,7 @@ mod tests {
             8 => buf.extend_from_slice(&(collection_size as u64).to_le_bytes()),
             _ => panic!("unsupported length_size"),
         }
+        buf.resize(header_size, 0);
 
         // Objects
         for (index, ref_count, data) in objects {
@@ -181,6 +255,7 @@ mod tests {
                 8 => buf.extend_from_slice(&(data.len() as u64).to_le_bytes()),
                 _ => panic!("unsupported"),
             }
+            buf.resize(buf.len() + (pad8(8 + ls) - (8 + ls)), 0);
             buf.extend_from_slice(data);
             // Pad to 8 bytes
             let padded = pad8(data.len());

crates/clawhdf5-format/src/group_v1.rs

@@ -45,9 +45,16 @@ pub fn resolve_v1_group_entries(
     )?;
 
     let mut entries = Vec::new();
+    let mut heap_checked = false;
     for snod_addr in snod_addrs {
         let snod = SymbolTableNode::parse(file_data, snod_addr as usize, offset_size)?;
         for entry in &snod.entries {
+            // Like libhdf5, look at the heap's free list only once a name is
+            // needed: an empty group with a damaged heap still lists.
+            if !heap_checked {
+                heap.validate_free_list(file_data, length_size)?;
+                heap_checked = true;
+            }
             let name = heap.read_string(file_data, entry.link_name_offset)?;
             entries.push(GroupEntry {
                 name,
@@ -73,6 +80,53 @@ pub fn find_v1_soft_link(
     offset_size: u8,
     length_size: u8,
 ) -> Result<Option<String>, FormatError> {
+    let mut found = None;
+    for_each_v1_soft_link(
+        file_data,
+        sym_table_msg,
+        offset_size,
+        length_size,
+        |link_name| link_name == name,
+        |_, target| {
+            found = Some(target);
+            false
+        },
+    )?;
+    Ok(found)
+}
+
+/// Every soft link in a v1 group, as `(name, target path)`.
+pub fn v1_soft_links(
+    file_data: &[u8],
+    sym_table_msg: &SymbolTableMessage,
+    offset_size: u8,
+    length_size: u8,
+) -> Result<Vec<(String, String)>, FormatError> {
+    let mut links = Vec::new();
+    for_each_v1_soft_link(
+        file_data,
+        sym_table_msg,
+        offset_size,
+        length_size,
+        |_| true,
+        |name, target| {
+            links.push((String::from(name), target));
+            true
+        },
+    )?;
+    Ok(links)
+}
+
+/// Visit the soft links of a v1 group whose name passes `wanted`, with their
+/// target paths, until `visit` returns false.
+fn for_each_v1_soft_link(
+    file_data: &[u8],
+    sym_table_msg: &SymbolTableMessage,
+    offset_size: u8,
+    length_size: u8,
+    wanted: impl Fn(&str) -> bool,
+    mut visit: impl FnMut(&str, String) -> bool,
+) -> Result<(), FormatError> {
     let heap = LocalHeap::parse(
         file_data,
         sym_table_msg.local_heap_address as usize,
@@ -85,13 +139,19 @@ pub fn find_v1_soft_link(
         offset_size,
         length_size,
     )?;
+    let mut heap_checked = false;
     for snod_addr in snod_addrs {
         let snod = SymbolTableNode::parse(file_data, snod_addr as usize, offset_size)?;
         for entry in &snod.entries {
             if entry.cache_type != CACHE_TYPE_SOFT_LINK {
                 continue;
             }
-            if heap.read_string(file_data, entry.link_name_offset)? != name {
+            if !heap_checked {
+                heap.validate_free_list(file_data, length_size)?;
+                heap_checked = true;
+            }
+            let name = heap.read_string(file_data, entry.link_name_offset)?;
+            if !wanted(&name) {
                 continue;
             }
             let value_offset = u32::from_le_bytes([
@@ -100,12 +160,19 @@ pub fn find_v1_soft_link(
                 entry.scratch_pad[2],
                 entry.scratch_pad[3],
             ]);
-            return heap
-                .read_string(file_data, u64::from(value_offset))
-                .map(Some);
+            let target = heap.read_string(file_data, u64::from(value_offset))?;
+            if !visit(&name, target) {
+                return Ok(());
             }
         }
-    Ok(None)
+    }
+    Ok(())
+}
+
+/// Whether a v1 symbol-table entry is a soft link (no object header of its
+/// own; its target path is in the local heap).
+pub fn is_v1_soft_link(entry: &GroupEntry) -> bool {
+    entry.cache_type == CACHE_TYPE_SOFT_LINK
 }
 
 /// Extract the SymbolTableMessage from an object header's messages.

crates/clawhdf5-format/src/group_v2.rs

@@ -38,6 +38,24 @@ pub fn resolve_v2_group_entries(
     }
 }
 
+/// First user-defined link type (HDF5 reserves 2-63; 64 is external).
+const FIRST_USER_DEFINED_LINK_TYPE: u8 = 65;
+
+/// Parse a Link message, or `None` for a user-defined link (type 65-255).
+///
+/// A user-defined link's target is only meaningful to the application that
+/// registered its class, so, like libhdf5 without that class, we cannot
+/// follow it. Leaving it out lets the rest of the group be listed and
+/// resolved instead of one such link failing the whole group; reserved
+/// types (2-63) are still an error.
+fn parse_link(data: &[u8], offset_size: u8) -> Result<Option<LinkMessage>, FormatError> {
+    match LinkMessage::parse(data, offset_size) {
+        Ok(link) => Ok(Some(link)),
+        Err(FormatError::InvalidLinkType(t)) if t >= FIRST_USER_DEFINED_LINK_TYPE => Ok(None),
+        Err(e) => Err(e),
+    }
+}
+
 /// Extract link entries from Link messages directly in the object header (compact storage).
 fn resolve_compact_entries(
     object_header: &ObjectHeader,
@@ -46,7 +64,9 @@ fn resolve_compact_entries(
     let mut entries = Vec::new();
     for msg in &object_header.messages {
         if msg.msg_type == MessageType::Link {
-            let link = LinkMessage::parse(&msg.data, offset_size)?;
+            let Some(link) = parse_link(&msg.data, offset_size)? else {
+                continue;
+            };
             if let LinkTarget::Hard {
                 object_header_address,
             } = link.link_target
@@ -98,7 +118,9 @@ fn for_each_dense_link(
 
         // Read managed object from fractal heap
         let link_data = fh.read_managed_object(file_data, id_bytes, offset_size)?;
-        visit(LinkMessage::parse(&link_data, offset_size)?);
+        if let Some(link) = parse_link(&link_data, offset_size)? {
+            visit(link);
+        }
     }
     Ok(())
 }
@@ -178,7 +200,9 @@ fn find_symbolic_link(
     } else {
         for msg in &object_header.messages {
             if msg.msg_type == MessageType::Link {
-                let link = LinkMessage::parse(&msg.data, offset_size)?;
+                let Some(link) = parse_link(&msg.data, offset_size)? else {
+                    continue;
+                };
                 if link.name == name && is_symbolic(&link.link_target) {
                     found = Some(link.link_target);
                 }
@@ -231,32 +255,135 @@ pub fn resolve_path_any(
     superblock: &Superblock,
     path: &str,
 ) -> Result<u64, FormatError> {
-    resolve_path_following_links(file_data, superblock, path, 0)
+    resolve_path_following_links(
+        file_data,
+        superblock,
+        superblock.root_group_address,
+        path,
+        0,
+    )
+}
+
+/// Resolve `path` relative to the group at `group_address` (an absolute path
+/// starts at the root group instead), following soft links. This is how a
+/// relative soft link's target is resolved: from the group holding the link.
+pub fn resolve_path_from(
+    file_data: &[u8],
+    superblock: &Superblock,
+    group_address: u64,
+    path: &str,
+) -> Result<u64, FormatError> {
+    let start = if path.starts_with('/') {
+        superblock.root_group_address
+    } else {
+        group_address
+    };
+    resolve_path_following_links(file_data, superblock, start, path, 0)
+}
+
+/// The children of the group at `group_address` that can be opened, as h5py
+/// lists them: hard links, and soft links resolved to the object they point
+/// at (under the soft link's own name). Links that cannot be followed are
+/// left out rather than failing the listing — a dangling or cyclic soft link
+/// (h5py lists its name but cannot open it), an external link (another
+/// file), and a user-defined link. An object header that is not a group has
+/// no children.
+///
+/// Any other error, such as a corrupt structure met while resolving a soft
+/// link, is returned.
+pub fn resolve_group_children(
+    file_data: &[u8],
+    superblock: &Superblock,
+    group_address: u64,
+) -> Result<Vec<GroupEntry>, FormatError> {
+    let os = superblock.offset_size;
+    let ls = superblock.length_size;
+    let header = ObjectHeader::parse(file_data, group_address as usize, os, ls)?;
+
+    let mut entries = Vec::new();
+    let mut soft = Vec::new();
+    if is_v1_group(&header) {
+        let sym_msg = header
+            .messages
+            .iter()
+            .find(|m| m.msg_type == MessageType::SymbolTable)
+            .ok_or_else(|| FormatError::PathNotFound(String::from("no symbol table message")))?;
+        let stm = SymbolTableMessage::parse(&sym_msg.data, os)?;
+        let all = group_v1::resolve_v1_group_entries(file_data, &stm, os, ls)?;
+        if all.iter().any(group_v1::is_v1_soft_link) {
+            soft = group_v1::v1_soft_links(file_data, &stm, os, ls)?;
+        }
+        entries.extend(all.into_iter().filter(|e| !group_v1::is_v1_soft_link(e)));
+    } else if is_v2_group(&header) {
+        let mut visit = |link: LinkMessage| match link.link_target {
+            LinkTarget::Hard {
+                object_header_address,
+            } => entries.push(GroupEntry {
+                name: link.name,
+                object_header_address,
+                cache_type: 0,
+            }),
+            LinkTarget::Soft { target_path } => soft.push((link.name, target_path)),
+            LinkTarget::External { .. } => {}
+        };
+        let link_info = find_link_info(&header, os)?;
+        if let Some(fh_addr) = link_info.fractal_heap_address {
+            for_each_dense_link(file_data, &link_info, fh_addr, os, ls, visit)?;
+        } else {
+            for msg in &header.messages {
+                if msg.msg_type == MessageType::Link
+                    && let Some(link) = parse_link(&msg.data, os)?
+                {
+                    visit(link);
+                }
+            }
+        }
+    }
+
+    for (name, target) in soft {
+        match resolve_path_from(file_data, superblock, group_address, &target) {
+            Ok(object_header_address) => entries.push(GroupEntry {
+                name,
+                object_header_address,
+                cache_type: 0,
+            }),
+            // Dangling, cyclic, or ending in another file: not openable here.
+            Err(
+                FormatError::PathNotFound(_)
+                | FormatError::NestingDepthExceeded
+                | FormatError::ExternalLinkUnsupported { .. },
+            ) => {}
+            Err(e) => return Err(e),
+        }
+    }
+    Ok(entries)
 }
 
 /// Soft links followed while resolving one path. Guards against link cycles
 /// (`a -> b -> a`), which are legal to create.
 const MAX_SOFT_LINK_DEPTH: u8 = 16;
 
+/// Walk `path` from the group at `start`, following soft links.
 fn resolve_path_following_links(
     file_data: &[u8],
     superblock: &Superblock,
+    start: u64,
     path: &str,
     depth: u8,
 ) -> Result<u64, FormatError> {
-    let components: Vec<&str> = path.split('/').filter(|s| !s.is_empty()).collect();
+    let components: Vec<&str> = path
+        .split('/')
+        .filter(|s| !s.is_empty() && *s != ".")
+        .collect();
     if components.is_empty() {
-        return Ok(superblock.root_group_address);
+        return Ok(start);
     }
 
     let os = superblock.offset_size;
     let ls = superblock.length_size;
 
-    let root_header =
-        ObjectHeader::parse(file_data, superblock.root_group_address as usize, os, ls)?;
-
-    let mut current_addr = superblock.root_group_address;
-    let mut current_header = root_header;
+    let mut current_addr = start;
+    let mut current_header = ObjectHeader::parse(file_data, start as usize, os, ls)?;
 
     for (i, component) in components.iter().enumerate() {
         let entries = resolve_group_entries(file_data, &current_header, os, ls)?;
@@ -280,20 +407,17 @@ fn resolve_path_following_links(
                         }
                         // A relative target is relative to the group holding
                         // the link; then the rest of the original path.
-                        let mut full = String::new();
-                        if !target_path.starts_with('/') {
-                            for parent in &components[..i] {
-                                full.push('/');
-                                full.push_str(parent);
-                            }
-                        }
-                        full.push('/');
-                        full.push_str(&target_path);
+                        let from = if target_path.starts_with('/') {
+                            superblock.root_group_address
+                        } else {
+                            current_addr
+                        };
+                        let mut full = target_path;
                         for rest in &components[i + 1..] {
                             full.push('/');
                             full.push_str(rest);
                         }
-                        resolve_path_following_links(file_data, superblock, &full, depth + 1)
+                        resolve_path_following_links(file_data, superblock, from, &full, depth + 1)
                     }
                     Some(LinkTarget::External {
                         filename,

crates/clawhdf5-format/src/lib.rs

@@ -26,12 +26,13 @@
 //! use clawhdf5_format::{signature, superblock, object_header, group_v2,
 //!     datatype, dataspace, data_layout, data_read, message_type::MessageType};
 //!
-//! let file_data = std::fs::read("output.h5").unwrap();
-//! let sig = signature::find_signature(&file_data).unwrap();
-//! let sb  = superblock::Superblock::parse(&file_data, sig).unwrap();
-//! let addr = group_v2::resolve_path_any(&file_data, &sb, "data").unwrap();
+//! let bytes = std::fs::read("output.h5").unwrap();
+//! // Addresses are relative to the superblock: skip any user block.
+//! let (_user_block, file_data) = signature::split_user_block(&bytes).unwrap();
+//! let sb  = superblock::Superblock::parse(file_data, 0).unwrap();
+//! let addr = group_v2::resolve_path_any(file_data, &sb, "data").unwrap();
 //! let hdr = object_header::ObjectHeader::parse(
-//!     &file_data, addr as usize, sb.offset_size, sb.length_size).unwrap();
+//!     file_data, addr as usize, sb.offset_size, sb.length_size).unwrap();
 //! ```
 //!
 //! # Features
@@ -42,6 +43,14 @@
 //! | `checksum` | yes | Jenkins lookup3 checksum validation |
 //! | `deflate` | yes | Deflate (gzip) compression via `flate2` |
 //! | `provenance` | yes | SHINES provenance — SHA-256 hashing & verification |
+//! | `lzf` | yes | LZF filter (32000), h5py's `compression="lzf"` |
+//! | `bitshuffle` | no | Bitshuffle filter (32008), none/LZ4/Zstandard |
+//! | `bzip2` | no | bzip2 filter (307) |
+//! | `blosc` | no | Blosc 1 filter (32001) |
+//! | `plugin-filters` | no | The four above |
+//!
+//! Filters are looked up by ID in [`filter_registry`], which also takes
+//! codecs registered at run time for other IDs.
 
 #![cfg_attr(not(feature = "std"), no_std)]
 
@@ -52,8 +61,10 @@ pub mod attribute;
 pub mod attribute_info;
 pub mod btree_v1;
 pub mod btree_v2;
+mod bulk_alloc;
 pub mod checksum;
 pub mod chunk_cache;
+mod chunk_grid;
 pub mod chunk_index;
 pub mod chunked_read;
 pub mod chunked_write;
@@ -69,11 +80,21 @@ pub mod extensible_array;
 pub mod file_writer;
 pub mod fill_value;
 pub mod filter_pipeline;
+pub mod filter_registry;
 pub mod filters;
+#[cfg(any(feature = "bitshuffle", feature = "blosc"))]
+mod filters_bitshuffle;
+#[cfg(feature = "blosc")]
+pub mod filters_blosc;
+#[cfg(feature = "bzip2")]
+mod filters_bzip2;
+#[cfg(feature = "lzf")]
+pub mod filters_lzf;
 mod filters_szip;
 pub mod fixed_array;
 pub mod float16;
 pub mod fractal_heap;
+mod gather;
 pub mod global_heap;
 pub mod group_info;
 pub mod group_v1;
@@ -98,8 +119,20 @@ pub mod shared_message;
 pub mod signature;
 pub mod superblock;
 pub mod symbol_table;
+#[cfg(all(
+    test,
+    any(
+        feature = "lzf",
+        feature = "bitshuffle",
+        feature = "bzip2",
+        feature = "blosc"
+    )
+))]
+mod test_fuzz;
 pub mod type_builders;
+pub mod vds;
 pub mod vl_data;
+mod writer_tree;
 
 #[cfg(feature = "provenance")]
 pub mod provenance;

crates/clawhdf5-format/src/local_heap.rs

@@ -87,6 +87,57 @@ impl LocalHeap {
         })
     }
 
+    /// Walk the free list the way libhdf5 does when it loads a heap's data
+    /// (`H5HL__fl_deserialize`), rejecting a heap whose free list points
+    /// outside the data segment. libhdf5 refuses such a heap ("bad heap free
+    /// list"), and names read from it would be garbage.
+    ///
+    /// libhdf5 only loads a heap when it needs a name from it (an empty
+    /// group's broken heap goes unnoticed), so call this before the first
+    /// [`Self::read_string`], not on parse.
+    ///
+    /// The end of the list is `H5HL_FREE_NULL` (1); an all-ones value (the
+    /// undefined address) is accepted as "no free list" too.
+    pub fn validate_free_list(&self, file_data: &[u8], length_size: u8) -> Result<(), FormatError> {
+        const FREE_NULL: u64 = 1;
+        let ls = length_size as usize;
+        let undefined = if ls >= 8 {
+            u64::MAX
+        } else {
+            (1u64 << (8 * ls)) - 1
+        };
+        let size = self.data_segment_size;
+        let seg = self.data_segment_address;
+        let mut next = self.free_list_head_offset;
+        // Each free block holds two lengths, so a list longer than this
+        // revisits a block: a cycle.
+        let max_blocks = size / (2 * ls as u64) + 1;
+        let mut walked = 0u64;
+        while next != FREE_NULL && next != undefined {
+            if next >= size || walked >= max_blocks {
+                return Err(FormatError::InvalidLocalHeapFreeList);
+            }
+            walked += 1;
+            let at = seg
+                .checked_add(next)
+                .and_then(|a| usize::try_from(a).ok())
+                .ok_or(FormatError::InvalidLocalHeapFreeList)?;
+            let block_offset = next;
+            next = read_offset(file_data, at, length_size)?;
+            if next == 0 {
+                return Err(FormatError::InvalidLocalHeapFreeList);
+            }
+            let block_size = read_offset(file_data, at + ls, length_size)?;
+            if block_offset
+                .checked_add(block_size)
+                .is_none_or(|end| end > size)
+            {
+                return Err(FormatError::InvalidLocalHeapFreeList);
+            }
+        }
+        Ok(())
+    }
+
     /// Read a null-terminated string from the heap's data segment at the given byte offset.
     pub fn read_string(&self, file_data: &[u8], string_offset: u64) -> Result<String, FormatError> {
         let seg_addr = self.data_segment_address as usize;
@@ -162,8 +213,8 @@ mod tests {
         // data_segment_size
         write_val(&mut file, pos, data_seg_size as u64, length_size);
         pos += length_size as usize;
-        // free_list_head_offset
-        write_val(&mut file, pos, 0xFFFFFFFF, length_size);
+        // free_list_head_offset: H5HL_FREE_NULL (no free space)
+        write_val(&mut file, pos, 1, length_size);
         pos += length_size as usize;
         // data_segment_address
         write_val(&mut file, pos, data_seg_offset as u64, offset_size);
@@ -243,6 +294,50 @@ mod tests {
         assert_eq!(s, "test");
     }
 
+    /// Heap with data segment `[a, b, c, 0-padding]` whose free list starts
+    /// at `head` and has one block `(next, size)` at offset 8.
+    fn heap_with_free_block(head: u64, next: u64, size: u64) -> Vec<u8> {
+        let mut file = build_heap_file(0, 100, &["abcdefg"], 8, 8);
+        file.resize(200, 0);
+        write_val(&mut file, 8, 32, 8); // data segment size
+        write_val(&mut file, 16, head, 8);
+        write_val(&mut file, 108, next, 8);
+        write_val(&mut file, 116, size, 8);
+        file
+    }
+
+    #[test]
+    fn free_list_inside_the_segment_is_accepted() {
+        let file = heap_with_free_block(8, 1, 24);
+        let heap = LocalHeap::parse(&file, 0, 8, 8).unwrap();
+        heap.validate_free_list(&file, 8).unwrap();
+        assert_eq!(heap.read_string(&file, 0).unwrap(), "abcdefg");
+        // An all-ones head is "no free list" too.
+        let file = heap_with_free_block(u64::MAX, 0, 0);
+        let heap = LocalHeap::parse(&file, 0, 8, 8).unwrap();
+        assert!(heap.validate_free_list(&file, 8).is_ok());
+    }
+
+    #[test]
+    fn bad_free_list_is_rejected_like_libhdf5() {
+        for (head, next, size, why) in [
+            (40, 1, 8, "head past the segment"),
+            (8, 1, 25, "block runs past the segment"),
+            (8, 0, 8, "next offset of zero"),
+            (8, 8, 8, "cycle"),
+            (8, 999, 8, "next past the segment"),
+        ] {
+            let file = heap_with_free_block(head, next, size);
+            // The header itself parses; the free list is checked on use.
+            let heap = LocalHeap::parse(&file, 0, 8, 8).unwrap();
+            assert_eq!(
+                heap.validate_free_list(&file, 8).unwrap_err(),
+                FormatError::InvalidLocalHeapFreeList,
+                "{why}"
+            );
+        }
+    }
+
     #[test]
     fn invalid_version() {
         let mut file = build_heap_file(0, 100, &["x"], 8, 8);

crates/clawhdf5-format/src/object_header.rs

@@ -108,10 +108,20 @@ impl ObjectHeader {
             return Err(FormatError::InvalidObjectHeaderVersion(version));
         }
 
-        let num_messages = LittleEndian::read_u16(&data[offset + 2..offset + 4]);
+        let num_messages = LittleEndian::read_u16(&data[offset + 2..offset + 4]) as usize;
         let reference_count = LittleEndian::read_u32(&data[offset + 4..offset + 8]);
         let header_data_size = LittleEndian::read_u32(&data[offset + 8..offset + 12]) as usize;
 
+        // libhdf5 (H5O__prefix_deserialize): a header with messages needs room
+        // for at least one message header, and one without has an empty chunk.
+        if (num_messages > 0 && header_data_size < V1_MSG_HEADER_SIZE)
+            || (num_messages == 0 && header_data_size > 0)
+        {
+            return Err(FormatError::InvalidObjectHeader(
+                "bad object header chunk size",
+            ));
+        }
+
         // Pad to 8-byte alignment: header prefix is 12 bytes, pad to 16
         let padding = 4; // pad 12-byte prefix to 16-byte alignment
         let msg_start = offset
@@ -124,69 +134,23 @@ impl ObjectHeader {
         ensure_len(data, msg_start, header_data_size)?;
 
         let mut messages = Vec::new();
-        let mut pos = msg_start;
-        let msg_end =
-            msg_start
-                .checked_add(header_data_size)
-                .ok_or(FormatError::UnexpectedEof {
-                    expected: usize::MAX,
-                    available: data.len(),
-                })?;
-
-        for _ in 0..num_messages {
-            if pos + 8 > msg_end {
-                break;
-            }
-            let msg_type_raw = LittleEndian::read_u16(&data[pos..pos + 2]);
-            let msg_data_size = LittleEndian::read_u16(&data[pos + 2..pos + 4]) as usize;
-            let msg_flags = data[pos + 4];
-            // reserved(3) at pos+5..pos+8
-            pos += 8;
-
-            ensure_len(data, pos, msg_data_size)?;
-            let msg_type = MessageType::from_u16(msg_type_raw);
-
-            // Check if unknown + must-understand (bit 3 of msg_flags)
-            if let MessageType::Unknown(id) = msg_type
-                && msg_flags & 0x08 != 0
-            {
-                return Err(FormatError::UnsupportedMessage(id));
-            }
-
-            if msg_type != MessageType::Nil {
-                messages.push(HeaderMessage {
-                    msg_type,
-                    size: msg_data_size,
-                    flags: msg_flags,
-                    creation_order: None,
-                    data: data[pos..pos + msg_data_size].to_vec(),
-                });
-            }
-
-            pos += msg_data_size;
-
-            // Follow continuations
-            if msg_type == MessageType::ObjectHeaderContinuation {
-                let cont_msg_data = &messages
-                    .last()
-                    .ok_or(FormatError::InvalidObjectHeaderSignature)?
-                    .data;
-                if cont_msg_data.len() >= (offset_size as usize + length_size as usize) {
-                    let cont_offset = read_offset(cont_msg_data, 0, offset_size)? as usize;
-                    let cont_length =
-                        read_offset(cont_msg_data, offset_size as usize, length_size)? as usize;
-                    // Parse continuation block (v1: just raw messages, no signature)
-                    let cont_msgs = Self::parse_v1_continuation(
+        let chunk0_count = Self::parse_v1_chunk(
             data,
-                        cont_offset,
-                        cont_length,
+            msg_start,
+            header_data_size,
             offset_size,
             length_size,
-                        32, // max continuation depth
+            MAX_V1_CONTINUATION_DEPTH,
+            &mut messages,
         )?;
-                    messages.extend(cont_msgs);
-                }
-            }
+        // libhdf5 reads every message in the first chunk and refuses a header
+        // whose prefix claims fewer than that (continuation chunks are read
+        // later and not held to the count). Stopping after the claimed number
+        // silently dropped the rest.
+        if chunk0_count > num_messages {
+            return Err(FormatError::InvalidObjectHeader(
+                "bad object header message count",
+            ));
         }
 
         Ok(ObjectHeader {
@@ -201,76 +165,87 @@ impl ObjectHeader {
         })
     }
 
-    fn parse_v1_continuation(
+    /// Parse the messages of one version-1 chunk (`length` bytes at
+    /// `offset`, no signature), following continuation messages as they are
+    /// met. Returns how many messages (NIL ones included) this chunk itself
+    /// holds.
+    ///
+    /// A version-1 chunk is filled with messages whose sizes are multiples of
+    /// 8; libhdf5 refuses a message that is not aligned, that runs past the
+    /// end of the chunk, or leftover bytes too few for a message header (a
+    /// "gap", which only version 2 allows).
+    #[allow(clippy::too_many_arguments)]
+    fn parse_v1_chunk(
         data: &[u8],
         offset: usize,
         length: usize,
         offset_size: u8,
         length_size: u8,
         depth_remaining: u16,
-    ) -> Result<Vec<HeaderMessage>, FormatError> {
+        messages: &mut Vec<HeaderMessage>,
+    ) -> Result<usize, FormatError> {
         if depth_remaining == 0 {
             return Err(FormatError::NestingDepthExceeded);
         }
         ensure_len(data, offset, length)?;
-        let mut messages = Vec::new();
+        let end = offset + length;
         let mut pos = offset;
-        let end = offset.saturating_add(length);
+        let mut count = 0usize;
 
-        while pos + 8 <= end {
+        while pos < end {
+            if end - pos < V1_MSG_HEADER_SIZE {
+                return Err(FormatError::InvalidObjectHeader(
+                    "gap found in early version of file format",
+                ));
+            }
             let msg_type_raw = LittleEndian::read_u16(&data[pos..pos + 2]);
             let msg_data_size = LittleEndian::read_u16(&data[pos + 2..pos + 4]) as usize;
             let msg_flags = data[pos + 4];
-            pos += 8;
+            // reserved(3) at pos+5..pos+8
+            pos += V1_MSG_HEADER_SIZE;
 
-            if pos + msg_data_size > end {
-                break;
+            if !msg_data_size.is_multiple_of(8) {
+                return Err(FormatError::InvalidObjectHeader("message not aligned"));
             }
+            if msg_data_size > end - pos {
+                return Err(FormatError::InvalidObjectHeader(
+                    "message size exceeds buffer end",
+                ));
+            }
+            let body = &data[pos..pos + msg_data_size];
+            check_message(1, msg_type_raw, msg_flags, body, offset_size, length_size)?;
+            count += 1;
 
             let msg_type = MessageType::from_u16(msg_type_raw);
-
-            if let MessageType::Unknown(id) = msg_type
-                && msg_flags & 0x08 != 0
-            {
-                return Err(FormatError::UnsupportedMessage(id));
-            }
-
             if msg_type != MessageType::Nil {
                 messages.push(HeaderMessage {
                     msg_type,
                     size: msg_data_size,
                     flags: msg_flags,
                     creation_order: None,
-                    data: data[pos..pos + msg_data_size].to_vec(),
+                    data: body.to_vec(),
                 });
             }
-
             pos += msg_data_size;
 
-            // Recursive continuations
+            // Follow continuations (v1 continuation chunks are just raw
+            // messages, no signature); check_message has checked the body.
             if msg_type == MessageType::ObjectHeaderContinuation {
-                let cont_msg_data = &messages
-                    .last()
-                    .ok_or(FormatError::InvalidObjectHeaderSignature)?
-                    .data;
-                if cont_msg_data.len() >= (offset_size as usize + length_size as usize) {
-                    let cont_offset = read_offset(cont_msg_data, 0, offset_size)? as usize;
-                    let cont_length =
-                        read_offset(cont_msg_data, offset_size as usize, length_size)? as usize;
-                    let cont_msgs = Self::parse_v1_continuation(
+                let cont_offset = read_offset(body, 0, offset_size)? as usize;
+                let cont_length = read_offset(body, offset_size as usize, length_size)? as usize;
+                Self::parse_v1_chunk(
                     data,
                     cont_offset,
                     cont_length,
                     offset_size,
                     length_size,
                     depth_remaining - 1,
+                    messages,
                 )?;
-                    messages.extend(cont_msgs);
-                }
             }
         }
 
-        Ok(messages)
+        Ok(count)
     }
 
     fn parse_v2(
@@ -287,6 +262,11 @@ impl ObjectHeader {
             return Err(FormatError::InvalidObjectHeaderVersion(version));
         }
         let flags = data[offset + 5];
+        if flags & !V2_HDR_ALL_FLAGS != 0 {
+            return Err(FormatError::InvalidObjectHeader(
+                "unknown object header status flag(s)",
+            ));
+        }
 
         let mut pos = offset + 6;
 
@@ -306,7 +286,14 @@ impl ObjectHeader {
         // Optional attribute storage thresholds (flags bit 4)
         if flags & 0x10 != 0 {
             ensure_len(data, pos, 4)?;
-            // max_compact_attrs(2) + min_dense_attrs(2) — read but don't store for now
+            // max_compact_attrs(2) + min_dense_attrs(2) — checked, not stored
+            let max_compact = LittleEndian::read_u16(&data[pos..pos + 2]);
+            let min_dense = LittleEndian::read_u16(&data[pos + 2..pos + 4]);
+            if max_compact < min_dense {
+                return Err(FormatError::InvalidObjectHeader(
+                    "bad object header attribute phase change values",
+                ));
+            }
             pos += 4;
         }
 
@@ -321,6 +308,14 @@ impl ObjectHeader {
         ensure_len(data, pos, chunk_size_width as usize)?;
         let chunk0_size = read_offset(data, pos, chunk_size_width)? as usize;
         pos += chunk_size_width as usize;
+        // Bit 2: attribute creation order tracked → messages include creation order field
+        let has_creation_order = flags & 0x04 != 0;
+        let msg_header_size = if has_creation_order { 6 } else { 4 };
+        if chunk0_size > 0 && chunk0_size < msg_header_size {
+            return Err(FormatError::InvalidObjectHeader(
+                "bad object header chunk size",
+            ));
+        }
 
         let chunk0_msg_start = pos;
         let chunk0_msg_end = pos
@@ -344,9 +339,6 @@ impl ObjectHeader {
             }
         }
 
-        // Bit 2: attribute creation order tracked → messages include creation order field
-        let has_creation_order = flags & 0x04 != 0;
-
         // Parse messages from chunk0
         let mut messages = Vec::new();
         let mut continuations = Vec::new();
@@ -405,8 +397,20 @@ impl ObjectHeader {
     ) -> Result<(), FormatError> {
         let msg_header_size = if has_creation_order { 6 } else { 4 };
         let mut pos = start;
+        let mut null_count = 0usize;
 
-        while pos + msg_header_size <= end {
+        while pos < end {
+            // Leftover bytes too few for a message header are a gap, which
+            // libhdf5 allows only in a chunk without NIL messages (a writer
+            // that leaves a gap had no NIL message to put the space in).
+            if end - pos < msg_header_size {
+                if null_count != 0 {
+                    return Err(FormatError::InvalidObjectHeader(
+                        "gap in chunk with no null messages",
+                    ));
+                }
+                break;
+            }
             let msg_type_raw = data[pos] as u16;
             let msg_data_size = LittleEndian::read_u16(&data[pos + 1..pos + 3]) as usize;
             let msg_flags = data[pos + 3];
@@ -417,36 +421,36 @@ impl ObjectHeader {
             };
             pos += msg_header_size;
 
-            if pos + msg_data_size > end {
-                // Could be padding at end of chunk
-                break;
+            // `end` is where the messages stop and the checksum starts.
+            // libhdf5 bounds a message by the chunk including its checksum,
+            // but a message that runs into the checksum still fails there:
+            // its loop stops at the checksum, and reading the checksum from
+            // past its start overruns the chunk ("ran off end of input
+            // buffer while decoding"). Both refuse it; only the text
+            // differs.
+            if msg_data_size > end - pos {
+                return Err(FormatError::InvalidObjectHeader(
+                    "message size exceeds buffer end",
+                ));
             }
+            let body = &data[pos..pos + msg_data_size];
+            check_message(2, msg_type_raw, msg_flags, body, offset_size, length_size)?;
 
             let msg_type = MessageType::from_u16(msg_type_raw);
-
-            if let MessageType::Unknown(id) = msg_type
-                && msg_flags & 0x08 != 0
-            {
-                return Err(FormatError::UnsupportedMessage(id));
-            }
-
-            let msg_data = data[pos..pos + msg_data_size].to_vec();
-
             if msg_type == MessageType::ObjectHeaderContinuation {
-                // Parse continuation offset/length from message data
-                if msg_data.len() >= (offset_size as usize + length_size as usize) {
-                    let cont_off = read_offset(&msg_data, 0, offset_size)? as usize;
-                    let cont_len =
-                        read_offset(&msg_data, offset_size as usize, length_size)? as usize;
+                // check_message has checked the body holds both fields.
+                let cont_off = read_offset(body, 0, offset_size)? as usize;
+                let cont_len = read_offset(body, offset_size as usize, length_size)? as usize;
                 continuations.push((cont_off, cont_len));
-                }
-            } else if msg_type != MessageType::Nil {
+            } else if msg_type == MessageType::Nil {
+                null_count += 1;
+            } else {
                 messages.push(HeaderMessage {
                     msg_type,
                     size: msg_data_size,
                     flags: msg_flags,
                     creation_order,
-                    data: msg_data,
+                    data: body.to_vec(),
                 });
             }
 
@@ -509,6 +513,151 @@ impl ObjectHeader {
     }
 }
 
+/// Size of a version-1 message header: type(2) + size(2) + flags(1) + reserved(3).
+const V1_MSG_HEADER_SIZE: usize = 8;
+
+/// How deep version-1 continuation chunks may chain (malformed-data guard).
+const MAX_V1_CONTINUATION_DEPTH: u16 = 32;
+
+/// Every defined version-2 object header status flag (libhdf5
+/// `H5O_HDR_ALL_FLAGS`): chunk-0 size width (bits 0-1), attribute creation
+/// order tracked/indexed, attribute phase-change values, times stored.
+const V2_HDR_ALL_FLAGS: u8 = 0x3F;
+
+// Header message flag bits (libhdf5 `H5O_MSG_FLAG_*`). Bit 0 (constant) needs
+// no check. Bit 3 (fail if unknown and the file is opened for writing) never
+// fails a read: the parser only ever reads, as libhdf5 ignores it for a
+// read-only open.
+const MSG_FLAG_SHARED: u8 = 0x02;
+const MSG_FLAG_DONTSHARE: u8 = 0x04;
+const MSG_FLAG_FAIL_IF_UNKNOWN_AND_OPEN_FOR_WRITE: u8 = 0x08;
+const MSG_FLAG_MARK_IF_UNKNOWN: u8 = 0x10;
+const MSG_FLAG_WAS_UNKNOWN: u8 = 0x20;
+const MSG_FLAG_SHAREABLE: u8 = 0x40;
+/// Fail if the message is unknown, whatever the access mode.
+const MSG_FLAG_FAIL_IF_UNKNOWN_ALWAYS: u8 = 0x80;
+
+/// Message type ids libhdf5 has a class for (`H5O_msg_class_g`): 0x00-0x18
+/// except 0x09 (a test-only "bogus" message). Anything else is an unknown
+/// message.
+fn is_known_message(id: u16) -> bool {
+    id <= 0x18 && id != 0x09
+}
+
+/// Message classes that may be shared (`H5O_SHARE_IS_SHARABLE`): dataspace,
+/// datatype, the two fill-value messages, filter pipeline and attribute.
+fn is_shareable_message(id: u16) -> bool {
+    matches!(id, 0x01 | 0x03 | 0x04 | 0x05 | 0x0B | 0x0C)
+}
+
+/// Check one header message the way libhdf5 does while it loads an object
+/// header (`H5O__chunk_deserialize`), so an object libhdf5 refuses to open is
+/// refused here too instead of being read from a corrupt header:
+///
+/// - contradictory flag combinations;
+/// - an unknown message the file says no reader may skip (bit 7). This had
+///   bits 3 and 7 the wrong way round once, failing objects libhdf5 reads
+///   and reading ones it refuses (`tbogus.h5`);
+/// - a known message whose class cannot be shared, flagged shared or
+///   shareable (`cve-2016-4332`);
+/// - the messages libhdf5 decodes while loading the header, whose decode
+///   errors fail the load: continuation, reference count (which a version-1
+///   header cannot hold), and both modification-time messages.
+fn check_message(
+    header_version: u8,
+    id: u16,
+    flags: u8,
+    body: &[u8],
+    offset_size: u8,
+    length_size: u8,
+) -> Result<(), FormatError> {
+    let bad_flags = FormatError::InvalidObjectHeader("bad flag combination for message");
+    if flags & MSG_FLAG_SHARED != 0 && flags & MSG_FLAG_DONTSHARE != 0 {
+        return Err(bad_flags);
+    }
+    if flags & MSG_FLAG_WAS_UNKNOWN != 0
+        && (flags & MSG_FLAG_FAIL_IF_UNKNOWN_AND_OPEN_FOR_WRITE != 0
+            || flags & MSG_FLAG_MARK_IF_UNKNOWN == 0)
+    {
+        return Err(bad_flags);
+    }
+
+    if !is_known_message(id) {
+        if flags & MSG_FLAG_FAIL_IF_UNKNOWN_ALWAYS != 0 {
+            return Err(FormatError::UnsupportedMessage(id));
+        }
+        return Ok(());
+    }
+    if flags & (MSG_FLAG_SHARED | MSG_FLAG_SHAREABLE) != 0 && !is_shareable_message(id) {
+        return Err(FormatError::InvalidObjectHeader(
+            "message of unshareable class flagged as shareable",
+        ));
+    }
+
+    let overrun = FormatError::InvalidObjectHeader("ran off end of input buffer while decoding");
+    match id {
+        // Continuation: address + length, and the chunk cannot be empty.
+        0x10 => {
+            if body.len() < offset_size as usize + length_size as usize {
+                return Err(overrun);
+            }
+            if read_offset(body, offset_size as usize, length_size)? == 0 {
+                return Err(FormatError::InvalidObjectHeader(
+                    "invalid continuation chunk size (0)",
+                ));
+            }
+        }
+        // Reference count: version-2 headers only; version 0 then a u32.
+        0x16 => {
+            if header_version == 1 {
+                return Err(FormatError::InvalidObjectHeader(
+                    "object header version does not support reference count message",
+                ));
+            }
+            match body.first() {
+                None => return Err(overrun),
+                Some(0) => {}
+                Some(_) => {
+                    return Err(FormatError::InvalidObjectHeader(
+                        "bad version number for reference count message",
+                    ));
+                }
+            }
+            if body.len() < 5 {
+                return Err(overrun);
+            }
+        }
+        // Old modification time: "YYYYMMDDhhmmss" and 2 reserved bytes.
+        0x0E => {
+            if body.len() < 16 {
+                return Err(overrun);
+            }
+            if !body[..14].iter().all(u8::is_ascii_digit) {
+                return Err(FormatError::InvalidObjectHeader(
+                    "badly formatted modification time message",
+                ));
+            }
+        }
+        // New modification time: version 1, 3 reserved bytes, u32 seconds.
+        0x12 => {
+            match body.first() {
+                None => return Err(overrun),
+                Some(1) => {}
+                Some(_) => {
+                    return Err(FormatError::InvalidObjectHeader(
+                        "bad version number for mtime message",
+                    ));
+                }
+            }
+            if body.len() < 8 {
+                return Err(overrun);
+            }
+        }
+        _ => {}
+    }
+    Ok(())
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -523,11 +672,14 @@ mod tests {
         // Calculate total header message data size
         let mut msg_bytes = Vec::new();
         for (mtype, mdata, mflags) in messages {
+            // v1 message sizes are multiples of 8 (the data is zero-padded).
+            let padded = mdata.len().div_ceil(8) * 8;
             msg_bytes.extend_from_slice(&mtype.to_le_bytes()); // type(2)
-            msg_bytes.extend_from_slice(&(mdata.len() as u16).to_le_bytes()); // size(2)
+            msg_bytes.extend_from_slice(&(padded as u16).to_le_bytes()); // size(2)
             msg_bytes.push(*mflags); // flags(1)
             msg_bytes.extend_from_slice(&[0u8; 3]); // reserved(3)
             msg_bytes.extend_from_slice(mdata); // data
+            msg_bytes.resize(msg_bytes.len() + padded - mdata.len(), 0);
         }
 
         let mut buf = Vec::new();
@@ -617,9 +769,10 @@ mod tests {
         let hdr = ObjectHeader::parse(&data, 0, 8, 8).unwrap();
         assert_eq!(hdr.messages.len(), 2);
         assert_eq!(hdr.messages[0].msg_type, MessageType::Dataspace);
-        assert_eq!(hdr.messages[0].data, vec![1, 2, 3, 4]);
+        // v1 message data is padded to a multiple of 8 bytes.
+        assert_eq!(hdr.messages[0].data, vec![1, 2, 3, 4, 0, 0, 0, 0]);
         assert_eq!(hdr.messages[1].msg_type, MessageType::DataLayout);
-        assert_eq!(hdr.messages[1].data, vec![5, 6]);
+        assert_eq!(hdr.messages[1].data[..2], [5, 6]);
     }
 
     #[test]
@@ -632,14 +785,263 @@ mod tests {
     }
 
     #[test]
-    fn parse_v1_unknown_must_understand_errors() {
-        // Bit 3 of msg_flags = must understand
-        let messages = [(0x00FFu16, &[0xAA][..], 0x08u8)];
+    fn parse_v1_unknown_fail_always_errors() {
+        // Bit 7 of msg_flags = fail if unknown, whatever the access mode.
+        let messages = [(0x00FFu16, &[0xAA][..], 0x80u8)];
         let data = build_v1_header(&messages, 8, 8);
         let err = ObjectHeader::parse(&data, 0, 8, 8).unwrap_err();
         assert_eq!(err, FormatError::UnsupportedMessage(0x00FF));
     }
 
+    #[test]
+    fn parse_v1_unknown_fail_on_write_is_ignored_when_reading() {
+        // Bit 3 = fail if unknown *and the file is opened for writing*. This
+        // parser only reads, so libhdf5 (read-only) opens such an object and
+        // so must we. Bits 4/5 (mark if unknown / was unknown) never fail.
+        for flags in [0x08u8, 0x10, 0x30] {
+            let messages = [(0x00FFu16, &[0xAA][..], flags)];
+            let data = build_v1_header(&messages, 8, 8);
+            let hdr = ObjectHeader::parse(&data, 0, 8, 8).unwrap();
+            assert_eq!(hdr.messages[0].msg_type, MessageType::Unknown(0x00FF));
+        }
+    }
+
+    #[test]
+    fn contradictory_message_flags_are_refused() {
+        // libhdf5: "bad flag combination for message" for shared + don't
+        // share, was-unknown without mark-if-unknown, and was-unknown with
+        // fail-if-unknown-on-write.
+        for flags in [0x06u8, 0x20, 0x38] {
+            let data = build_v1_header(&[(0x00FFu16, &[0xAA][..], flags)], 8, 8);
+            assert_eq!(
+                ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+                FormatError::InvalidObjectHeader("bad flag combination for message"),
+                "flags {flags:#x}"
+            );
+            let data = build_v2_header(0x00, &[(0xF0, &[1, 2], flags)], None);
+            assert_eq!(
+                ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+                FormatError::InvalidObjectHeader("bad flag combination for message"),
+                "flags {flags:#x}"
+            );
+        }
+    }
+
+    #[test]
+    fn unshareable_message_flagged_shareable_is_refused() {
+        // A layout (0x08) or modification time (0x12) message cannot be
+        // shared; bit 1 (shared) or bit 6 (shareable) on one is corruption
+        // (cve-2016-4332). A datatype (0x03) may be shareable.
+        let mtime = [1u8, 0, 0, 0, 0x10, 0x20, 0x30, 0x40];
+        for (id, flags) in [(0x08u16, 0x40u8), (0x08, 0x02), (0x12, 0x40)] {
+            let data = build_v1_header(&[(id, &mtime[..], flags)], 8, 8);
+            assert_eq!(
+                ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+                FormatError::InvalidObjectHeader(
+                    "message of unshareable class flagged as shareable"
+                ),
+                "id {id:#x} flags {flags:#x}"
+            );
+        }
+        let data = build_v1_header(&[(0x03, &[0u8; 8][..], 0x40)], 8, 8);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+        // An unknown message is never checked for shareability.
+        let data = build_v1_header(&[(0x00FF, &[0u8; 8][..], 0x40)], 8, 8);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+    }
+
+    #[test]
+    fn v1_message_must_be_aligned() {
+        // cve-2018-13873: a v1 message whose size is not a multiple of 8.
+        let mut data = build_v1_header(&[(0x01, &[0u8; 8][..], 0)], 8, 8);
+        data[16 + 2] = 7; // size field of the only message
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("message not aligned")
+        );
+    }
+
+    #[test]
+    fn message_overrunning_its_chunk_is_refused() {
+        // It used to end the chunk quietly, dropping this message and any
+        // after it.
+        let mut data = build_v1_header(&[(0x01, &[0u8; 8][..], 0)], 8, 8);
+        data[16 + 2] = 16;
+        data.resize(data.len() + 64, 0);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("message size exceeds buffer end")
+        );
+        let mut data = build_v2_header(0x00, &[(0x01, &[1, 2], 0)], None);
+        data[7 + 1] = 9; // size of the only message (after OHDR, ver, flags, chunk size)
+        let chk = crate::checksum::jenkins_lookup3(&data[..data.len() - 4]);
+        let n = data.len();
+        data[n - 4..].copy_from_slice(&chk.to_le_bytes());
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("message size exceeds buffer end")
+        );
+    }
+
+    #[test]
+    fn v1_gap_after_last_message_is_refused() {
+        // Fewer than 8 bytes left over: a gap, which only version 2 allows.
+        let mut data = build_v1_header(&[(0x01, &[0u8; 8][..], 0)], 8, 8);
+        data[8] += 4; // header_data_size
+        data.extend_from_slice(&[0u8; 4]);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("gap found in early version of file format")
+        );
+    }
+
+    #[test]
+    fn v1_chunk_holding_more_messages_than_the_prefix_says_is_refused() {
+        // cve-2024-32619: the prefix says 1 message, the chunk holds 2. The
+        // second used to be dropped silently.
+        let mut data = build_v1_header(&[(0x01, &[0u8; 8][..], 0), (0x03, &[0u8; 8][..], 0)], 8, 8);
+        data[2] = 1;
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("bad object header message count")
+        );
+        // Fewer in the chunk than the prefix says is fine (the rest may be in
+        // continuation chunks; libhdf5 only enforces that with strict checks).
+        data[2] = 3;
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap().messages.len(),
+            2
+        );
+    }
+
+    #[test]
+    fn v1_prefix_chunk_size_must_fit_the_message_count() {
+        let mut data = build_v1_header(&[], 8, 8);
+        data[8] = 8; // no messages but a non-empty chunk
+        data.extend_from_slice(&[0u8; 8]);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("bad object header chunk size")
+        );
+    }
+
+    #[test]
+    fn v1_header_cannot_hold_a_reference_count_message() {
+        // cve-2018-11204.
+        let data = build_v1_header(&[(0x16, &[0, 2, 0, 0, 0][..], 0)], 8, 8);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader(
+                "object header version does not support reference count message"
+            )
+        );
+        let data = build_v2_header(0x00, &[(0x16, &[0, 2, 0, 0, 0], 0)], None);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+    }
+
+    #[test]
+    fn modification_time_messages_are_decoded_with_the_header() {
+        // cve-2024-33873 (version 0) and cve-2024-33874 (empty message).
+        for (body, why) in [
+            (
+                &[0u8, 0, 0, 0, 1, 2, 3, 4][..],
+                "bad version number for mtime message",
+            ),
+            (&[][..], "ran off end of input buffer while decoding"),
+        ] {
+            let data = build_v2_header(0x00, &[(0x12, body, 0)], None);
+            assert_eq!(
+                ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+                FormatError::InvalidObjectHeader(why)
+            );
+        }
+        let data = build_v2_header(0x00, &[(0x12, &[1, 0, 0, 0, 1, 2, 3, 4], 0)], None);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+        // The old (0x0E) message is 14 ASCII digits and 2 reserved bytes.
+        let data = build_v1_header(&[(0x0E, &b"20110414214255\0\0"[..], 0)], 8, 8);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+        let data = build_v1_header(&[(0x0E, &b"2011041421425x\0\0"[..], 0)], 8, 8);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("badly formatted modification time message")
+        );
+    }
+
+    #[test]
+    fn continuation_message_must_hold_a_nonempty_chunk() {
+        let mut cont = [0u8; 16];
+        cont[..8].copy_from_slice(&64u64.to_le_bytes());
+        let data = build_v2_header(0x00, &[(0x10, &cont, 0)], None);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("invalid continuation chunk size (0)")
+        );
+        let data = build_v2_header(0x00, &[(0x10, &cont[..8], 0)], None);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("ran off end of input buffer while decoding")
+        );
+    }
+
+    #[test]
+    fn v2_prefix_is_checked() {
+        let data = build_v2_header(0x40, &[(0x01, &[1], 0)], None);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("unknown object header status flag(s)")
+        );
+        // build_v2_header writes max_compact 8, min_dense 6; swap them.
+        let mut data = build_v2_header(0x10, &[(0x01, &[1], 0)], None);
+        data[6] = 6;
+        data[8] = 8;
+        let chk = crate::checksum::jenkins_lookup3(&data[..data.len() - 4]);
+        let n = data.len();
+        data[n - 4..].copy_from_slice(&chk.to_le_bytes());
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("bad object header attribute phase change values")
+        );
+    }
+
+    #[test]
+    fn v2_gap_is_allowed_only_without_nil_messages() {
+        // Three bytes after the last message: a gap (a message header is 4).
+        let mut data = build_v2_header(0x00, &[(0x01, &[1, 2, 3], 0), (0x03, &[], 0)], None);
+        // Turn the empty datatype message (4 header bytes) into a 3-byte gap
+        // by shrinking the chunk.
+        let n = data.len();
+        data.truncate(n - 5);
+        data[6] -= 1;
+        let chk = crate::checksum::jenkins_lookup3(&data);
+        data.extend_from_slice(&chk.to_le_bytes());
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap().messages.len(),
+            1
+        );
+
+        let mut data = build_v2_header(0x00, &[(0x00, &[1, 2, 3], 0), (0x03, &[], 0)], None);
+        let n = data.len();
+        data.truncate(n - 5);
+        data[6] -= 1;
+        let chk = crate::checksum::jenkins_lookup3(&data);
+        data.extend_from_slice(&chk.to_le_bytes());
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::InvalidObjectHeader("gap in chunk with no null messages")
+        );
+    }
+
+    #[test]
+    fn parse_v2_unknown_message_flags() {
+        let data = build_v2_header(0x00, &[(0xF0, &[1, 2], 0x08)], None);
+        assert!(ObjectHeader::parse(&data, 0, 8, 8).is_ok());
+        let data = build_v2_header(0x00, &[(0xF0, &[1, 2], 0x80)], None);
+        assert_eq!(
+            ObjectHeader::parse(&data, 0, 8, 8).unwrap_err(),
+            FormatError::UnsupportedMessage(0xF0)
+        );
+    }
+
     #[test]
     fn parse_v2_no_timestamps_one_message() {
         let data = build_v2_header(0x00, &[(0x01, &[10, 20], 0)], None);

crates/clawhdf5-format/src/object_header_writer.rs

@@ -1,11 +1,17 @@
 //! Object header writer for v2 format.
 
 #[cfg(not(feature = "std"))]
-use alloc::vec::Vec;
+use alloc::{format, vec::Vec};
 
 use crate::checksum::jenkins_lookup3;
+use crate::error::FormatError;
 use crate::message_type::MessageType;
 
+/// Largest message payload a v2 object header can describe: the per-message
+/// size field is 2 bytes. A bigger message cannot be encoded at all — writing
+/// its size truncated to 16 bits produced files libhdf5 refuses.
+pub const MAX_MESSAGE_SIZE: usize = u16::MAX as usize;
+
 /// Writer for v2 object headers with proper checksums.
 pub struct ObjectHeaderWriter {
     messages: Vec<(MessageType, Vec<u8>, u8)>, // (type, data, msg_flags)
@@ -30,7 +36,22 @@ impl ObjectHeaderWriter {
     }
 
     /// Serialize the complete v2 object header (OHDR + messages + checksum).
-    pub fn serialize(&self) -> Vec<u8> {
+    ///
+    /// Fails with [`FormatError::SerializationError`] when a message is larger
+    /// than [`MAX_MESSAGE_SIZE`] (e.g. an attribute over ~64 KiB, which would
+    /// need dense attribute storage), rather than writing a corrupt header.
+    pub fn serialize(&self) -> Result<Vec<u8>, FormatError> {
+        if let Some((msg_type, data, _)) = self
+            .messages
+            .iter()
+            .find(|(_, data, _)| data.len() > MAX_MESSAGE_SIZE)
+        {
+            return Err(FormatError::SerializationError(format!(
+                "{msg_type:?} message is {} bytes; an object header message holds at most \
+                 {MAX_MESSAGE_SIZE} bytes",
+                data.len()
+            )));
+        }
         // Calculate total message bytes: each message has type(1) + size(2) + flags(1) + data
         let msg_bytes_total: usize = self
             .messages
@@ -80,7 +101,7 @@ impl ObjectHeaderWriter {
         let checksum = jenkins_lookup3(&buf);
         buf.extend_from_slice(&checksum.to_le_bytes());
 
-        buf
+        Ok(buf)
     }
 }
 
@@ -125,15 +146,22 @@ impl BatchObjectHeaderWriter {
 
     /// Compute the serialized size of each header without actually serializing.
     /// Returns sizes in the same order as headers were added.
-    pub fn compute_sizes(&self) -> Vec<usize> {
-        self.headers.iter().map(|h| h.serialize().len()).collect()
+    pub fn compute_sizes(&self) -> Result<Vec<usize>, FormatError> {
+        self.headers
+            .iter()
+            .map(|h| h.serialize().map(|b| b.len()))
+            .collect()
     }
 
     /// Serialize all headers into a single contiguous buffer.
     /// Returns `(combined_bytes, offsets)` where `offsets[i]` is the byte
     /// offset of header `i` within the combined buffer.
-    pub fn serialize_all(&self) -> (Vec<u8>, Vec<usize>) {
-        let serialized: Vec<Vec<u8>> = self.headers.iter().map(|h| h.serialize()).collect();
+    pub fn serialize_all(&self) -> Result<(Vec<u8>, Vec<usize>), FormatError> {
+        let serialized: Vec<Vec<u8>> = self
+            .headers
+            .iter()
+            .map(|h| h.serialize())
+            .collect::<Result<_, _>>()?;
         let total: usize = serialized.iter().map(|s| s.len()).sum();
         let mut buf = Vec::with_capacity(total);
         let mut offsets = Vec::with_capacity(serialized.len());
@@ -141,7 +169,7 @@ impl BatchObjectHeaderWriter {
             offsets.push(buf.len());
             buf.extend_from_slice(s);
         }
-        (buf, offsets)
+        Ok((buf, offsets))
     }
 }
 
@@ -159,7 +187,7 @@ mod tests {
     #[test]
     fn empty_header_roundtrip() {
         let writer = ObjectHeaderWriter::new();
-        let bytes = writer.serialize();
+        let bytes = writer.serialize().unwrap();
         let hdr = ObjectHeader::parse(&bytes, 0, 8, 8).unwrap();
         assert_eq!(hdr.version, 2);
         assert_eq!(hdr.messages.len(), 0);
@@ -170,7 +198,7 @@ mod tests {
         let mut writer = ObjectHeaderWriter::new();
         writer.add_message(MessageType::Dataspace, vec![1, 2, 3, 4]);
         writer.add_message(MessageType::Datatype, vec![5, 6]);
-        let bytes = writer.serialize();
+        let bytes = writer.serialize().unwrap();
         let hdr = ObjectHeader::parse(&bytes, 0, 8, 8).unwrap();
         assert_eq!(hdr.messages.len(), 2);
         assert_eq!(hdr.messages[0].msg_type, MessageType::Dataspace);
@@ -184,12 +212,30 @@ mod tests {
         let mut writer = ObjectHeaderWriter::new();
         // Add a message with >255 bytes of payload
         writer.add_message(MessageType::Datatype, vec![0xAA; 300]);
-        let bytes = writer.serialize();
+        let bytes = writer.serialize().unwrap();
         let hdr = ObjectHeader::parse(&bytes, 0, 8, 8).unwrap();
         assert_eq!(hdr.messages.len(), 1);
         assert_eq!(hdr.messages[0].data.len(), 300);
     }
 
+    #[test]
+    fn oversized_message_is_an_error_not_a_truncated_size() {
+        // 65535 bytes is the largest encodable payload.
+        let mut writer = ObjectHeaderWriter::new();
+        writer.add_message(MessageType::Attribute, vec![0; MAX_MESSAGE_SIZE]);
+        let bytes = writer.serialize().unwrap();
+        let hdr = ObjectHeader::parse(&bytes, 0, 8, 8).unwrap();
+        assert_eq!(hdr.messages[0].data.len(), MAX_MESSAGE_SIZE);
+
+        // One byte more used to be written with its size wrapped to 0.
+        let mut writer = ObjectHeaderWriter::new();
+        writer.add_message(MessageType::Attribute, vec![0; MAX_MESSAGE_SIZE + 1]);
+        assert!(matches!(
+            writer.serialize(),
+            Err(FormatError::SerializationError(_))
+        ));
+    }
+
     #[test]
     fn batch_writer_serialize_all() {
         let mut batch = BatchObjectHeaderWriter::new();
@@ -204,7 +250,7 @@ mod tests {
         batch.add(w2);
         assert_eq!(batch.len(), 2);
 
-        let (buf, offsets) = batch.serialize_all();
+        let (buf, offsets) = batch.serialize_all().unwrap();
         assert_eq!(offsets.len(), 2);
         assert_eq!(offsets[0], 0);
 
@@ -222,7 +268,7 @@ mod tests {
     fn batch_writer_empty() {
         let batch = BatchObjectHeaderWriter::new();
         assert!(batch.is_empty());
-        let (buf, offsets) = batch.serialize_all();
+        let (buf, offsets) = batch.serialize_all().unwrap();
         assert!(buf.is_empty());
         assert!(offsets.is_empty());
     }

crates/clawhdf5-format/src/parallel_read.rs

@@ -10,7 +10,7 @@
 use crate::chunked_read::ChunkInfo;
 use crate::error::FormatError;
 use crate::filter_pipeline::FilterPipeline;
-use crate::filters::decompress_chunk;
+use crate::filters::decompress_chunk_exact;
 use crate::lane_partition::{self, LaneStats, PartitionStats};
 
 /// Threshold: only use parallel decompression when chunk count exceeds this.
@@ -27,6 +27,20 @@ pub fn should_use_parallel(chunk_count: usize) -> bool {
     chunk_count > PARALLEL_THRESHOLD
 }
 
+/// Whether handing a read's chunks to rayon can decode them faster than the
+/// calling thread would alone.
+///
+/// `false` when the pool the work would go to (the current pool inside a
+/// rayon worker, else the global one) has a single thread. Handing work to
+/// that pool is then worse than useless: the caller blocks while the one
+/// worker decodes, and every other thread reading at the same time queues
+/// behind the same worker, so N reader threads decode on one core. (That is
+/// how full reads with `--decode-threads 1` stopped scaling at about 2x in
+/// the `concurrent_read` benchmark.)
+pub fn pool_can_parallelise() -> bool {
+    rayon::current_num_threads() > 1
+}
+
 /// Decompress chunks in parallel using lane-partitioned assignment.
 ///
 /// Instead of naive `par_iter`, chunks are deterministically assigned to lanes
@@ -84,11 +98,14 @@ pub fn decompress_chunks_lane_partitioned(
                 }
                 let raw_chunk = &file_data[c_addr..c_addr + size];
 
-                let decompressed = if chunk_info.filter_mask == 0 {
-                    decompress_chunk(raw_chunk, pipeline, chunk_total_bytes, element_size)?
-                } else {
-                    raw_chunk.to_vec()
-                };
+                let decompressed = decompress_chunk_exact(
+                    raw_chunk,
+                    pipeline,
+                    chunk_total_bytes,
+                    element_size,
+                    chunk_info.filter_mask,
+                    &chunk_info.offsets,
+                )?;
 
                 stats.chunks_processed += 1;
                 stats.compressed_bytes += size as u64;
@@ -158,11 +175,14 @@ pub fn decompress_chunks_parallel(
             }
             let raw_chunk = &file_data[c_addr..c_addr + size];
 
-            let decompressed = if chunk_info.filter_mask == 0 {
-                decompress_chunk(raw_chunk, pipeline, chunk_total_bytes, element_size)?
-            } else {
-                raw_chunk.to_vec()
-            };
+            let decompressed = decompress_chunk_exact(
+                raw_chunk,
+                pipeline,
+                chunk_total_bytes,
+                element_size,
+                chunk_info.filter_mask,
+                &chunk_info.offsets,
+            )?;
 
             Ok(DecompressedChunk {
                 index,
@@ -200,11 +220,14 @@ pub fn decompress_chunks_sequential(
         let raw_chunk = &file_data[c_addr..c_addr + size];
 
         let decompressed = if let Some(pl) = pipeline {
-            if chunk_info.filter_mask == 0 {
-                decompress_chunk(raw_chunk, pl, chunk_total_bytes, element_size)?
-            } else {
-                raw_chunk.to_vec()
-            }
+            decompress_chunk_exact(
+                raw_chunk,
+                pl,
+                chunk_total_bytes,
+                element_size,
+                chunk_info.filter_mask,
+                &chunk_info.offsets,
+            )?
         } else {
             raw_chunk.to_vec()
         };
@@ -212,3 +235,60 @@ pub fn decompress_chunks_sequential(
     }
     Ok(result)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::filter_pipeline::{FILTER_SHUFFLE, FilterDescription};
+
+    /// Eight shuffled 32-byte chunks; chunk 5 is stored short when `short`.
+    fn chunks(short: bool) -> (Vec<u8>, Vec<ChunkInfo>) {
+        let mut file = Vec::new();
+        let mut infos = Vec::new();
+        for i in 0..8u64 {
+            let len = if short && i == 5 { 16 } else { 32 };
+            infos.push(ChunkInfo {
+                chunk_size: len as u32,
+                filter_mask: 0,
+                offsets: vec![i * 8],
+                address: file.len() as u64,
+            });
+            file.extend(core::iter::repeat_n(i as u8, len));
+        }
+        (file, infos)
+    }
+
+    /// Every parallel decoder refuses a chunk that decodes short, naming it.
+    #[test]
+    fn short_decoded_chunk_is_an_error() {
+        let pipeline = FilterPipeline {
+            version: 2,
+            filters: vec![FilterDescription {
+                filter_id: FILTER_SHUFFLE,
+                name: None,
+                flags: 0,
+                client_data: vec![4],
+            }],
+        };
+        let (file, good) = chunks(false);
+        assert_eq!(
+            decompress_chunks_parallel(&file, &good, &pipeline, 32, 4).unwrap()[5],
+            [5u8; 32]
+        );
+        let (file, bad) = chunks(true);
+        let errs = [
+            decompress_chunks_lane_partitioned(&file, &bad, &pipeline, 32, 4, 1, Some(3))
+                .map(|_| ())
+                .unwrap_err(),
+            decompress_chunks_parallel(&file, &bad, &pipeline, 32, 4)
+                .map(|_| ())
+                .unwrap_err(),
+            decompress_chunks_sequential(&file, &bad, Some(&pipeline), 32, 4)
+                .map(|_| ())
+                .unwrap_err(),
+        ];
+        for e in errs {
+            assert!(e.to_string().contains("[40]"), "{e}");
+        }
+    }
+}

crates/clawhdf5-format/src/partial_read.rs

@@ -3,11 +3,13 @@
 //!
 //! [`crate::data_read::read_raw_data_selection`] used to decode the *entire*
 //! dataset and then pick elements out of it, so reading a 64x64 window of a
-//! large dataset took about as long as reading all of it. Here the selection's
-//! bounding box is materialised instead — only the rows of a contiguous
-//! dataset, or only the chunks, that overlap it — and the existing extractor
-//! runs over that small buffer with the selection translated to the box's
-//! origin. Extraction semantics are therefore exactly the full-read ones.
+//! large dataset took about as long as reading all of it. A contiguous
+//! dataset's selection is now copied straight out of the file, one `memcpy`
+//! per contiguous run of selected elements (`crate::gather`). For chunked
+//! data the selection's bounding box is materialised — only the chunks that
+//! overlap it — and the extractor runs over that small buffer with the
+//! selection translated to the box's origin. Extraction semantics are
+//! therefore exactly the full-read ones.
 
 #[cfg(not(feature = "std"))]
 use alloc::string as alloc_or_std;
@@ -22,7 +24,7 @@ use crate::data_read::extract_selection_from_buffer;
 use crate::dataspace::Dataspace;
 use crate::error::FormatError;
 use crate::filter_pipeline::FilterPipeline;
-use crate::filters::decompress_chunk;
+use crate::filters::{all_filters_skipped, decompress_chunk_exact};
 use crate::selection::Selection;
 
 /// The smallest axis-aligned box containing every selected element, as
@@ -250,10 +252,33 @@ pub fn read_selection(
     if dims.is_empty() || elem_size == 0 {
         return Ok(None);
     }
+    let total = dataspace.checked_num_elements()?;
+    // Contiguous data is addressable in place: copy the selection's runs
+    // straight out of it, whatever fraction of the dataset it covers, with no
+    // intermediate box (and no full copy for a large selection).
+    if let (
+        DataLayout::Contiguous {
+            address: Some(address),
+            ..
+        },
+        Selection::Hyperslab { .. } | Selection::Points(_),
+    ) = (layout, selection)
+    {
+        validate(selection, dims)?;
+        let base = usize::try_from(*address)
+            .map_err(|_| FormatError::Overflow("data address exceeds usize".into()))?;
+        let data = file_data
+            .get(base..)
+            .and_then(|d| d.get(..checked_byte_len(total, elem_size).ok()?))
+            .ok_or(FormatError::UnexpectedEof {
+                expected: base,
+                available: file_data.len(),
+            })?;
+        return crate::gather::gather::<u8>(data, dims, elem_size, selection).map(Some);
+    }
     let Some((box_start, box_extent)) = bounding_box(selection, dims) else {
         return Ok(None);
     };
-    let total = dataspace.checked_num_elements()?;
     let box_elements = box_extent
         .iter()
         .try_fold(1u64, |acc, &e| acc.checked_mul(e))
@@ -265,30 +290,6 @@ pub fn read_selection(
     let mut boxed = alloc_output(checked_byte_len(box_elements, elem_size)?)?;
 
     match layout {
-        DataLayout::Contiguous {
-            address: Some(address),
-            ..
-        } => {
-            let base = usize::try_from(*address)
-                .map_err(|_| FormatError::Overflow("data address exceeds usize".into()))?;
-            let data = file_data
-                .get(base..)
-                .and_then(|d| d.get(..checked_byte_len(total, elem_size).ok()?))
-                .ok_or(FormatError::UnexpectedEof {
-                    expected: base,
-                    available: file_data.len(),
-                })?;
-            let origin = vec![0u64; dims.len()];
-            copy_overlap(
-                data,
-                &origin,
-                dims,
-                &mut boxed,
-                &box_start,
-                &box_extent,
-                elem_size,
-            );
-        }
         DataLayout::Chunked {
             btree_address: Some(_),
             ..
@@ -325,12 +326,19 @@ pub fn read_selection(
                         expected: at.saturating_add(chunk.chunk_size as usize),
                         available: file_data.len(),
                     })?;
-                // Mirrors the full-read path: a non-zero filter mask means the
-                // chunk was stored unfiltered.
+                // Mirrors the full-read path: filter-mask bit i set means
+                // filter i was not applied to this chunk.
                 let decoded;
                 let data: &[u8] = match pipeline {
-                    Some(pl) if chunk.filter_mask == 0 => {
-                        decoded = decompress_chunk(raw, pl, chunk_bytes, elem_size as u32)?;
+                    Some(pl) if !all_filters_skipped(pl, chunk.filter_mask) => {
+                        decoded = decompress_chunk_exact(
+                            raw,
+                            pl,
+                            chunk_bytes,
+                            elem_size as u32,
+                            chunk.filter_mask,
+                            &chunk.offsets[..rank],
+                        )?;
                         &decoded
                     }
                     _ => raw,

crates/clawhdf5-format/src/property_list.rs

@@ -43,7 +43,7 @@ impl Default for DatasetCreateProps {
             fletcher32: false,
             lz4: false,
             zstd_level: None,
-            fill_time: FillTime::Alloc,
+            fill_time: FillTime::IfSet,
             compact: false,
             alignment: 0,
         }
@@ -335,7 +335,7 @@ mod tests {
     fn dcpl_defaults() {
         let dcpl = DatasetCreateProps::new();
         assert!(dcpl.chunk_dims.is_none());
-        assert_eq!(dcpl.fill_time, FillTime::Alloc);
+        assert_eq!(dcpl.fill_time, FillTime::IfSet);
         assert!(!dcpl.compact);
     }
 

crates/clawhdf5-format/src/selection.rs

@@ -229,44 +229,47 @@ impl Selection {
     /// self-describing in length, so the count lets a caller walk a packed list
     /// of selections — as the Virtual Dataset global-heap block does).
     ///
-    /// Only the forms needed for VDS assembly are decoded: `ALL`, `NONE`, and
-    /// **regular** hyperslabs serialized at **version 3** (the encoding HDF5
-    /// 1.10+/2.0 emit). Point selections, irregular hyperslabs, and older
-    /// hyperslab versions return an error rather than mis-decoding.
+    /// Decodes `ALL`, `NONE`, and hyperslabs at every version libhdf5 writes
+    /// (1: irregular, 4-byte coordinates — the default-format encoding; 2:
+    /// regular, 8-byte; 3: either, variable width). A regular hyperslab maps
+    /// to [`Selection::Hyperslab`]; an *irregular* one (a union of blocks)
+    /// maps to a single-block hyperslab when it has one block, and otherwise to
+    /// [`Selection::Points`] listing the union in row-major order (the order
+    /// libhdf5 iterates it in). Unlimited counts/blocks decode as `u64::MAX`
+    /// (see [`SerializedSelection::decode`] for the raw form). Point
+    /// selections are refused: libhdf5 does not allow them in virtual datasets
+    /// either.
     pub fn decode_serialized(data: &[u8]) -> Result<(Selection, usize), FormatError> {
-        if data.len() < 8 {
-            return Err(FormatError::UnexpectedEof {
-                expected: 8,
-                available: data.len(),
-            });
+        let (raw, len) = SerializedSelection::decode(data)?;
+        let sel = match raw {
+            SerializedSelection::All => Selection::All,
+            SerializedSelection::None => Selection::None,
+            SerializedSelection::Regular {
+                start,
+                stride,
+                count,
+                block,
+            } => Selection::Hyperslab {
+                start,
+                stride,
+                count,
+                block,
+            },
+            SerializedSelection::Blocks { rank, starts, ends } => {
+                if starts.len() == rank {
+                    let block = starts.iter().zip(&ends).map(|(&s, &e)| e - s + 1).collect();
+                    Selection::Hyperslab {
+                        start: starts,
+                        stride: vec![1; rank],
+                        count: vec![1; rank],
+                        block,
                     }
-        let sel_type = u32::from_le_bytes([data[0], data[1], data[2], data[3]]);
-        let version = u32::from_le_bytes([data[4], data[5], data[6], data[7]]);
-
-        match sel_type {
-            // ALL / NONE: type(4) + version(4) + reserved(4) + length(4) = 16 bytes.
-            3 | 0 => {
-                if data.len() < 16 {
-                    return Err(FormatError::UnexpectedEof {
-                        expected: 16,
-                        available: data.len(),
-                    });
-                }
-                let sel = if sel_type == 3 {
-                    Selection::All
                 } else {
-                    Selection::None
+                    Selection::Points(blocks_union_coords(rank, &starts, &ends)?)
+                }
+            }
         };
-                Ok((sel, 16))
-            }
-            2 => decode_hyperslab_serialized(data, version),
-            1 => Err(FormatError::ChunkedReadError(
-                "VDS point selections are not supported".into(),
-            )),
-            _ => Err(FormatError::ChunkedReadError(
-                "unknown dataspace selection type".into(),
-            )),
-        }
+        Ok((sel, len))
     }
 
     /// Enumerate the selected element indices of a **1-D** dataspace of the
@@ -314,6 +317,11 @@ impl Selection {
                         "VDS selection rank does not match dataspace rank".into(),
                     ));
                 }
+                if count.iter().chain(block.iter()).any(|&v| v == UNLIMITED) {
+                    return Err(FormatError::ChunkedReadError(
+                        "unlimited selection must be clipped before it is enumerated".into(),
+                    ));
+                }
                 // Selected coordinates along each dimension, in order.
                 let mut per_dim: Vec<Vec<u64>> = Vec::with_capacity(rank);
                 for d in 0..rank {
@@ -400,59 +408,174 @@ impl Selection {
     }
 }
 
-/// Decode an `H5S_SEL_HYPER` selection in its serialized form. Only version-3
-/// **regular** hyperslabs are supported.
-fn decode_hyperslab_serialized(
-    data: &[u8],
-    version: u32,
-) -> Result<(Selection, usize), FormatError> {
-    if version != 3 {
-        return Err(FormatError::ChunkedReadError(
-            "only version-3 hyperslab selections are supported".into(),
-        ));
+/// Hyperslab count/block value meaning "unlimited" (`H5S_UNLIMITED`).
+pub const UNLIMITED: u64 = u64::MAX;
+
+/// Largest number of elements an irregular selection is expanded to when it
+/// is converted to a point list by [`Selection::decode_serialized`].
+const MAX_EXPANDED_POINTS: u64 = 1 << 26;
+
+/// A selection exactly as `H5S_select_serialize` stores it, before it is
+/// applied to any dataspace.
+///
+/// Unlike [`Selection`] this keeps an irregular hyperslab as its list of
+/// blocks, and a regular hyperslab's count/block may be [`UNLIMITED`] (the
+/// unlimited selections used by unlimited and "printf" virtual dataset
+/// mappings).
+#[derive(Debug, Clone, PartialEq)]
+pub enum SerializedSelection {
+    /// `H5S_SEL_ALL`.
+    All,
+    /// `H5S_SEL_NONE`.
+    None,
+    /// A regular hyperslab. `count[d]` or `block[d]` may be [`UNLIMITED`].
+    Regular {
+        start: Vec<u64>,
+        stride: Vec<u64>,
+        count: Vec<u64>,
+        block: Vec<u64>,
+    },
+    /// An irregular hyperslab: the union of `starts.len() / rank` blocks, each
+    /// given by its first (`starts`) and last (`ends`, inclusive) coordinate,
+    /// flattened block-major.
+    Blocks {
+        rank: usize,
+        starts: Vec<u64>,
+        ends: Vec<u64>,
+    },
+}
+
+fn sel_err(msg: &str) -> FormatError {
+    FormatError::ChunkedReadError(msg.into())
+}
+
+/// Bounds-checked little-endian reader over a serialized selection.
+struct SelReader<'a> {
+    data: &'a [u8],
+    pos: usize,
+}
+
+impl SelReader<'_> {
+    fn take(&mut self, n: usize) -> Result<&[u8], FormatError> {
+        let end = self.pos.checked_add(n).filter(|&e| e <= self.data.len());
+        let end = end.ok_or(FormatError::UnexpectedEof {
+            expected: self.pos.saturating_add(n),
+            available: self.data.len(),
+        })?;
+        let s = &self.data[self.pos..end];
+        self.pos = end;
+        Ok(s)
     }
-    // type(4) ver(4) flags(1) enc_size(1) rank(4) [start,stride,count,block]*rank
-    if data.len() < 14 {
-        return Err(FormatError::UnexpectedEof {
-            expected: 14,
-            available: data.len(),
-        });
+
+    fn uint(&mut self, size: usize) -> Result<u64, FormatError> {
+        let bytes = self.take(size)?;
+        Ok(bytes
+            .iter()
+            .enumerate()
+            .fold(0u64, |v, (i, &b)| v | (b as u64) << (i * 8)))
     }
-    let flags = data[8];
-    let enc_size = data[9] as usize;
-    // Bit 0 set => regular hyperslab. Irregular hyperslabs list explicit blocks.
-    if flags & 0x01 == 0 {
-        return Err(FormatError::ChunkedReadError(
-            "irregular VDS hyperslab selections are not supported".into(),
-        ));
+
+    fn remaining(&self) -> usize {
+        self.data.len() - self.pos
     }
-    if enc_size != 2 && enc_size != 4 && enc_size != 8 {
-        return Err(FormatError::ChunkedReadError(
-            "unsupported hyperslab coordinate encoding size".into(),
-        ));
-    }
-    let rank = u32::from_le_bytes([data[10], data[11], data[12], data[13]]) as usize;
-    // HDF5 caps dataspace rank at 32 (H5S_MAX_RANK). Reject anything larger so a
-    // corrupt rank can't drive a huge allocation or read loop.
-    if rank > 32 {
-        return Err(FormatError::ChunkedReadError(
-            "hyperslab selection rank exceeds maximum (32)".into(),
-        ));
-    }
-    let mut pos = 14;
-    let read_coord = |data: &[u8], pos: usize| -> Result<u64, FormatError> {
-        if pos + enc_size > data.len() {
-            return Err(FormatError::UnexpectedEof {
-                expected: pos + enc_size,
-                available: data.len(),
-            });
-        }
-        let mut v = 0u64;
-        for (i, &b) in data[pos..pos + enc_size].iter().enumerate() {
-            v |= (b as u64) << (i * 8);
-        }
-        Ok(v)
+}
+
+impl SerializedSelection {
+    /// Decode a serialized selection, returning it and the number of bytes it
+    /// occupies. Mirrors libhdf5's `H5S_select_deserialize`: `ALL`/`NONE` and
+    /// hyperslab versions 1-3 are decoded; point selections (which libhdf5
+    /// refuses in virtual datasets) and malformed input are errors.
+    pub fn decode(data: &[u8]) -> Result<(SerializedSelection, usize), FormatError> {
+        let mut r = SelReader { data, pos: 0 };
+        let sel_type = r.uint(4)?;
+        let version = r.uint(4)?;
+        match sel_type {
+            // ALL / NONE: type(4) + version(4) + reserved(4) + length(4).
+            0 | 3 => {
+                r.take(8)?;
+                let sel = if sel_type == 3 {
+                    SerializedSelection::All
+                } else {
+                    SerializedSelection::None
                 };
+                Ok((sel, r.pos))
+            }
+            2 => {
+                let sel = decode_hyperslab(&mut r, version)?;
+                Ok((sel, r.pos))
+            }
+            1 => Err(sel_err(
+                "VDS point selections are not supported (libhdf5 rejects them too)",
+            )),
+            _ => Err(sel_err("unknown dataspace selection type")),
+        }
+    }
+
+    /// The single dimension in which this selection is unlimited, if any.
+    pub fn unlimited_dim(&self) -> Option<usize> {
+        match self {
+            SerializedSelection::Regular { count, block, .. } => count
+                .iter()
+                .zip(block)
+                .position(|(&c, &b)| c == UNLIMITED || b == UNLIMITED),
+            _ => None,
+        }
+    }
+
+    /// The rank the selection was serialized with (`None` for ALL/NONE, which
+    /// carry no rank).
+    pub fn rank(&self) -> Option<usize> {
+        match self {
+            SerializedSelection::Regular { start, .. } => Some(start.len()),
+            SerializedSelection::Blocks { rank, .. } => Some(*rank),
+            _ => None,
+        }
+    }
+}
+
+/// `H5S__hyper_deserialize`: after the type and version words.
+fn decode_hyperslab(r: &mut SelReader, version: u64) -> Result<SerializedSelection, FormatError> {
+    const REGULAR: u8 = 0x01;
+    let (flags, enc_size) = match version {
+        // v1: reserved(4) + length(4), always irregular, 4-byte coordinates.
+        1 => {
+            r.take(8)?;
+            (0u8, 4usize)
+        }
+        // v2: flags(1) + length(4), 8-byte coordinates.
+        2 => {
+            let flags = r.take(1)?[0];
+            r.take(4)?;
+            (flags, 8)
+        }
+        // v3: flags(1) + encoding size(1).
+        3 => {
+            let flags = r.take(1)?[0];
+            let enc = r.take(1)?[0] as usize;
+            (flags, enc)
+        }
+        _ => return Err(sel_err("unsupported hyperslab selection version")),
+    };
+    if flags & !REGULAR != 0 {
+        return Err(sel_err("unknown hyperslab selection flags"));
+    }
+    if !matches!(enc_size, 2 | 4 | 8) {
+        return Err(sel_err("unsupported hyperslab coordinate encoding size"));
+    }
+    let rank = r.uint(4)? as usize;
+    // HDF5 caps dataspace rank at 32 (H5S_MAX_RANK). Reject anything else so a
+    // corrupt rank can't drive a huge allocation or read loop.
+    if rank == 0 || rank > 32 {
+        return Err(sel_err("hyperslab selection rank must be 1..=32"));
+    }
+    // The all-ones value of the encoding width means "unlimited".
+    let unlim_raw = if enc_size == 8 {
+        u64::MAX
+    } else {
+        (1u64 << (enc_size * 8)) - 1
+    };
+
+    if flags & REGULAR != 0 {
         let (mut start, mut stride, mut count, mut block) = (
             Vec::with_capacity(rank),
             Vec::with_capacity(rank),
@@ -460,24 +583,104 @@ fn decode_hyperslab_serialized(
             Vec::with_capacity(rank),
         );
         for _ in 0..rank {
-        start.push(read_coord(data, pos)?);
-        pos += enc_size;
-        stride.push(read_coord(data, pos)?);
-        pos += enc_size;
-        count.push(read_coord(data, pos)?);
-        pos += enc_size;
-        block.push(read_coord(data, pos)?);
-        pos += enc_size;
+            start.push(r.uint(enc_size)?);
+            stride.push(r.uint(enc_size)?);
+            let c = r.uint(enc_size)?;
+            count.push(if c == unlim_raw { UNLIMITED } else { c });
+            let b = r.uint(enc_size)?;
+            block.push(if b == unlim_raw { UNLIMITED } else { b });
         }
-    Ok((
-        Selection::Hyperslab {
+        let unlimited = count
+            .iter()
+            .zip(&block)
+            .filter(|&(&c, &b)| c == UNLIMITED || b == UNLIMITED)
+            .count();
+        if unlimited > 1 {
+            return Err(sel_err(
+                "hyperslab selection is unlimited in more than one dimension",
+            ));
+        }
+        for d in 0..rank {
+            // Overlapping blocks are not a valid regular hyperslab.
+            if count[d] > 1 && block[d] != UNLIMITED && block[d] > stride[d] {
+                return Err(sel_err("regular hyperslab blocks overlap"));
+            }
+        }
+        return Ok(SerializedSelection::Regular {
             start,
             stride,
             count,
             block,
-        },
-        pos,
-    ))
+        });
+    }
+
+    // Irregular: number of blocks, then each block's start and end corners.
+    let nblocks = r.uint(enc_size)?;
+    let per_block = (rank * 2 * enc_size) as u64;
+    // Untrusted count: it must fit in what is left of the buffer.
+    if nblocks
+        .checked_mul(per_block)
+        .is_none_or(|need| need > r.remaining() as u64)
+    {
+        return Err(FormatError::UnexpectedEof {
+            expected: r
+                .pos
+                .saturating_add(nblocks.saturating_mul(per_block) as usize),
+            available: r.data.len(),
+        });
+    }
+    let n = nblocks as usize * rank;
+    let (mut starts, mut ends) = (Vec::with_capacity(n), Vec::with_capacity(n));
+    for _ in 0..nblocks {
+        for _ in 0..rank {
+            starts.push(r.uint(enc_size)?);
+        }
+        for _ in 0..rank {
+            ends.push(r.uint(enc_size)?);
+        }
+    }
+    if starts.iter().zip(&ends).any(|(s, e)| e < s) {
+        return Err(sel_err("hyperslab block ends before it starts"));
+    }
+    Ok(SerializedSelection::Blocks { rank, starts, ends })
+}
+
+/// The coordinates of the union of the given blocks, in row-major order.
+fn blocks_union_coords(
+    rank: usize,
+    starts: &[u64],
+    ends: &[u64],
+) -> Result<Vec<Vec<u64>>, FormatError> {
+    let mut total = 0u64;
+    for (s, e) in starts.chunks_exact(rank).zip(ends.chunks_exact(rank)) {
+        let vol = s
+            .iter()
+            .zip(e)
+            .try_fold(1u64, |acc, (&s, &e)| acc.checked_mul(e - s + 1));
+        total = vol
+            .and_then(|v| total.checked_add(v))
+            .filter(|&t| t <= MAX_EXPANDED_POINTS)
+            .ok_or_else(|| sel_err("irregular hyperslab selection is too large to expand"))?;
+    }
+    let mut out = Vec::with_capacity(total as usize);
+    for (s, e) in starts.chunks_exact(rank).zip(ends.chunks_exact(rank)) {
+        let mut cur = s.to_vec();
+        'block: loop {
+            out.push(cur.clone());
+            for d in (0..rank).rev() {
+                if cur[d] < e[d] {
+                    cur[d] += 1;
+                    continue 'block;
+                }
+                cur[d] = s[d];
+            }
+            break;
+        }
+    }
+    // Lexicographic order of coordinates is row-major order.
+    out.sort_unstable();
+    out.dedup();
+    Ok(out)
 }
 
 // ---------------------------------------------------------------------------
@@ -642,11 +845,100 @@ mod tests {
     }
 
     #[test]
-    fn decode_irregular_hyperslab_rejected() {
+    fn decode_truncated_irregular_hyperslab_is_error() {
+        // Irregular, rank 1, but the block count is missing.
         let bytes = [0x02u8, 0, 0, 0, 0x03, 0, 0, 0, 0x00, 0x02, 0x01, 0, 0, 0];
         assert!(Selection::decode_serialized(&bytes).is_err());
     }
 
+    /// Version 1 as libhdf5 writes it for the default (earliest) format bounds:
+    /// type, version, reserved(4), length(4), rank(4), nblocks(4), then each
+    /// block's start and inclusive end corner as 4-byte values.
+    fn v1_blocks(rank: u32, blocks: &[(&[u32], &[u32])]) -> Vec<u8> {
+        let mut b = Vec::new();
+        for w in [2u32, 1, 0, 0, rank, blocks.len() as u32] {
+            b.extend_from_slice(&w.to_le_bytes());
+        }
+        for (s, e) in blocks {
+            for v in s.iter().chain(e.iter()) {
+                b.extend_from_slice(&v.to_le_bytes());
+            }
+        }
+        b
+    }
+
+    #[test]
+    fn decode_v1_irregular_single_block() {
+        // Exactly what h5py/HDF5 2.0 writes for `[0:4]` with default libver.
+        let bytes = v1_blocks(1, &[(&[0], &[3])]);
+        let (sel, used) = Selection::decode_serialized(&bytes).unwrap();
+        assert_eq!(used, bytes.len());
+        assert_eq!(sel.iter_linear_1d(8).unwrap(), vec![0, 1, 2, 3]);
+    }
+
+    #[test]
+    fn decode_v1_irregular_union_is_row_major() {
+        // Blocks given out of order and overlapping still enumerate once each,
+        // in row-major order (libhdf5 iterates the union, not the list).
+        let bytes = v1_blocks(2, &[(&[1, 0], &[1, 1]), (&[0, 2], &[1, 2])]);
+        let (sel, used) = Selection::decode_serialized(&bytes).unwrap();
+        assert_eq!(used, bytes.len());
+        // (0,2) (1,0) (1,1) (1,2) in a 2x3 space.
+        assert_eq!(sel.iter_linear(&[2, 3]).unwrap(), vec![2, 3, 4, 5]);
+    }
+
+    #[test]
+    fn decode_v2_regular_with_unlimited_count() {
+        // v2: flags(1) + length(4), then 8-byte start/stride/count/block.
+        let mut b = Vec::new();
+        b.extend_from_slice(&2u32.to_le_bytes());
+        b.extend_from_slice(&2u32.to_le_bytes());
+        b.push(0x01);
+        b.extend_from_slice(&36u32.to_le_bytes());
+        b.extend_from_slice(&1u32.to_le_bytes());
+        for v in [0u64, 10, u64::MAX, 10] {
+            b.extend_from_slice(&v.to_le_bytes());
+        }
+        let (raw, used) = SerializedSelection::decode(&b).unwrap();
+        assert_eq!(used, b.len());
+        assert_eq!(raw.unlimited_dim(), Some(0));
+        assert_eq!(
+            raw,
+            SerializedSelection::Regular {
+                start: vec![0],
+                stride: vec![10],
+                count: vec![UNLIMITED],
+                block: vec![10],
+            }
+        );
+        // An unclipped unlimited selection cannot be enumerated.
+        let (sel, _) = Selection::decode_serialized(&b).unwrap();
+        assert!(sel.iter_linear_1d(100).is_err());
+    }
+
+    #[test]
+    fn decode_v3_two_byte_all_ones_is_unlimited() {
+        let bytes = [
+            0x02, 0, 0, 0, 0x03, 0, 0, 0, 0x01, 0x02, 0x01, 0, 0, 0, //
+            0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0xFF, 0xFF,
+        ];
+        let (raw, _) = SerializedSelection::decode(&bytes).unwrap();
+        assert_eq!(raw.unlimited_dim(), Some(0));
+    }
+
+    #[test]
+    fn decode_irregular_block_count_beyond_buffer_is_error() {
+        let mut b = v1_blocks(1, &[(&[0], &[3])]);
+        b[20..24].copy_from_slice(&u32::MAX.to_le_bytes());
+        assert!(Selection::decode_serialized(&b).is_err());
+    }
+
+    #[test]
+    fn decode_point_selection_is_refused() {
+        let bytes = [1u8, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
+        assert!(Selection::decode_serialized(&bytes).is_err());
+    }
+
     #[test]
     fn iter_linear_2d_block_row_major() {
         // A 2x2 block at the top-left of a 4x4 space => linear 0,1,4,5.

crates/clawhdf5-format/src/shared_message.rs

@@ -154,13 +154,29 @@ pub fn is_shared(msg_flags: u8) -> bool {
 ///
 /// When the shared flag is set on a message, the data contains a reference
 /// instead of the actual message content.
+///
+/// Assumes the file's length size equals its offset size, which only matters
+/// for version-1 references; use [`parse_shared_ref_sized`] when the
+/// superblock's length size is known.
 pub fn parse_shared_ref(data: &[u8], offset_size: u8) -> Result<SharedMessageRef, FormatError> {
+    parse_shared_ref_sized(data, offset_size, offset_size)
+}
+
+/// [`parse_shared_ref`] with the superblock's length size, which locates the
+/// object header address in a version-1 reference.
+pub fn parse_shared_ref_sized(
+    data: &[u8],
+    offset_size: u8,
+    length_size: u8,
+) -> Result<SharedMessageRef, FormatError> {
     ensure_len(data, 0, 2)?;
     let version = data[0];
     let ref_type = data[1];
 
     // Layouts (HDF5 spec IV.A.2 "Shared Message", and libhdf5's decoder):
-    //   v1: version, type, reserved(6), address     — always "committed"
+    //   v1: version, type, reserved(6), then an old-style symbol table
+    //       entry: link-name offset(length_size), object header address,
+    //       cache type(4), reserved(4), scratch(16)  — always "committed"
     //   v2: version, type, address                  — always "committed"
     //   v3: version, type, then a fractal-heap ID if type == SOHM, otherwise
     //       an address
@@ -177,7 +193,7 @@ pub fn parse_shared_ref(data: &[u8], offset_size: u8) -> Result<SharedMessageRef
         })
     };
     match version {
-        1 => address_at(2 + 6),
+        1 => address_at(2 + 6 + length_size as usize),
         2 => address_at(2),
         3 if ref_type == SHARE_TYPE_SOHM => {
             ensure_len(data, 2, FHEAP_ID_LEN)?;
@@ -225,9 +241,12 @@ pub fn parse_sohm_table_message(
 
 /// Parse the SOHM table structure (signature "SMTB") from the file.
 ///
-/// Each index entry: index_type(1) + mesg_types(2) + min_mesg_size(4) +
-///   list_max(2) + btree_min(2) + num_messages(2) + index_addr(offset_size) +
-///   heap_addr(offset_size)
+/// Each index entry: version(1) + index_type(1) + mesg_types(2) +
+///   min_mesg_size(4) + list_max(2) + btree_min(2) + num_messages(2) +
+///   index_addr(offset_size) + heap_addr(offset_size)
+///
+/// The leading per-index version byte (0) was missing here, so every field
+/// after it was read one byte off — verified against an HDF5 2.0 file.
 pub fn parse_sohm_table(
     file_data: &[u8],
     table_addr: usize,
@@ -240,11 +259,16 @@ pub fn parse_sohm_table(
     }
     let mut pos = table_addr + 4;
     let os = offset_size as usize;
-    let entry_size = 1 + 2 + 4 + 2 + 2 + 2 + os + os; // 13 + 2*offset_size
+    let entry_size = 1 + 1 + 2 + 4 + 2 + 2 + 2 + os + os; // 14 + 2*offset_size
 
     let mut indexes = Vec::with_capacity(nindexes as usize);
     for _ in 0..nindexes {
         ensure_len(file_data, pos, entry_size)?;
+        let version = file_data[pos];
+        if version != 0 {
+            return Err(FormatError::InvalidSohmTableVersion(version));
+        }
+        pos += 1;
         let index_type = file_data[pos];
         pos += 1;
         let mesg_types = u16::from_le_bytes([file_data[pos], file_data[pos + 1]]);
@@ -381,6 +405,68 @@ pub fn parse_sohm_btree_entries(
 // ---- SOHM resolution ----
 
 /// Find the SOHM index that handles the given message type.
+/// Load a file's SOHM table: superblock → superblock extension → Shared
+/// Message Table message → SMTB. `Ok(None)` when the file has no superblock
+/// extension or no shared-message table.
+pub fn load_sohm_table(
+    file_data: &[u8],
+    offset_size: u8,
+    length_size: u8,
+) -> Result<Option<SohmTable>, FormatError> {
+    let sig = crate::signature::find_signature(file_data)?;
+    let sb = crate::superblock::Superblock::parse(file_data, sig)?;
+    let Some(ext_addr) = sb
+        .superblock_extension_address
+        .filter(|&a| !is_undefined(a, offset_size))
+    else {
+        return Ok(None);
+    };
+    let ext = ObjectHeader::parse(file_data, ext_addr as usize, offset_size, length_size)?;
+    let Some(msg) = ext
+        .messages
+        .iter()
+        .find(|m| m.msg_type == MessageType::SharedMessageTable)
+    else {
+        return Ok(None);
+    };
+    let table_msg = parse_sohm_table_message(&msg.data, offset_size)?;
+    parse_sohm_table(
+        file_data,
+        table_msg.table_address as usize,
+        table_msg.nindexes,
+        offset_size,
+    )
+    .map(Some)
+}
+
+/// Like [`message_data`], but also follows references into the file's SOHM
+/// heap (shared object header messages), loading the SOHM table on demand.
+pub fn message_data_with_sohm<'a>(
+    file_data: &[u8],
+    msg: &'a crate::object_header::HeaderMessage,
+    offset_size: u8,
+    length_size: u8,
+) -> Result<Cow<'a, [u8]>, FormatError> {
+    if !is_shared(msg.flags) {
+        return Ok(Cow::Borrowed(&msg.data));
+    }
+    let shared_ref = parse_shared_ref_sized(&msg.data, offset_size, length_size)?;
+    let table = if shared_ref.heap_id.is_some() {
+        load_sohm_table(file_data, offset_size, length_size)?
+    } else {
+        None
+    };
+    resolve_shared_message_with_sohm(
+        file_data,
+        &shared_ref,
+        msg.msg_type,
+        offset_size,
+        length_size,
+        table.as_ref(),
+    )
+    .map(Cow::Owned)
+}
+
 fn find_index_for_msg_type(table: &SohmTable, msg_type: MessageType) -> Option<&SohmIndex> {
     let type_bit = 1u16 << msg_type.to_u16();
     table
@@ -444,7 +530,7 @@ pub fn message_data<'a>(
     if !is_shared(msg.flags) {
         return Ok(Cow::Borrowed(&msg.data));
     }
-    let shared_ref = parse_shared_ref(&msg.data, offset_size)?;
+    let shared_ref = parse_shared_ref_sized(&msg.data, offset_size, length_size)?;
     resolve_shared_message(
         file_data,
         &shared_ref,
@@ -459,7 +545,8 @@ pub fn message_data<'a>(
 ///
 /// For type 1/3 (shared in another object header), reads the target object header
 /// and finds the message of the specified type.
-/// For type 2 (SOHM), uses the fractal heap from the SOHM table.
+/// For type 2 (SOHM), uses the fractal heap from the file's SOHM table,
+/// loaded from the superblock extension on demand.
 pub fn resolve_shared_message(
     file_data: &[u8],
     shared_ref: &SharedMessageRef,
@@ -467,13 +554,18 @@ pub fn resolve_shared_message(
     offset_size: u8,
     length_size: u8,
 ) -> Result<Vec<u8>, FormatError> {
+    let table = if shared_ref.heap_id.is_some() {
+        load_sohm_table(file_data, offset_size, length_size)?
+    } else {
+        None
+    };
     resolve_shared_message_with_sohm(
         file_data,
         shared_ref,
         target_msg_type,
         offset_size,
         length_size,
-        None,
+        table.as_ref(),
     )
 }
 
@@ -579,15 +671,26 @@ mod tests {
 
     #[test]
     fn parse_v1_ref() {
-        let mut data = Vec::new();
-        data.push(1); // version
-        data.push(0); // type
-        data.extend_from_slice(&[0u8; 6]); // reserved
-        data.extend_from_slice(&0x5678u64.to_le_bytes());
+        // Datatype message of `/group1/dset2` in HDF5's `tcompound.h5`
+        // (written in 2000): version 1, six reserved bytes, then an old-style
+        // symbol table entry — link-name offset 0x10, object header address
+        // 0x590 (the committed datatype `/type1`), cache type, reserved and
+        // scratch.
+        let mut data = vec![1, 0, 0, 0, 0, 0, 0, 0];
+        data.extend_from_slice(&0x10u64.to_le_bytes());
+        data.extend_from_slice(&0x590u64.to_le_bytes());
+        data.extend_from_slice(&[0; 24]);
 
-        let shared = parse_shared_ref(&data, 8).unwrap();
+        let shared = parse_shared_ref_sized(&data, 8, 8).unwrap();
         assert_eq!(shared.version, 1);
-        assert_eq!(shared.object_header_address, Some(0x5678));
+        assert_eq!(shared.object_header_address, Some(0x590));
+
+        // The name offset is a length: 4 bytes here, then an 8-byte address.
+        let mut data = vec![1, 0, 0, 0, 0, 0, 0, 0];
+        data.extend_from_slice(&0x10u32.to_le_bytes());
+        data.extend_from_slice(&0x590u64.to_le_bytes());
+        let shared = parse_shared_ref_sized(&data, 8, 4).unwrap();
+        assert_eq!(shared.object_header_address, Some(0x590));
     }
 
     #[test]
@@ -707,6 +810,7 @@ mod tests {
         let mut buf = Vec::new();
         buf.extend_from_slice(b"SMTB");
         for idx in indexes {
+            buf.push(0); // version
             buf.push(idx.index_type);
             buf.extend_from_slice(&idx.mesg_types.to_le_bytes());
             buf.extend_from_slice(&idx.min_mesg_size.to_le_bytes());

crates/clawhdf5-format/src/signature.rs

@@ -11,6 +11,16 @@ pub const HDF5_SIGNATURE: [u8; 8] = [0x89, b'H', b'D', b'F', b'\r', b'\n', 0x1A,
 /// (powers of two starting at 512, plus offset 0).
 ///
 /// Returns the byte offset where the signature was found.
+///
+/// A non-zero offset means the file starts with a *user block*, and every
+/// address inside the file is relative to the superblock's position, not to
+/// byte 0 (libhdf5 uses the signature's position as the base address even
+/// when the stored base-address field disagrees). The parsers in this crate
+/// take addresses as indices into `file_data`, so they must be handed the
+/// bytes from the signature on — use [`split_user_block`]. [`Superblock::parse`]
+/// refuses a non-zero offset for this reason.
+///
+/// [`Superblock::parse`]: crate::superblock::Superblock::parse
 pub fn find_signature(data: &[u8]) -> Result<usize, FormatError> {
     // Check offset 0
     if data.len() >= 8 && data[..8] == HDF5_SIGNATURE {
@@ -29,6 +39,17 @@ pub fn find_signature(data: &[u8]) -> Result<usize, FormatError> {
     Err(FormatError::SignatureNotFound)
 }
 
+/// Split a file into its user block and its HDF5 bytes.
+///
+/// Returns `(user_block, hdf5)`: `user_block` is everything before the
+/// superblock signature (empty for most files) and `hdf5` is the rest, in
+/// which every HDF5 address is a plain index. Pass `hdf5` as `file_data` to
+/// every parser in this crate, and parse the superblock at offset 0 of it.
+pub fn split_user_block(data: &[u8]) -> Result<(&[u8], &[u8]), FormatError> {
+    let offset = find_signature(data)?;
+    Ok(data.split_at(offset))
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -88,6 +109,21 @@ mod tests {
         assert_eq!(find_signature(&data), Err(FormatError::SignatureNotFound));
     }
 
+    #[test]
+    fn split_user_block_rebases_at_the_signature() {
+        let mut data = vec![7u8; 1024];
+        data[512..520].copy_from_slice(&HDF5_SIGNATURE);
+        let (ub, hdf5) = split_user_block(&data).unwrap();
+        assert_eq!(ub.len(), 512);
+        assert_eq!(hdf5.len(), 512);
+        assert_eq!(&hdf5[..8], &HDF5_SIGNATURE);
+
+        data[..8].copy_from_slice(&HDF5_SIGNATURE);
+        let (ub, hdf5) = split_user_block(&data).unwrap();
+        assert!(ub.is_empty());
+        assert_eq!(hdf5.len(), 1024);
+    }
+
     #[test]
     fn signature_prefers_earliest() {
         // Signature at both 0 and 512, should return 0

crates/clawhdf5-format/src/superblock.rs

@@ -39,7 +39,13 @@ pub struct Superblock {
     pub superblock_extension_address: Option<u64>,
     /// CRC32C checksum (v2/v3 only).
     pub checksum: Option<u32>,
-    /// Page size for page-buffer mode (v4 only). `None` for v0–v3.
+    /// Page size of the non-standard "version 4" superblock layout (v4 only).
+    /// `None` for v0–v3.
+    ///
+    /// HDF5 has no superblock version 4 — libhdf5 refuses it. A real paged
+    /// file is a v2/v3 superblock whose extension holds a File Space Info
+    /// message (what `FileWriter::with_page_size` writes). This field is kept
+    /// only so such files written by older clawhdf5 versions still parse.
     pub page_size: Option<u32>,
 }
 
@@ -94,6 +100,42 @@ pub mod swmr_flags {
 }
 
 impl Superblock {
+    /// Where the HDF5 data ends, relative to the superblock, for a file of
+    /// `file_len` bytes whose superblock is at `user_block` (both counted
+    /// from the start of the file), with libhdf5's truncation check
+    /// (`H5F__super_read`).
+    ///
+    /// The superblock records the end of the file's data as an absolute
+    /// address. A file shorter than that was truncated, and libhdf5 refuses
+    /// to open it ("truncated file"); so does this, with
+    /// [`FormatError::TruncatedFile`]. Bytes past that address are not part
+    /// of the file: libhdf5 fails any read of them ("addr overflow" /
+    /// "address plus size exceeds file eoa"), so a reader should parse only
+    /// the data up to the returned end. As libhdf5 does for a SWMR reader,
+    /// the check is skipped for a version-3 superblock whose writer is still
+    /// writing it in SWMR mode (it extends the file as it goes); the data
+    /// then ends at the end of the file.
+    ///
+    /// When the superblock's recorded base address differs from where the
+    /// superblock actually is (a user block added or removed after the file
+    /// was written), libhdf5 moves the recorded end of file by the same
+    /// amount, and so does this.
+    pub fn data_end(&self, user_block: u64, file_len: u64) -> Result<u64, FormatError> {
+        let eof =
+            i128::from(self.eof_address) - i128::from(self.base_address) + i128::from(user_block);
+        if eof < 0 || eof > i128::from(file_len) {
+            if self.version >= 3 && self.is_swmr_write() {
+                return Ok(file_len.saturating_sub(user_block));
+            }
+            return Err(FormatError::TruncatedFile {
+                stored_eof: u64::try_from(eof).unwrap_or(self.eof_address),
+                actual_len: file_len,
+            });
+        }
+        // 0 <= eof <= file_len, so it fits a u64.
+        Ok((eof as u64).saturating_sub(user_block))
+    }
+
     /// Whether the file was opened with write access when the superblock was written.
     pub fn is_write_access(&self) -> bool {
         self.consistency_flags & swmr_flags::WRITE_ACCESS != 0
@@ -127,8 +169,9 @@ impl Superblock {
 
     /// Serialize this superblock to bytes.
     ///
-    /// Writes v2/v3 format, or v4 (with `page_size`) when `self.version == 4`.
-    /// Computes and appends Jenkins lookup3 checksum.
+    /// Writes v2/v3 format, or the non-standard v4 (with `page_size`) when
+    /// `self.version == 4` — which no HDF5 library opens; see
+    /// [`Self::page_size`]. Computes and appends Jenkins lookup3 checksum.
     pub fn serialize(&self) -> Vec<u8> {
         let mut buf = Vec::with_capacity(48);
         buf.extend_from_slice(&HDF5_SIGNATURE);
@@ -167,8 +210,18 @@ impl Superblock {
 
     /// Parse a superblock from `data` starting at `signature_offset`.
     ///
-    /// The signature must be present at the given offset.
+    /// The signature must be present at the given offset, and that offset
+    /// must be 0: every address in an HDF5 file is relative to the
+    /// superblock, so when a file has a user block (signature at 512, 1024,
+    /// …) the caller must pass the bytes from the signature on — see
+    /// [`crate::signature::split_user_block`] — and use that slice as
+    /// `file_data` everywhere. A non-zero offset is refused with
+    /// [`FormatError::UserBlockNotStripped`] because the addresses in the
+    /// returned superblock would otherwise be applied to the wrong bytes.
     pub fn parse(data: &[u8], signature_offset: usize) -> Result<Superblock, FormatError> {
+        if signature_offset != 0 {
+            return Err(FormatError::UserBlockNotStripped(signature_offset as u64));
+        }
         let d = data
             .get(signature_offset..)
             .ok_or(FormatError::UnexpectedEof {
@@ -520,6 +573,30 @@ mod tests {
         buf
     }
 
+    #[test]
+    fn data_end_refuses_truncated_files_like_libhdf5() {
+        // build_v2_bytes records base 0, end of file 2048.
+        let sb = Superblock::parse(&build_v2_bytes(8, 2), 0).unwrap();
+        assert_eq!(sb.data_end(0, 2048), Ok(2048));
+        // Bytes past the recorded end are not part of the file.
+        assert_eq!(sb.data_end(0, 4096), Ok(2048));
+        assert_eq!(
+            sb.data_end(0, 2047),
+            Err(FormatError::TruncatedFile {
+                stored_eof: 2048,
+                actual_len: 2047
+            })
+        );
+        // A user block added in front after the file was written (the
+        // recorded base address is still 0): the end moves with it.
+        assert_eq!(sb.data_end(512, 2560), Ok(2048));
+        assert!(sb.data_end(512, 2559).is_err());
+        // A v3 superblock still being written in SWMR mode is not checked.
+        let mut swmr = Superblock::parse(&build_v2_bytes(8, 3), 0).unwrap();
+        swmr.consistency_flags = swmr_flags::WRITE_ACCESS | swmr_flags::SWMR_WRITE;
+        assert_eq!(swmr.data_end(0, 1000), Ok(1000));
+    }
+
     #[test]
     fn parse_v0_8byte_offsets() {
         let data = build_v0_bytes(8);
@@ -669,7 +746,16 @@ mod tests {
         let mut data = vec![0u8; 1024];
         let v0 = build_v0_bytes(8);
         data[512..512 + v0.len()].copy_from_slice(&v0);
-        let sb = Superblock::parse(&data, 512).unwrap();
+        // Addresses are relative to the superblock, so parsing in place
+        // (where they would be applied to the whole buffer) is refused...
+        assert_eq!(
+            Superblock::parse(&data, 512),
+            Err(FormatError::UserBlockNotStripped(512))
+        );
+        // ...and the caller parses the bytes from the signature on.
+        let (ub, hdf5) = crate::signature::split_user_block(&data).unwrap();
+        assert_eq!(ub.len(), 512);
+        let sb = Superblock::parse(hdf5, 0).unwrap();
         assert_eq!(sb.version, 0);
         assert_eq!(sb.root_group_address, 96);
     }

crates/clawhdf5-format/src/test_fuzz.rs

@@ -0,0 +1,149 @@
+//! Mutation fuzzing for the filter decoders (tests only).
+//!
+//! A decoder fed a random or mutated frame may fail, but must not panic —
+//! tests build with overflow checks and debug assertions, so an unchecked
+//! subtraction, multiplication or shift on a header field, or an
+//! out-of-range slice, fails the test — and must not return more than its
+//! output limit.
+
+#[cfg(not(feature = "std"))]
+extern crate alloc;
+#[cfg(not(feature = "std"))]
+use alloc::vec::Vec;
+
+use crate::error::FormatError;
+
+/// xorshift64*: deterministic, so a failure reproduces.
+pub(crate) struct Rng(u64);
+
+impl Rng {
+    pub(crate) fn new(seed: u64) -> Rng {
+        Rng(seed.max(1))
+    }
+
+    pub(crate) fn next_u64(&mut self) -> u64 {
+        let mut x = self.0;
+        x ^= x >> 12;
+        x ^= x << 25;
+        x ^= x >> 27;
+        self.0 = x;
+        x.wrapping_mul(0x2545_F491_4F6C_DD1D)
+    }
+
+    /// Uniform in `0..n` (`n` > 0).
+    pub(crate) fn below(&mut self, n: usize) -> usize {
+        (self.next_u64() % n as u64) as usize
+    }
+
+    pub(crate) fn bytes(&mut self, n: usize) -> Vec<u8> {
+        (0..n).map(|_| self.next_u64() as u8).collect()
+    }
+
+    /// A u32 that tends to hit edge cases in size and offset fields.
+    fn interesting_u32(&mut self, len: usize) -> u32 {
+        match self.below(10) {
+            0 => 0,
+            1 => 1,
+            2 => self.below(20) as u32,
+            3 => 15 + self.below(3) as u32,
+            4 => u32::MAX - self.below(16) as u32,
+            5 => 1 << self.below(32),
+            6 => (len as u32)
+                .wrapping_add(self.below(9) as u32)
+                .wrapping_sub(4),
+            7 => i32::MAX as u32,
+            _ => self.next_u64() as u32,
+        }
+    }
+}
+
+/// One to four random edits of `seed`.
+pub(crate) fn mutate(rng: &mut Rng, seed: &[u8]) -> Vec<u8> {
+    let mut v = seed.to_vec();
+    for _ in 0..1 + rng.below(4) {
+        let len = v.len();
+        match rng.below(9) {
+            0 if len > 0 => {
+                let i = rng.below(len);
+                v[i] ^= 1 << rng.below(8);
+            }
+            1 if len > 0 => {
+                let i = rng.below(len);
+                v[i] = rng.next_u64() as u8;
+            }
+            2 if len > 0 => {
+                let i = rng.below(len);
+                v[i] = [0, 0xff, 0x7f, 0x80, 0x20, 0x1f][rng.below(6)];
+            }
+            // A size or offset field: little- or big-endian, anywhere, but
+            // most often in the first 32 bytes where headers live.
+            3 | 4 if len >= 4 => {
+                let span = if rng.below(2) == 0 { len.min(32) } else { len };
+                let i = rng.below(span - 3);
+                let x = rng.interesting_u32(len);
+                let b = if rng.below(2) == 0 {
+                    x.to_le_bytes()
+                } else {
+                    x.to_be_bytes()
+                };
+                v[i..i + 4].copy_from_slice(&b);
+            }
+            5 if len > 0 => v.truncate(rng.below(len)),
+            6 => {
+                let n = 1 + rng.below(64);
+                let extra = rng.bytes(n);
+                v.extend_from_slice(&extra);
+            }
+            7 if len > 1 => {
+                let a = rng.below(len);
+                let b = a + rng.below(len - a);
+                let copy = v[a..b].to_vec();
+                let at = rng.below(len);
+                v.splice(at..at, copy);
+            }
+            _ if len > 0 => {
+                let i = rng.below(len);
+                v[i] = v[i].wrapping_add(1 + rng.below(3) as u8);
+            }
+            _ => v.push(rng.next_u64() as u8),
+        }
+    }
+    v
+}
+
+/// Feed `iters` inputs to `decode`: mostly mutations of `seeds`, some pure
+/// noise and some truncated seeds. Asserts only "no panic, output within
+/// `limit`".
+pub(crate) fn fuzz_decoder(
+    seed: u64,
+    seeds: &[Vec<u8>],
+    iters: usize,
+    limit: usize,
+    mut decode: impl FnMut(&[u8]) -> Result<Vec<u8>, FormatError>,
+) {
+    assert!(!seeds.is_empty());
+    let mut rng = Rng::new(seed);
+    for s in seeds {
+        // The seeds themselves must be valid, or the fuzz explores nothing.
+        decode(s).expect("seed frame must decode");
+    }
+    for _ in 0..iters {
+        let input = match rng.below(16) {
+            0 => {
+                let n = rng.below(96);
+                rng.bytes(n)
+            }
+            1 => {
+                let s = &seeds[rng.below(seeds.len())];
+                s[..rng.below(s.len() + 1)].to_vec()
+            }
+            _ => {
+                let s = &seeds[rng.below(seeds.len())];
+                mutate(&mut rng, s)
+            }
+        };
+        if let Ok(out) = decode(&input) {
+            assert!(out.len() <= limit, "decoded {} > limit {limit}", out.len());
+        }
+    }
+}

crates/clawhdf5-format/src/type_builders.rs

@@ -15,31 +15,83 @@ use crate::datatype::{
 
 /// Controls when fill values are written to dataset storage.
 ///
-/// Corresponds to the HDF5 fill value message's "fill time" field.
+/// Corresponds to the HDF5 fill value message's "fill time" field
+/// (`H5D_fill_time_t`).
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
 pub enum FillTime {
-    /// Never write fill values (0x02). Avoids initialization overhead
-    /// for datasets that will be fully written before any read.
+    /// Never write fill values (`H5D_FILL_TIME_NEVER`). Avoids
+    /// initialization overhead for datasets that will be fully written
+    /// before any read.
     Never,
-    /// Write fill values at allocation time (0x0a). This is the default
-    /// and matches the HDF5 C library's behavior.
-    #[default]
+    /// Write fill values when storage is allocated (`H5D_FILL_TIME_ALLOC`).
     Alloc,
-    /// Write fill values only when the fill value has been explicitly set (0x06).
+    /// Write fill values at allocation only if one was set explicitly
+    /// (`H5D_FILL_TIME_IFSET`). The default, as in the HDF5 C library.
+    #[default]
     IfSet,
 }
 
+/// Space allocation time written with every fill value message: late
+/// (`H5D_ALLOC_TIME_LATE`), bits 0-1 of the flags byte.
+const ALLOC_TIME_LATE: u8 = 2;
+
 impl FillTime {
-    /// Serialize to the byte used in the fill value message (version 3).
+    /// Serialize to the flags byte of a version 3 fill value message: the
+    /// space allocation time (late) in bits 0-1 and the fill time in bits
+    /// 2-3 (`H5D_FILL_TIME_ALLOC` = 0, `NEVER` = 1, `IFSET` = 2).
+    ///
+    /// This used to put `Never` in the ALLOC slot, `Alloc` in IFSET and
+    /// `IfSet` in NEVER, so libhdf5 saw every choice as a different one.
     pub fn to_byte(self) -> u8 {
+        ALLOC_TIME_LATE | (self.code() << 2)
+    }
+
+    /// Decode the fill time from a version 3 fill value message's flags.
+    pub fn from_byte(flags: u8) -> Option<FillTime> {
+        match (flags >> 2) & 0x03 {
+            0 => Some(FillTime::Alloc),
+            1 => Some(FillTime::Never),
+            2 => Some(FillTime::IfSet),
+            _ => None,
+        }
+    }
+
+    fn code(self) -> u8 {
         match self {
-            FillTime::Never => 0x02,
-            FillTime::Alloc => 0x0a,
-            FillTime::IfSet => 0x06,
+            FillTime::Alloc => 0,
+            FillTime::Never => 1,
+            FillTime::IfSet => 2,
         }
     }
 }
 
+/// Serialize a version 3 Fill Value message for a dataset of `dt`: the fill
+/// time, and the user-defined fill value if there is one (bit 5).
+pub(crate) fn fill_value_message(
+    fill_time: FillTime,
+    value: Option<&[u8]>,
+    dt: &Datatype,
+) -> Result<Vec<u8>, crate::error::FormatError> {
+    let mut msg = vec![3, fill_time.to_byte()];
+    if let Some(value) = value {
+        if matches!(dt, Datatype::VariableLength { .. }) {
+            return Err(crate::error::FormatError::SerializationError(
+                "a fill value for a variable-length datatype is not supported".into(),
+            ));
+        }
+        if value.len() != dt.type_size() as usize {
+            return Err(crate::error::FormatError::DataSizeMismatch {
+                expected: dt.type_size() as usize,
+                actual: value.len(),
+            });
+        }
+        msg[1] |= 0x20; // fill value defined
+        msg.extend_from_slice(&(value.len() as u32).to_le_bytes());
+        msg.extend_from_slice(value);
+    }
+    Ok(msg)
+}
+
 // ---- Datatype constructors ----
 
 pub fn make_f64_type() -> Datatype {
@@ -332,7 +384,11 @@ pub(crate) fn build_attr_message(name: &str, value: &AttrValue) -> AttributeMess
             raw_data: data.clone(),
         },
         AttrValue::String(s) => {
-            let bytes = s.as_bytes();
+            // A fixed-length string type must be at least 1 byte: libhdf5
+            // rejects size 0 ("invalid datatype size") and with it every
+            // attribute on the object. h5py stores "" as one NUL byte.
+            let mut bytes = s.as_bytes().to_vec();
+            bytes.resize(bytes.len().max(1), 0);
             AttributeMessage {
                 name: name.to_string(),
                 datatype: Datatype::String {
@@ -341,11 +397,12 @@ pub(crate) fn build_attr_message(name: &str, value: &AttrValue) -> AttributeMess
                     charset: CharacterSet::Utf8,
                 },
                 dataspace: scalar_ds(),
-                raw_data: bytes.to_vec(),
+                raw_data: bytes,
             }
         }
         AttrValue::StringArray(arr) => {
-            let max_len = arr.iter().map(|s| s.len()).max().unwrap_or(0);
+            // At least 1 byte per element, as for a single string.
+            let max_len = arr.iter().map(|s| s.len()).max().unwrap_or(0).max(1);
             let mut raw = Vec::new();
             for s in arr {
                 let mut b = s.as_bytes().to_vec();
@@ -431,8 +488,10 @@ pub struct DatasetBuilder {
     pub(crate) data: Option<Vec<u8>>,
     pub(crate) attrs: Vec<(String, AttrValue)>,
     pub(crate) chunk_options: ChunkOptions,
-    /// Controls when fill values are written. Default is `FillTime::Alloc`.
+    /// Controls when fill values are written. Default is `FillTime::IfSet`.
     pub(crate) fill_time: FillTime,
+    /// User-defined fill value: one element's bytes, as stored.
+    pub(crate) fill_value: Option<Vec<u8>>,
     /// Use compact (inline) storage: data is stored in the object header.
     /// Only valid when raw data is <= 65536 bytes and dataset is not chunked.
     pub(crate) compact: bool,
@@ -459,6 +518,7 @@ impl DatasetBuilder {
             attrs: Vec::new(),
             chunk_options: ChunkOptions::default(),
             fill_time: FillTime::default(),
+            fill_value: None,
             compact: false,
             alignment: 0,
             virtual_sources: None,
@@ -635,8 +695,13 @@ impl DatasetBuilder {
         self
     }
 
+    /// Set attribute `name`. Setting it again replaces the earlier value,
+    /// as `attrs[name] = v` does in h5py.
     pub fn set_attr(&mut self, name: &str, value: AttrValue) -> &mut Self {
-        self.attrs.push((name.to_string(), value));
+        match self.attrs.iter_mut().find(|(n, _)| n == name) {
+            Some(slot) => slot.1 = value,
+            None => self.attrs.push((name.to_string(), value)),
+        }
         self
     }
 
@@ -671,7 +736,67 @@ impl DatasetBuilder {
         self
     }
 
-    /// Enable Pcodec lossless numerical compression (clawhdf5 filter ID 32023).
+    /// Compress with a plugin filter ([`PluginFilter`]), in the format the
+    /// libhdf5 plugin reads (h5py, hdf5plugin). Implies chunked storage.
+    /// Each filter needs its cargo feature (`lzf`, ...); writing fails with
+    /// `UnsupportedFilter` without it.
+    ///
+    /// [`PluginFilter`]: crate::chunked_write::PluginFilter
+    pub fn with_plugin_filter(&mut self, filter: crate::chunked_write::PluginFilter) -> &mut Self {
+        self.chunk_options.plugin = Some(filter);
+        self
+    }
+
+    /// Enable LZF compression (filter 32000) — h5py's built-in
+    /// `compression="lzf"`. Implies chunked storage; shuffle is applied
+    /// first unless `.without_shuffle()`. Requires the `lzf` cargo feature.
+    pub fn with_lzf(&mut self) -> &mut Self {
+        self.with_plugin_filter(crate::chunked_write::PluginFilter::Lzf)
+    }
+
+    /// Enable bitshuffle (filter 32008) with `compression` after the bit
+    /// transpose, in bitshuffle's default block size. Implies chunked
+    /// storage; no byte shuffle is added. Requires the `bitshuffle` cargo
+    /// feature.
+    pub fn with_bitshuffle(
+        &mut self,
+        compression: crate::chunked_write::BitshuffleCompression,
+    ) -> &mut Self {
+        self.with_plugin_filter(crate::chunked_write::PluginFilter::Bitshuffle {
+            block_size: 0,
+            compression,
+        })
+    }
+
+    /// Enable bzip2 (filter 307) at block size `level` (1-9). Implies
+    /// chunked storage; shuffle is applied first unless
+    /// `.without_shuffle()`. Requires the `bzip2` cargo feature.
+    pub fn with_bzip2(&mut self, level: u32) -> &mut Self {
+        self.with_plugin_filter(crate::chunked_write::PluginFilter::Bzip2 { level })
+    }
+
+    /// Enable Blosc (filter 32001) with `codec` at `level` (0-9) after
+    /// `shuffle`. Implies chunked storage; no extra HDF5 shuffle is added.
+    /// Requires the `blosc` cargo feature.
+    pub fn with_blosc(
+        &mut self,
+        codec: crate::chunked_write::BloscCodec,
+        level: u32,
+        shuffle: crate::chunked_write::BloscShuffle,
+    ) -> &mut Self {
+        self.with_plugin_filter(crate::chunked_write::PluginFilter::Blosc {
+            codec,
+            level,
+            shuffle,
+        })
+    }
+
+    /// Enable Pcodec lossless numerical compression (private clawhdf5 filter
+    /// ID 480).
+    ///
+    /// **Not interoperable:** pcodec has no registered HDF5 filter ID and no
+    /// libhdf5 plugin, so h5py and other HDF5 readers cannot read the
+    /// dataset — only clawhdf5 built with the `pcodec` feature can.
     ///
     /// Pcodec achieves 30–94% better compression ratio than Zstd for f32/f64
     /// columns at 1–5 GiB/s decompression speed (arXiv:2502.06112). Requires
@@ -715,10 +840,20 @@ impl DatasetBuilder {
         self
     }
 
+    /// Set the dataset's fill value: what readers return for storage that
+    /// was never written (e.g. after the dataset is extended). `value` is one
+    /// element's bytes as stored — the dataset datatype's size and byte order
+    /// (`(-1i32).to_le_bytes()` for an `i32` dataset). A size mismatch, or a
+    /// variable-length datatype, makes `finish` fail.
+    pub fn with_fill_value(&mut self, value: &[u8]) -> &mut Self {
+        self.fill_value = Some(value.to_vec());
+        self
+    }
+
     /// Use compact (inline) storage for this dataset.
     ///
     /// The raw data is stored directly in the dataset's object header rather
-    /// than as a separate data blob. Only effective when raw data <= 65536 bytes
+    /// than as a separate data blob. Only effective when raw data <= 65531 bytes
     /// and the dataset is not chunked.
     pub fn compact(&mut self) -> &mut Self {
         self.compact = true;
@@ -773,34 +908,117 @@ impl DatasetBuilder {
 
 // ---- Group builder ----
 
-/// Builder for groups.
+/// One entry of a [`GroupBuilder`], kept in the order it was added (the
+/// order a group that tracks creation order lists its links in).
+pub(crate) enum GroupItem {
+    Dataset(Box<DatasetBuilder>),
+    Group(GroupBuilder),
+    /// A soft link: `name` resolves to whatever `target` names when read.
+    Soft {
+        name: String,
+        target: String,
+    },
+    /// An extra hard link to the object at `target` (a path in this file).
+    Hard {
+        name: String,
+        target: String,
+    },
+    /// An external link to `path` in the file `file`.
+    External {
+        name: String,
+        file: String,
+        path: String,
+    },
+}
+
+/// Builder for a group: its datasets, subgroups, links and attributes.
+///
+/// Names are paths relative to the group: `create_dataset("a/b/x")` creates
+/// the groups `a` and `a/b` as needed, as h5py does. A group added where a
+/// group of the same path already exists (added by another builder, or
+/// created as an intermediate group) is merged into it, like h5py's
+/// `require_group`; any other name used twice in a group is an error when the
+/// file is written. A path component must not be empty or `"."`.
 pub struct GroupBuilder {
     pub(crate) name: String,
-    pub(crate) datasets: Vec<DatasetBuilder>,
+    pub(crate) items: Vec<GroupItem>,
     pub(crate) attrs: Vec<(String, AttrValue)>,
-    /// (link_name, target_file, target_path)
-    pub(crate) external_links: Vec<(String, String, String)>,
+    /// Track (and index) link creation order; `None` follows the file's
+    /// default (`FileWriter::track_order`).
+    pub(crate) track_order: Option<bool>,
 }
 
 impl GroupBuilder {
     pub(crate) fn new(name: &str) -> Self {
         Self {
             name: name.to_string(),
-            datasets: Vec::new(),
+            items: Vec::new(),
             attrs: Vec::new(),
-            external_links: Vec::new(),
+            track_order: None,
         }
     }
 
+    /// Create a dataset in this group. `name` may be a relative path
+    /// (`"a/b/x"`); missing intermediate groups are created.
     pub fn create_dataset(&mut self, name: &str) -> &mut DatasetBuilder {
-        self.datasets.push(DatasetBuilder::new(name));
-        self.datasets.last_mut().unwrap()
+        self.items
+            .push(GroupItem::Dataset(Box::new(DatasetBuilder::new(name))));
+        match self.items.last_mut() {
+            Some(GroupItem::Dataset(d)) => d,
+            _ => unreachable!("just pushed a dataset"),
+        }
+    }
+
+    /// Start a subgroup of this group. Like `FileWriter::create_group`, the
+    /// builder is detached: fill it, then pass `finish()`'s result to
+    /// [`Self::add_group`]. `name` may be a relative path.
+    pub fn create_group(&self, name: &str) -> GroupBuilder {
+        GroupBuilder::new(name)
+    }
+
+    /// Add a finished subgroup to this group.
+    pub fn add_group(&mut self, group: FinishedGroup) -> &mut Self {
+        self.items.push(GroupItem::Group(group.group));
+        self
     }
 
     pub fn set_attr(&mut self, name: &str, value: AttrValue) {
         self.attrs.push((name.to_string(), value));
     }
 
+    /// Track the creation order of this group's links, and index it, as
+    /// h5py's `track_order=True` does: libhdf5 (and h5py) then list the
+    /// group's members in the order they were added rather than by name.
+    /// Applies to links only, not to attributes.
+    pub fn track_order(&mut self, track: bool) -> &mut Self {
+        self.track_order = Some(track);
+        self
+    }
+
+    /// Add a soft link `name` to the path `target` (absolute, or relative to
+    /// this group), like h5py's `grp[name] = h5py.SoftLink(target)`. The
+    /// target need not exist.
+    pub fn add_soft_link(&mut self, name: &str, target: &str) -> &mut Self {
+        self.items.push(GroupItem::Soft {
+            name: name.to_string(),
+            target: target.to_string(),
+        });
+        self
+    }
+
+    /// Add another hard link `name` to the group or dataset at `target`
+    /// (absolute, or relative to this group), like h5py's
+    /// `grp[name] = f[target]`. The target must be written in the same file;
+    /// its path may go through other hard links, but not through soft or
+    /// external links.
+    pub fn add_hard_link(&mut self, name: &str, target: &str) -> &mut Self {
+        self.items.push(GroupItem::Hard {
+            name: name.to_string(),
+            target: target.to_string(),
+        });
+        self
+    }
+
     /// Add an external link: a named pointer to an object in another HDF5 file.
     pub fn add_external_link(
         &mut self,
@@ -808,30 +1026,21 @@ impl GroupBuilder {
         target_file: &str,
         target_path: &str,
     ) -> &mut Self {
-        self.external_links.push((
-            name.to_string(),
-            target_file.to_string(),
-            target_path.to_string(),
-        ));
+        self.items.push(GroupItem::External {
+            name: name.to_string(),
+            file: target_file.to_string(),
+            path: target_path.to_string(),
+        });
         self
     }
 
     /// Consume the builder, returning a FinishedGroup to add to FileWriter.
     pub fn finish(self) -> FinishedGroup {
-        FinishedGroup {
-            name: self.name,
-            datasets: self.datasets,
-            attrs: self.attrs,
-            external_links: self.external_links,
-        }
+        FinishedGroup { group: self }
     }
 }
 
 /// A finished group ready for the file writer.
 pub struct FinishedGroup {
-    pub(crate) name: String,
-    pub(crate) datasets: Vec<DatasetBuilder>,
-    pub(crate) attrs: Vec<(String, AttrValue)>,
-    /// (link_name, target_file, target_path)
-    pub(crate) external_links: Vec<(String, String, String)>,
+    pub(crate) group: GroupBuilder,
 }

crates/clawhdf5-format/src/vl_data.rs

@@ -5,10 +5,12 @@
 //! `sequence_length(4 LE) + collection_address(offset_size LE) + object_index(4 LE)`.
 
 #[cfg(not(feature = "std"))]
-use alloc::{string::String, vec::Vec};
+use alloc::{collections::BTreeMap, format, string::String, vec, vec::Vec};
+#[cfg(feature = "std")]
+use std::collections::BTreeMap;
 
 use crate::error::FormatError;
-use crate::global_heap::GlobalHeapCollection;
+use crate::global_heap::{GlobalHeapCollection, GlobalHeapIndex};
 
 /// A parsed variable-length element reference (global heap ID).
 #[derive(Debug, Clone)]
@@ -109,7 +111,218 @@ fn is_undefined_address(addr: u64, offset_size: u8) -> bool {
     }
 }
 
+/// The size of one variable-length element in a file with `offset_size`-byte
+/// addresses: a sequence length (4), a global heap collection address and an
+/// object index (4). libhdf5 computes it this way rather than trusting the
+/// datatype message (`H5T_set_loc`).
+pub fn element_size(offset_size: u8) -> usize {
+    4 + offset_size as usize + 4
+}
+
+/// Refuse a variable-length datatype whose stored element size is not the
+/// one this file's offset size implies. Its elements would be laid out with
+/// a stride libhdf5 does not use, so every value after the first would be
+/// read from the wrong place.
+pub fn check_element_size(stored_size: u32, offset_size: u8) -> Result<(), FormatError> {
+    let expected = element_size(offset_size);
+    if stored_size as usize != expected {
+        return Err(FormatError::VlDataError(format!(
+            "variable-length datatype stores {stored_size}-byte elements; a file with \
+             {offset_size}-byte offsets uses {expected}"
+        )));
+    }
+    Ok(())
+}
+
+/// A collection's objects, located in the file data but not copied:
+/// `(index, offset, size)` of the first object with each index, sorted by
+/// index.
+struct CachedCollection {
+    objects: Vec<(u16, usize, usize)>,
+}
+
+impl CachedCollection {
+    fn new(index: GlobalHeapIndex) -> Self {
+        let mut objects: Vec<(u16, usize, usize)> = index
+            .objects
+            .iter()
+            .map(|o| (o.index, o.offset, o.size))
+            .collect();
+        // Stable, so the first object with a repeated index is kept.
+        objects.sort_by_key(|o| o.0);
+        objects.dedup_by_key(|o| o.0);
+        Self { objects }
+    }
+
+    /// What this entry costs to keep, in bytes (roughly).
+    fn cost(&self) -> usize {
+        64 + self.objects.len() * core::mem::size_of::<(u16, usize, usize)>()
+    }
+
+    fn get(&self, index: u32) -> Option<(usize, usize)> {
+        let index = u16::try_from(index).ok()?;
+        let i = self.objects.binary_search_by_key(&index, |o| o.0).ok()?;
+        Some((self.objects[i].1, self.objects[i].2))
+    }
+}
+
+/// How many bytes of collection indexes a [`VlResolver`] keeps before it
+/// drops them and starts again. Values are never copied into the cache, so
+/// this bounds what a read retains however many collections it visits.
+const CACHE_BUDGET: usize = 32 << 20;
+
+/// Resolves variable-length elements against a file's global heap, parsing
+/// each heap collection once however many elements point into it.
+///
+/// Values follow libhdf5: an element whose heap address is 0 is null (an
+/// empty string or sequence), and an element whose heap object is not
+/// exactly `length × base size` bytes is an error ("Expected global heap
+/// object size does not match"), not a truncated or padded value.
+///
+/// Memory stays bounded on hostile files: the cache holds where each
+/// object lies, not a copy of it, up to a fixed budget; and collections
+/// that overlap one another are refused (libhdf5 never writes them), so a
+/// file cannot make the resolver parse the same bytes as the objects of
+/// many collections.
+pub struct VlResolver<'a> {
+    file_data: &'a [u8],
+    offset_size: u8,
+    length_size: u8,
+    cache: BTreeMap<u64, CachedCollection>,
+    cached_bytes: usize,
+    budget: usize,
+    /// Start → end of every collection parsed so far (kept when the cache
+    /// is dropped, to check overlaps).
+    extents: BTreeMap<usize, usize>,
+}
+
+impl<'a> VlResolver<'a> {
+    /// A resolver over `file_data` (the file from its superblock on), with
+    /// the superblock's offset and length sizes.
+    pub fn new(file_data: &'a [u8], offset_size: u8, length_size: u8) -> Self {
+        Self {
+            file_data,
+            offset_size,
+            length_size,
+            cache: BTreeMap::new(),
+            cached_bytes: 0,
+            budget: CACHE_BUDGET,
+            extents: BTreeMap::new(),
+        }
+    }
+
+    /// The size of one element in this file (see [`element_size`]).
+    pub fn element_size(&self) -> usize {
+        element_size(self.offset_size)
+    }
+
+    /// Split `raw` into elements; its length must be a whole number of them.
+    fn elements(&self, raw: &[u8]) -> Result<Vec<VlElement>, FormatError> {
+        let size = self.element_size();
+        if !raw.len().is_multiple_of(size) {
+            return Err(FormatError::VlDataError(format!(
+                "{} bytes is not a whole number of {size}-byte variable-length elements",
+                raw.len()
+            )));
+        }
+        parse_vl_references(raw, (raw.len() / size) as u64, self.offset_size)
+    }
+
+    /// The bytes of one element: `length × base_size` bytes from the heap,
+    /// or `None` for a null element.
+    fn resolve(
+        &mut self,
+        vl: &VlElement,
+        base_size: usize,
+    ) -> Result<Option<&'a [u8]>, FormatError> {
+        let addr = vl.collection_address;
+        if addr == 0 {
+            return Ok(None);
+        }
+        let data = self.object(vl)?;
+        let expected = (vl.length as usize)
+            .checked_mul(base_size)
+            .ok_or_else(|| FormatError::Overflow("variable-length element size".into()))?;
+        if data.len() != expected {
+            return Err(FormatError::VlDataError(format!(
+                "global heap object {} in the collection at {addr} holds {} bytes; the element \
+                 says {} × {base_size}",
+                vl.object_index,
+                data.len(),
+                vl.length
+            )));
+        }
+        Ok(Some(data))
+    }
+
+    /// One element (the first [`element_size`](Self::element_size) bytes of
+    /// `elem`) of a variable-length sequence whose base type is `base_size`
+    /// bytes: its `length × base_size` bytes, or `None` for a null element
+    /// (heap address 0).
+    pub fn element(
+        &mut self,
+        elem: &[u8],
+        base_size: usize,
+    ) -> Result<Option<&'a [u8]>, FormatError> {
+        let vl = parse_vl_references(elem, 1, self.offset_size)?;
+        self.resolve(&vl[0], base_size)
+    }
+
+    /// One variable-length string element: its bytes up to the first NUL,
+    /// or `None` for a null element (h5dump prints it as `NULL`, h5py
+    /// returns it as empty).
+    pub fn string_element(&mut self, elem: &[u8]) -> Result<Option<&'a [u8]>, FormatError> {
+        Ok(self.element(elem, 1)?.map(cut_at_nul))
+    }
+
+    /// The strings of the variable-length string elements in `raw`, as
+    /// bytes. A string ends at its first NUL, as libhdf5 returns it (it
+    /// converts each to a C string); a null element is empty.
+    pub fn string_bytes(&mut self, raw: &[u8]) -> Result<Vec<Vec<u8>>, FormatError> {
+        self.elements(raw)?
+            .iter()
+            .map(|vl| Ok(self.resolve(vl, 1)?.map(cut_at_nul).unwrap_or(&[]).to_vec()))
+            .collect()
+    }
+
+    /// The strings of the variable-length string elements in `raw`, decoded
+    /// as UTF-8 with invalid sequences replaced by U+FFFD (see
+    /// [`string_bytes`](Self::string_bytes) for the exact bytes).
+    pub fn strings(&mut self, raw: &[u8]) -> Result<Vec<String>, FormatError> {
+        Ok(self
+            .string_bytes(raw)?
+            .into_iter()
+            .map(|b| match String::from_utf8(b) {
+                Ok(s) => s,
+                Err(e) => String::from_utf8_lossy(e.as_bytes()).into_owned(),
+            })
+            .collect())
+    }
+
+    /// The sequences of the variable-length sequence elements in `raw`, each
+    /// as its `length × base_size` bytes in the base type's encoding.
+    pub fn sequences(&mut self, raw: &[u8], base_size: usize) -> Result<Vec<Vec<u8>>, FormatError> {
+        if base_size == 0 {
+            return Err(FormatError::VlDataError(
+                "variable-length sequence of a zero-size base type".into(),
+            ));
+        }
+        self.elements(raw)?
+            .iter()
+            .map(|vl| Ok(self.resolve(vl, base_size)?.unwrap_or(&[]).to_vec()))
+            .collect()
+    }
+}
+
+/// A string's bytes up to its first NUL.
+fn cut_at_nul(s: &[u8]) -> &[u8] {
+    &s[..s.iter().position(|&b| b == 0).unwrap_or(s.len())]
+}
+
 /// Resolve VL strings from raw data by looking up each element in the global heap.
+///
+/// Reads the first `num_elements` elements of `raw`. Strings end at their
+/// first NUL and invalid UTF-8 is replaced, as in [`VlResolver::strings`].
 pub fn read_vl_strings(
     file_data: &[u8],
     raw_data: &[u8],
@@ -117,38 +330,33 @@ pub fn read_vl_strings(
     offset_size: u8,
     length_size: u8,
 ) -> Result<Vec<String>, FormatError> {
-    let refs = parse_vl_references(raw_data, num_elements, offset_size)?;
-    let mut result = Vec::with_capacity(refs.len());
-
-    for vl in &refs {
-        if vl.length == 0 && is_undefined_address(vl.collection_address, offset_size) {
-            result.push(String::new());
-            continue;
-        }
-        if vl.length == 0 && vl.collection_address == 0 {
-            result.push(String::new());
-            continue;
-        }
-
-        let coll =
-            GlobalHeapCollection::parse(file_data, vl.collection_address as usize, length_size)?;
-        let obj = coll.get_object(vl.object_index as u16).ok_or(
-            FormatError::GlobalHeapObjectNotFound {
-                collection_address: vl.collection_address,
-                index: vl.object_index as u16,
-            },
-        )?;
-
-        // The object data is the raw string bytes
-        let len = (vl.length as usize).min(obj.data.len());
-        let s = String::from_utf8_lossy(&obj.data[..len]).into_owned();
-        result.push(s);
-    }
-
-    Ok(result)
+    let raw = first_elements(raw_data, num_elements, offset_size)?;
+    VlResolver::new(file_data, offset_size, length_size).strings(raw)
 }
 
-/// Resolve VL byte sequences from raw data.
+/// The first `num_elements` elements of `raw`, or an error if it is shorter.
+fn first_elements(raw: &[u8], num_elements: u64, offset_size: u8) -> Result<&[u8], FormatError> {
+    let total = usize::try_from(num_elements)
+        .ok()
+        .and_then(|n| n.checked_mul(element_size(offset_size)))
+        .ok_or(FormatError::UnexpectedEof {
+            expected: usize::MAX,
+            available: raw.len(),
+        })?;
+    raw.get(..total).ok_or(FormatError::UnexpectedEof {
+        expected: total,
+        available: raw.len(),
+    })
+}
+
+/// Resolve VL sequences from raw data, returning each element's bytes.
+///
+/// Each element is the sequence's full encoding — element count × base type
+/// size bytes, in the base type's byte order — so a sequence of `i32` yields
+/// four bytes per value. Decode it with the base type (e.g.
+/// [`crate::data_read::read_as_i64`]). This does not know the base type, so
+/// it returns each heap object whole; [`VlResolver::sequences`] also checks
+/// the object's size against the element's length.
 pub fn read_vl_bytes(
     file_data: &[u8],
     raw_data: &[u8],
@@ -157,33 +365,97 @@ pub fn read_vl_bytes(
     length_size: u8,
 ) -> Result<Vec<Vec<u8>>, FormatError> {
     let refs = parse_vl_references(raw_data, num_elements, offset_size)?;
+    let mut resolver = VlResolver::new(file_data, offset_size, length_size);
     let mut result = Vec::with_capacity(refs.len());
 
     for vl in &refs {
-        if vl.length == 0
-            && (is_undefined_address(vl.collection_address, offset_size)
-                || vl.collection_address == 0)
-        {
+        // A heap address of 0 is a null element, as in VlResolver.
+        if vl.collection_address == 0 {
             result.push(Vec::new());
             continue;
         }
-
-        let coll =
-            GlobalHeapCollection::parse(file_data, vl.collection_address as usize, length_size)?;
-        let obj = coll.get_object(vl.object_index as u16).ok_or(
-            FormatError::GlobalHeapObjectNotFound {
-                collection_address: vl.collection_address,
-                index: vl.object_index as u16,
-            },
-        )?;
-
-        let len = (vl.length as usize).min(obj.data.len());
-        result.push(obj.data[..len].to_vec());
+        // The heap object holds the whole sequence. `vl.length` counts
+        // elements, not bytes, so it is only the byte length when the base
+        // type is one byte wide.
+        let obj = resolver.object(vl)?;
+        result.push(obj.to_vec());
     }
 
     Ok(result)
 }
 
+impl<'a> VlResolver<'a> {
+    /// The heap object `vl` points to, whatever its size; its collection is
+    /// parsed on first use.
+    fn object(&mut self, vl: &VlElement) -> Result<&'a [u8], FormatError> {
+        let addr = vl.collection_address;
+        // libhdf5 writes a null element with address 0, never the undefined
+        // address, and fails to read one ("addr undefined") even when its
+        // length is 0; we returned an empty value.
+        if is_undefined_address(addr, self.offset_size) {
+            return Err(FormatError::VlDataError(format!(
+                "variable-length element (length {}) has the undefined global heap address",
+                vl.length
+            )));
+        }
+        if !self.cache.contains_key(&addr) {
+            let offset = usize::try_from(addr).map_err(|_| FormatError::UnexpectedEof {
+                expected: usize::MAX,
+                available: self.file_data.len(),
+            })?;
+            let index =
+                GlobalHeapCollection::parse_index(self.file_data, offset, self.length_size)?;
+            // parse_index checked that the collection lies in the file.
+            let end = offset + index.collection_size as usize;
+            self.check_overlap(offset, end)?;
+            let coll = CachedCollection::new(index);
+            if self.cached_bytes.saturating_add(coll.cost()) > self.budget {
+                self.cache.clear();
+                self.cached_bytes = 0;
+            }
+            self.cached_bytes += coll.cost();
+            self.cache.insert(addr, coll);
+        }
+        let (start, size) = self.cache[&addr].get(vl.object_index).ok_or(
+            FormatError::GlobalHeapObjectNotFound {
+                collection_address: addr,
+                index: vl.object_index as u16,
+            },
+        )?;
+        Ok(&self.file_data[start..start + size])
+    }
+
+    /// Record the collection at `start..end`, refusing one that overlaps a
+    /// collection already read. libhdf5 allocates each collection its own
+    /// block; overlapping ones only come from a crafted file, where they let
+    /// every byte be parsed again as the objects of each collection.
+    fn check_overlap(&mut self, start: usize, end: usize) -> Result<(), FormatError> {
+        if let Some(&known) = self.extents.get(&start) {
+            return if known == end {
+                Ok(())
+            } else {
+                Err(FormatError::VlDataError(format!(
+                    "global heap collection at {start} changed size"
+                )))
+            };
+        }
+        let before = self.extents.range(..start).next_back();
+        let after = self.extents.range(start..).next();
+        let clash = match (before, after) {
+            (Some((&s, &e)), _) if e > start => Some(s),
+            (_, Some((&s, _))) if s < end => Some(s),
+            _ => None,
+        };
+        if let Some(other) = clash {
+            return Err(FormatError::VlDataError(format!(
+                "global heap collection at {start} overlaps the one at {other}"
+            )));
+        }
+        self.extents.insert(start, end);
+        Ok(())
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -278,16 +550,27 @@ mod tests {
     }
 
     #[test]
-    fn null_vl_element_empty_string() {
-        // length=0, address=undefined
-        let mut raw = Vec::new();
-        raw.extend_from_slice(&0u32.to_le_bytes()); // length=0
-        raw.extend_from_slice(&u64::MAX.to_le_bytes()); // undefined address
-        raw.extend_from_slice(&0u32.to_le_bytes()); // index
-
-        let file_data = vec![0u8; 16];
-        let strings = read_vl_strings(&file_data, &raw, 1, 8, 8).unwrap();
-        assert_eq!(strings, vec![""]);
+    fn an_undefined_heap_address_is_an_error_even_at_length_0() {
+        // libhdf5 fails the read ("addr undefined"); h5py and libhdf5 write
+        // a null element with address 0. We returned "".
+        let mut file_data = vec![0u8; 256];
+        build_gcol_at(&mut file_data, 64, &[(1, b"x")]);
+        for (os, undef) in [(8u8, u64::MAX), (4, 0xFFFF_FFFF)] {
+            for length in [0, 1] {
+                let mut raw = element(1, 64, 1, os);
+                raw.extend(element(length, undef, 1, os));
+                let mut r = VlResolver::new(&file_data, os, 8);
+                let e = r.string_bytes(&raw).unwrap_err().to_string();
+                assert!(e.contains("undefined"), "{e}");
+                assert!(r.sequences(&raw, 1).is_err());
+                assert!(r.string_element(&raw[raw.len() / 2..]).is_err());
+                let n = 2;
+                assert!(read_vl_strings(&file_data, &raw, n, os, 8).is_err());
+                assert!(read_vl_bytes(&file_data, &raw, n, os, 8).is_err());
+                // The defined element alone still reads.
+                assert_eq!(r.strings(&raw[..raw.len() / 2]).unwrap(), ["x"]);
+            }
+        }
     }
 
     #[test]
@@ -326,6 +609,126 @@ mod tests {
         assert_eq!(bytes, vec![vec![0xDE, 0xAD], vec![0xBE, 0xEF, 0xCA]]);
     }
 
+    fn element(length: u32, addr: u64, index: u32, offset_size: u8) -> Vec<u8> {
+        let mut raw = length.to_le_bytes().to_vec();
+        raw.extend_from_slice(&addr.to_le_bytes()[..offset_size as usize]);
+        raw.extend_from_slice(&index.to_le_bytes());
+        raw
+    }
+
+    #[test]
+    fn strings_end_at_the_first_nul() {
+        // libhdf5 hands each VL string over as a C string, so h5py sees
+        // "a\0b" as "a"; we used to return the NUL and what followed.
+        let mut file_data = vec![0u8; 512];
+        build_gcol_at(&mut file_data, 64, &[(1, b"a\0b"), (2, b"cd")]);
+        let mut raw = element(3, 64, 1, 8);
+        raw.extend(element(2, 64, 2, 8));
+        let mut r = VlResolver::new(&file_data, 8, 8);
+        assert_eq!(
+            r.string_bytes(&raw).unwrap(),
+            vec![b"a".to_vec(), b"cd".to_vec()]
+        );
+        assert_eq!(
+            read_vl_strings(&file_data, &raw, 2, 8, 8).unwrap(),
+            ["a", "cd"]
+        );
+    }
+
+    #[test]
+    fn a_heap_object_of_the_wrong_size_is_an_error() {
+        // libhdf5: "Expected global heap object size does not match". We
+        // used to return the object cut to the element's length.
+        let mut file_data = vec![0u8; 512];
+        build_gcol_at(&mut file_data, 64, &[(1, b"cdefgh"), (2, &[1, 0, 0, 0])]);
+        let mut r = VlResolver::new(&file_data, 8, 8);
+        assert!(r.string_bytes(&element(3, 64, 1, 8)).is_err());
+        assert!(r.string_bytes(&element(9, 64, 1, 8)).is_err());
+        assert!(read_vl_strings(&file_data, &element(3, 64, 1, 8), 1, 8, 8).is_err());
+        // A sequence of one i32 is 4 bytes; of two, 8.
+        assert_eq!(
+            r.sequences(&element(1, 64, 2, 8), 4).unwrap(),
+            vec![vec![1, 0, 0, 0]]
+        );
+        assert!(r.sequences(&element(2, 64, 2, 8), 4).is_err());
+        assert!(r.sequences(&element(1, 64, 2, 8), 0).is_err());
+    }
+
+    #[test]
+    fn address_zero_is_null_whatever_the_length() {
+        // libhdf5 treats a heap address of 0 as a null element.
+        let file_data = vec![0u8; 64];
+        let mut r = VlResolver::new(&file_data, 8, 8);
+        assert_eq!(
+            r.string_bytes(&element(5, 0, 1, 8)).unwrap(),
+            vec![Vec::<u8>::new()]
+        );
+        assert_eq!(
+            r.sequences(&element(5, 0, 1, 8), 4).unwrap(),
+            vec![Vec::<u8>::new()]
+        );
+    }
+
+    #[test]
+    fn four_byte_offsets_use_twelve_byte_elements() {
+        let mut file_data = vec![0u8; 512];
+        build_gcol_at(&mut file_data, 64, &[(1, b"one"), (2, b""), (3, b"three")]);
+        let mut raw = element(3, 64, 1, 4);
+        raw.extend(element(0, 64, 2, 4));
+        raw.extend(element(5, 64, 3, 4));
+        assert_eq!(raw.len(), 36);
+        let mut r = VlResolver::new(&file_data, 4, 8);
+        assert_eq!(r.element_size(), 12);
+        assert_eq!(r.strings(&raw).unwrap(), ["one", "", "three"]);
+        // Not a whole number of elements.
+        assert!(r.strings(&raw[..30]).is_err());
+    }
+
+    #[test]
+    fn the_cache_stays_within_its_budget_and_rereads_what_it_dropped() {
+        // Twenty collections of three objects each; a budget that holds
+        // about two of them. Reading every element twice must still return
+        // the right strings after the cache is dropped.
+        let mut file_data = vec![0u8; 64];
+        let mut raw = Vec::new();
+        for c in 0..20u64 {
+            let at = file_data.len();
+            let names: Vec<String> = (0..3).map(|i| format!("c{c}o{i}")).collect();
+            let objs: Vec<(u16, &[u8])> = names
+                .iter()
+                .enumerate()
+                .map(|(i, n)| (i as u16 + 1, n.as_bytes()))
+                .collect();
+            build_gcol_at(&mut file_data, at, &objs);
+            for (i, n) in names.iter().enumerate() {
+                raw.extend(element(n.len() as u32, at as u64, i as u32 + 1, 8));
+            }
+        }
+        raw.extend(raw.clone());
+        let mut r = VlResolver::new(&file_data, 8, 8);
+        let one = CachedCollection {
+            objects: vec![(0, 0, 0); 3],
+        }
+        .cost();
+        r.budget = 2 * one + 1;
+        let want: Vec<String> = (0..2)
+            .flat_map(|_| (0..20).flat_map(|c| (0..3).map(move |i| format!("c{c}o{i}"))))
+            .collect();
+        for (k, chunk) in raw.chunks(16).enumerate() {
+            assert_eq!(r.strings(chunk).unwrap(), [want[k].clone()]);
+            assert!(r.cached_bytes <= r.budget);
+            assert!(r.cache.len() <= 2);
+        }
+    }
+
+    #[test]
+    fn element_size_is_checked_against_the_offset_size() {
+        assert!(check_element_size(16, 8).is_ok());
+        assert!(check_element_size(12, 4).is_ok());
+        assert!(check_element_size(16, 4).is_err());
+        assert!(check_element_size(524_304, 8).is_err());
+    }
+
     #[test]
     fn parse_vl_references_truncated_error() {
         let raw = vec![0u8; 10]; // too short for 1 element with offset_size=8

crates/clawhdf5-format/src/writer_tree.rs

@@ -0,0 +1,494 @@
+//! The group hierarchy `FileWriter` writes: builders flattened into a tree
+//! of groups, datasets and links, with path names expanded into
+//! intermediate groups, hard links resolved to objects, reference counts
+//! counted, and everything put in layout order.
+
+#[cfg(not(feature = "std"))]
+use alloc::{
+    collections::BTreeMap,
+    format,
+    string::{String, ToString},
+    vec,
+    vec::Vec,
+};
+#[cfg(feature = "std")]
+use std::collections::BTreeMap;
+
+use crate::error::FormatError;
+use crate::type_builders::{AttrValue, DatasetBuilder, GroupBuilder, GroupItem};
+
+/// Depth of the chain of unresolved hard links followed while resolving one
+/// hard-link target path (a bound on recursion; cycles are found exactly).
+const MAX_LINK_DEPTH: usize = 64;
+
+fn err(msg: String) -> FormatError {
+    FormatError::SerializationError(msg)
+}
+
+/// A link name must be one path component: not empty, not ".", and without
+/// '/' (a '/' separates components, so it cannot be part of a name).
+fn check_link_name(name: &str, path: &str) -> Result<(), FormatError> {
+    if name.is_empty() || name == "." || name.contains('/') {
+        return Err(err(format!(
+            "invalid object name {path:?}: every path component must be a \
+             non-empty name other than \".\""
+        )));
+    }
+    Ok(())
+}
+
+/// What a link in the final tree points at.
+#[derive(Debug, Clone, PartialEq)]
+pub(crate) enum LinkTo {
+    /// A group, by index into [`Tree::groups`] (layout order).
+    Group(usize),
+    /// A dataset, by index into [`Tree::datasets`] (layout order).
+    Dataset(usize),
+    Soft(String),
+    External {
+        file: String,
+        path: String,
+    },
+}
+
+pub(crate) struct Link {
+    pub(crate) name: String,
+    pub(crate) to: LinkTo,
+    /// Set when the group tracks creation order.
+    pub(crate) creation_order: Option<u64>,
+}
+
+pub(crate) struct Group {
+    pub(crate) attrs: Vec<(String, AttrValue)>,
+    /// Links in the order they are written.
+    pub(crate) links: Vec<Link>,
+    pub(crate) track_order: bool,
+    /// Number of hard links to this group (the root counts one for the
+    /// superblock's reference).
+    pub(crate) refcount: u32,
+}
+
+/// The flattened file: groups (root first) and datasets, both in the order
+/// they are laid out in the file.
+pub(crate) struct Tree {
+    pub(crate) groups: Vec<Group>,
+    pub(crate) datasets: Vec<(DatasetBuilder, u32)>,
+}
+
+// ---- construction ----
+
+enum Target {
+    Group(usize),
+    Dataset(usize),
+    Soft(String),
+    Hard(String),
+    External { file: String, path: String },
+}
+
+struct BuildGroup {
+    /// Full path, for messages.
+    path: String,
+    attrs: Vec<(String, AttrValue)>,
+    links: Vec<(String, Target)>,
+    by_name: BTreeMap<String, usize>,
+    track_order: Option<bool>,
+}
+
+struct Builder {
+    groups: Vec<BuildGroup>,
+    datasets: Vec<DatasetBuilder>,
+}
+
+fn join(parent: &str, name: &str) -> String {
+    if parent == "/" {
+        format!("/{name}")
+    } else {
+        format!("{parent}/{name}")
+    }
+}
+
+impl Builder {
+    fn new_group(&mut self, path: String) -> usize {
+        self.groups.push(BuildGroup {
+            path,
+            attrs: Vec::new(),
+            links: Vec::new(),
+            by_name: BTreeMap::new(),
+            track_order: None,
+        });
+        self.groups.len() - 1
+    }
+
+    /// Split `path` (relative to group `g`) into the group holding its last
+    /// component, creating missing intermediate groups, and that component.
+    fn parent_of<'p>(&mut self, g: usize, path: &'p str) -> Result<(usize, &'p str), FormatError> {
+        // An absolute path is accepted at the root only.
+        let rel = match path.strip_prefix('/') {
+            Some(rest) if g == 0 => rest,
+            Some(_) => {
+                return Err(err(format!(
+                    "invalid object name {path:?} in {}: absolute paths are accepted \
+                     only at the root",
+                    self.groups[g].path
+                )));
+            }
+            None => path,
+        };
+        let mut comps: Vec<&str> = rel.split('/').collect();
+        let last = comps.pop().unwrap_or("");
+        check_link_name(last, path)?;
+        let mut cur = g;
+        for c in comps {
+            check_link_name(c, path)?;
+            cur = match self.groups[cur].by_name.get(c).copied() {
+                Some(i) => match self.groups[cur].links[i].1 {
+                    Target::Group(child) => child,
+                    _ => {
+                        return Err(err(format!(
+                            "cannot create {path:?} in {}: {c:?} exists and is not a group",
+                            self.groups[g].path
+                        )));
+                    }
+                },
+                None => {
+                    let child = self.new_group(join(&self.groups[cur].path, c));
+                    self.push_link(cur, c, Target::Group(child))?;
+                    child
+                }
+            };
+        }
+        Ok((cur, last))
+    }
+
+    fn push_link(&mut self, g: usize, name: &str, to: Target) -> Result<(), FormatError> {
+        let grp = &mut self.groups[g];
+        if grp.by_name.contains_key(name) {
+            return Err(err(format!("{:?} already exists", join(&grp.path, name))));
+        }
+        grp.by_name.insert(name.to_string(), grp.links.len());
+        grp.links.push((name.to_string(), to));
+        Ok(())
+    }
+
+    /// Add `item` to group `g`.
+    fn add_item(&mut self, g: usize, item: GroupItem) -> Result<(), FormatError> {
+        match item {
+            GroupItem::Dataset(db) => {
+                let (parent, name) = self.parent_of(g, &db.name)?;
+                let name = name.to_string();
+                self.push_link(parent, &name, Target::Dataset(self.datasets.len()))?;
+                self.datasets.push(*db);
+            }
+            GroupItem::Group(gb) => self.add_group(g, gb)?,
+            GroupItem::Soft { name, target } => {
+                if target.is_empty() {
+                    return Err(err(format!("soft link {name:?} has an empty target")));
+                }
+                let (parent, last) = self.parent_of(g, &name)?;
+                self.push_link(parent, last, Target::Soft(target))?;
+            }
+            GroupItem::Hard { name, target } => {
+                let (parent, last) = self.parent_of(g, &name)?;
+                self.push_link(parent, last, Target::Hard(target))?;
+            }
+            GroupItem::External { name, file, path } => {
+                if file.is_empty() || path.is_empty() {
+                    return Err(err(format!(
+                        "external link {name:?} needs a file name and an object path"
+                    )));
+                }
+                let (parent, last) = self.parent_of(g, &name)?;
+                self.push_link(parent, last, Target::External { file, path })?;
+            }
+        }
+        Ok(())
+    }
+
+    /// Add the group `gb` (named by a path relative to group `g`), merging it
+    /// into a group already at that path.
+    fn add_group(&mut self, g: usize, gb: GroupBuilder) -> Result<(), FormatError> {
+        let (parent, last) = self.parent_of(g, &gb.name)?;
+        let idx = match self.groups[parent].by_name.get(last).copied() {
+            Some(i) => match self.groups[parent].links[i].1 {
+                Target::Group(child) => child,
+                _ => {
+                    return Err(err(format!(
+                        "{:?} already exists and is not a group",
+                        join(&self.groups[parent].path, last)
+                    )));
+                }
+            },
+            None => {
+                let child = self.new_group(join(&self.groups[parent].path, last));
+                self.push_link(parent, last, Target::Group(child))?;
+                child
+            }
+        };
+        self.merge_into(idx, gb)
+    }
+
+    /// Merge a builder's attributes, setting and items into group `idx`.
+    fn merge_into(&mut self, idx: usize, gb: GroupBuilder) -> Result<(), FormatError> {
+        // An attribute set again (by this builder or a merged one) takes the
+        // new value, as assigning `attrs[name]` in h5py does.
+        for (name, value) in gb.attrs {
+            let attrs = &mut self.groups[idx].attrs;
+            match attrs.iter_mut().find(|(n, _)| *n == name) {
+                Some(slot) => slot.1 = value,
+                None => attrs.push((name, value)),
+            }
+        }
+        if let Some(t) = gb.track_order {
+            match self.groups[idx].track_order {
+                Some(old) if old != t => {
+                    return Err(err(format!(
+                        "conflicting track_order settings for {}",
+                        self.groups[idx].path
+                    )));
+                }
+                _ => self.groups[idx].track_order = Some(t),
+            }
+        }
+        for item in gb.items {
+            self.add_item(idx, item)?;
+        }
+        Ok(())
+    }
+
+    /// The object a hard link's `target` path names, from group `from`.
+    ///
+    /// Hard links met on the way are resolved once and remembered in
+    /// `memo` (by group and link index), so a target that goes through
+    /// other hard links costs time linear in the links, not exponential; a
+    /// hard link met again while it is being resolved is a cycle.
+    fn resolve(
+        &self,
+        memo: &mut [Vec<Resolution>],
+        from: usize,
+        target: &str,
+        depth: usize,
+    ) -> Result<Obj, FormatError> {
+        if depth > MAX_LINK_DEPTH {
+            return Err(err(format!(
+                "hard link target {target:?}: more than {MAX_LINK_DEPTH} hard links \
+                 to follow"
+            )));
+        }
+        let (mut cur, rest) = match target.strip_prefix('/') {
+            Some(rest) => (0, rest),
+            None => (from, target),
+        };
+        if target.is_empty() {
+            return Err(err("a hard link needs a target path".to_string()));
+        }
+        let comps: Vec<&str> = rest
+            .split('/')
+            .filter(|c| !c.is_empty() && *c != ".")
+            .collect();
+        let mut obj = Obj::Group(cur);
+        for (i, c) in comps.iter().enumerate() {
+            let Obj::Group(g) = obj else {
+                return Err(err(format!(
+                    "hard link target {target:?}: {:?} is not a group",
+                    comps[..i].join("/")
+                )));
+            };
+            cur = g;
+            let grp = &self.groups[cur];
+            let Some(&li) = grp.by_name.get(*c) else {
+                return Err(err(format!(
+                    "hard link target {target:?} does not exist in the file"
+                )));
+            };
+            obj = match &grp.links[li].1 {
+                Target::Group(child) => Obj::Group(*child),
+                Target::Dataset(d) => Obj::Dataset(*d),
+                Target::Hard(p) => match memo[cur][li] {
+                    Resolution::Done(o) => o,
+                    Resolution::InProgress => {
+                        return Err(err(format!(
+                            "hard link target {target:?}: the hard link {:?} leads \
+                             back to itself (a cycle)",
+                            join(&grp.path, c)
+                        )));
+                    }
+                    Resolution::Todo => {
+                        memo[cur][li] = Resolution::InProgress;
+                        let o = self.resolve(memo, cur, p, depth + 1)?;
+                        memo[cur][li] = Resolution::Done(o);
+                        o
+                    }
+                },
+                Target::Soft(_) | Target::External { .. } => {
+                    return Err(err(format!(
+                        "hard link target {target:?} goes through a soft or external \
+                         link ({:?}); name the object by its hard-link path",
+                        join(&grp.path, c)
+                    )));
+                }
+            };
+        }
+        Ok(obj)
+    }
+}
+
+/// Where resolving one hard link has got to.
+#[derive(Clone, Copy)]
+enum Resolution {
+    Todo,
+    InProgress,
+    Done(Obj),
+}
+
+#[derive(Clone, Copy)]
+enum Obj {
+    Group(usize),
+    Dataset(usize),
+}
+
+/// Flatten the root group builder into a [`Tree`]. `default_track_order`
+/// applies to every group that does not set its own.
+pub(crate) fn build(root: GroupBuilder, default_track_order: bool) -> Result<Tree, FormatError> {
+    let mut b = Builder {
+        groups: Vec::new(),
+        datasets: Vec::new(),
+    };
+    b.new_group("/".to_string());
+    b.merge_into(0, root)?;
+
+    // Resolve hard links and count references.
+    let mut group_refs = vec![0u32; b.groups.len()];
+    let mut ds_refs = vec![0u32; b.datasets.len()];
+    group_refs[0] = 1; // the superblock's reference to the root
+    let mut memo: Vec<Vec<Resolution>> = b
+        .groups
+        .iter()
+        .map(|g| vec![Resolution::Todo; g.links.len()])
+        .collect();
+    let mut resolved: Vec<Vec<Option<Obj>>> = Vec::with_capacity(b.groups.len());
+    for (gi, g) in b.groups.iter().enumerate() {
+        let mut row = Vec::with_capacity(g.links.len());
+        for (li, (_, t)) in g.links.iter().enumerate() {
+            let obj = match t {
+                Target::Group(i) => Some(Obj::Group(*i)),
+                Target::Dataset(d) => Some(Obj::Dataset(*d)),
+                Target::Hard(p) => Some(match memo[gi][li] {
+                    Resolution::Done(o) => o,
+                    _ => {
+                        memo[gi][li] = Resolution::InProgress;
+                        let o = b.resolve(&mut memo, gi, p, 0)?;
+                        memo[gi][li] = Resolution::Done(o);
+                        o
+                    }
+                }),
+                Target::Soft(_) | Target::External { .. } => None,
+            };
+            match obj {
+                Some(Obj::Group(i)) => group_refs[i] += 1,
+                Some(Obj::Dataset(d)) => ds_refs[d] += 1,
+                None => {}
+            }
+            row.push(obj);
+        }
+        resolved.push(row);
+    }
+
+    // The order each group's links are written in: creation order when
+    // tracked; otherwise datasets, then groups, then other links (the order
+    // earlier versions wrote, so one-level files keep their layout).
+    let tracked: Vec<bool> = b
+        .groups
+        .iter()
+        .map(|g| g.track_order.unwrap_or(default_track_order))
+        .collect();
+    let link_order: Vec<Vec<usize>> = b
+        .groups
+        .iter()
+        .enumerate()
+        .map(|(gi, g)| {
+            let mut idx: Vec<usize> = (0..g.links.len()).collect();
+            if !tracked[gi] {
+                idx.sort_by_key(|&i| match g.links[i].1 {
+                    Target::Dataset(_) => 0,
+                    Target::Group(_) => 1,
+                    _ => 2,
+                });
+            }
+            idx
+        })
+        .collect();
+
+    // Layout order: groups depth-first from the root, following the links
+    // that created them; datasets group by group in that order.
+    let mut group_order = Vec::with_capacity(b.groups.len());
+    let mut stack = vec![0usize];
+    while let Some(g) = stack.pop() {
+        group_order.push(g);
+        let children: Vec<usize> = link_order[g]
+            .iter()
+            .filter_map(|&i| match b.groups[g].links[i].1 {
+                Target::Group(c) => Some(c),
+                _ => None,
+            })
+            .collect();
+        stack.extend(children.into_iter().rev());
+    }
+    let mut ds_order = Vec::with_capacity(b.datasets.len());
+    for &g in &group_order {
+        for &i in &link_order[g] {
+            if let Target::Dataset(d) = b.groups[g].links[i].1 {
+                ds_order.push(d);
+            }
+        }
+    }
+    let mut group_pos = vec![0usize; b.groups.len()];
+    for (pos, &g) in group_order.iter().enumerate() {
+        group_pos[g] = pos;
+    }
+    let mut ds_pos = vec![0usize; b.datasets.len()];
+    for (pos, &d) in ds_order.iter().enumerate() {
+        ds_pos[d] = pos;
+    }
+
+    let mut groups_by_id: Vec<Option<BuildGroup>> = b.groups.into_iter().map(Some).collect();
+    let mut groups = Vec::with_capacity(group_order.len());
+    for &g in &group_order {
+        let bg = groups_by_id[g].take().expect("each group is laid out once");
+        let mut targets: Vec<Option<(String, Target)>> = bg.links.into_iter().map(Some).collect();
+        let links = link_order[g]
+            .iter()
+            .map(|&i| {
+                let (name, t) = targets[i].take().expect("each link is written once");
+                let to = match (resolved[g][i], t) {
+                    (Some(Obj::Group(c)), _) => LinkTo::Group(group_pos[c]),
+                    (Some(Obj::Dataset(d)), _) => LinkTo::Dataset(ds_pos[d]),
+                    (None, Target::Soft(s)) => LinkTo::Soft(s),
+                    (None, Target::External { file, path }) => LinkTo::External { file, path },
+                    (None, _) => unreachable!("hard links are resolved"),
+                };
+                Link {
+                    name,
+                    to,
+                    creation_order: tracked[g].then_some(i as u64),
+                }
+            })
+            .collect();
+        groups.push(Group {
+            attrs: bg.attrs,
+            links,
+            track_order: tracked[g],
+            refcount: group_refs[g],
+        });
+    }
+    let mut ds_by_id: Vec<Option<DatasetBuilder>> = b.datasets.into_iter().map(Some).collect();
+    let datasets = ds_order
+        .iter()
+        .map(|&d| {
+            (
+                ds_by_id[d].take().expect("each dataset is laid out once"),
+                ds_refs[d],
+            )
+        })
+        .collect();
+    Ok(Tree { groups, datasets })
+}

crates/clawhdf5-format/tests/fixtures/filters/README.md

@@ -0,0 +1,13 @@
+# Filter conformance fixtures
+
+Files written by libhdf5 (and its registered filter plugins), used by the
+filter regression tests in `src/filters.rs` to compare our decoders against
+the values h5py/libhdf5 read from the same bytes. Chunk byte ranges quoted in
+the tests come from h5py's `DatasetID.get_chunk_info`.
+
+| File | Origin | Licence |
+|------|--------|---------|
+| `h5ex_d_lz4.h5` | HDF Group `HDF5Examples/C/H5FLT/tfiles/h5ex_d_lz4.h5` (hdf5 repository) | HDF5 licence (BSD-3-Clause style) |
+| `noencoder.h5` | HDF Group `test/testfiles/noencoder.h5` (hdf5 repository) | HDF5 licence (BSD-3-Clause style) |
+| `le_data.h5` | HDF Group `test/testfiles/le_data.h5` (hdf5 repository) | HDF5 licence (BSD-3-Clause style) |
+| `szip_h5py.h5` | Written for these tests with h5py 3 / libhdf5 2.0.0 (libaec szip): `f8` (8x10, chunks 4x10, `('nn', 8)`), `i8` (8x10, chunks 4x10, `('ec', 4)`), `u2` (70, chunks 35, `('nn', 8)`) | Same as this repository |

crates/clawhdf5-format/tests/fixtures/gen_shared_fill.py

@@ -0,0 +1,49 @@
+"""Generate shared_fill_value.h5: datasets whose Fill Value message is
+*shared*, in the two ways libhdf5 can share one.
+
+- /sohm_a, /sohm_b: the file has a shared-object-header-message (SOHM) index
+  for fill values, so libhdf5 stores the fill value (-7, int32) in the SOHM
+  heap and /sohm_b's header holds only a reference to it. Chunked, with only
+  the first chunk written, so the rest reads as the fill value.
+- /unwritten_a, /unwritten_b: the same, never written: no storage at all,
+  read entirely as the fill value.
+
+h5py has no API for SOHM indexes, so the file creation property list is
+configured by calling the libhdf5 bundled in the h5py wheel through ctypes.
+Written with h5py 3.16.0 / HDF5 2.0.0. Re-run only to regenerate:
+
+    python gen_shared_fill.py shared_fill_value.h5
+"""
+import ctypes
+import glob
+import os
+import sys
+
+import h5py
+import numpy as np
+
+libdir = os.path.join(os.path.dirname(os.path.dirname(h5py.__file__)), "h5py.libs")
+libs = [p for p in glob.glob(os.path.join(libdir, "libhdf5*.so*")) if "_hl" not in os.path.basename(p)]
+lib = ctypes.CDLL(libs[0])
+lib.H5open()
+
+H5O_SHMESG_FILL_FLAG = 1 << 0x0005
+
+fcpl = h5py.h5p.create(h5py.h5p.FILE_CREATE)
+lib.H5Pset_shared_mesg_nindexes.argtypes = [ctypes.c_int64, ctypes.c_uint]
+lib.H5Pset_shared_mesg_index.argtypes = [ctypes.c_int64, ctypes.c_uint, ctypes.c_uint, ctypes.c_uint]
+assert lib.H5Pset_shared_mesg_nindexes(fcpl.id, 1) >= 0
+assert lib.H5Pset_shared_mesg_index(fcpl.id, 0, H5O_SHMESG_FILL_FLAG, 0) >= 0
+
+fapl = h5py.h5p.create(h5py.h5p.FILE_ACCESS)
+fapl.set_libver_bounds(h5py.h5f.LIBVER_LATEST, h5py.h5f.LIBVER_LATEST)
+fid = h5py.h5f.create(sys.argv[1].encode(), h5py.h5f.ACC_TRUNC, fcpl=fcpl, fapl=fapl)
+with h5py.File(fid) as f:
+    # Chunked, with only the first chunk written: the rest reads as fill.
+    # libhdf5 keeps the first copy of a message in its own header; the second
+    # identical one (the `_b` datasets) is the SOHM reference.
+    for name in ("sohm_a", "sohm_b"):
+        d = f.create_dataset(name, shape=(8,), chunks=(4,), dtype="<i4", fillvalue=-7)
+        d[:4] = np.arange(4)
+    for name in ("unwritten_a", "unwritten_b"):
+        f.create_dataset(name, shape=(3,), dtype="<i4", fillvalue=-7)