.gitignore

@@ -7,3 +7,6 @@ weights/
 .venv
 __pycache__/
 .pytest_cache/
+
+# Scratch files the heavy tests generate (huge_chunks_interop)
+crates/*/tests/scratch/

crates/clawhdf5-format/src/chunk_cache.rs

@@ -899,7 +899,7 @@ mod tests {
 
     fn make_chunk(offsets: Vec<u64>, address: u64, size: u32) -> ChunkInfo {
         ChunkInfo {
-            chunk_size: size,
+            chunk_size: u64::from(size),
             filter_mask: 0,
             offsets,
             address,

crates/clawhdf5-format/src/chunk_index.rs

@@ -109,7 +109,7 @@ pub struct ChunkMapping {
     /// File byte address of the compressed chunk.
     pub file_offset: u64,
     /// Size of the compressed chunk in the file.
-    pub file_size: u32,
+    pub file_size: u64,
     /// Filter mask (0 = all filters applied).
     pub filter_mask: u32,
     /// Pre-computed row-copy operations for assembling this chunk into output.
@@ -338,7 +338,7 @@ mod tests {
 
     fn make_chunk(offsets: Vec<u64>, address: u64, size: u32) -> ChunkInfo {
         ChunkInfo {
-            chunk_size: size,
+            chunk_size: u64::from(size),
             filter_mask: 0,
             offsets,
             address,

crates/clawhdf5-format/src/chunked_read.rs

@@ -314,7 +314,7 @@ pub(crate) fn chunk_req(
     chunk_bytes: usize,
     wanted: bool,
 ) -> ExtentReq {
-    let len = c.chunk_size as usize;
+    let len = crate::addr::saturating_usize(c.chunk_size);
     ExtentReq {
         addr: c.address,
         len,
@@ -521,7 +521,7 @@ pub fn decompress_all_chunks_with_stats_in<S: Storage + ?Sized>(
 #[derive(Debug, Clone)]
 pub struct ChunkInfo {
     /// Size of chunk data in the file (after compression).
-    pub chunk_size: u32,
+    pub chunk_size: u64,
     /// Bitmask of filters that were NOT applied (0 = all applied).
     pub filter_mask: u32,
     /// N-dimensional offset of this chunk in dataset space.
@@ -610,7 +610,7 @@ fn stored_element_size(dt: &Datatype, offset_size: u8) -> u64 {
 /// (`H5D__chunk_set_sizes`: "stored datatype size in chunk layout does not
 /// match datatype description"). Reading it anyway laid the chunks out with
 /// the wrong element size.
-pub(crate) fn check_chunk_element_size(
+pub fn check_chunk_element_size(
     layout: &DataLayout,
     datatype: &Datatype,
     offset_size: u8,
@@ -1028,7 +1028,7 @@ fn parse_chunk_node<S: Storage + ?Sized>(
         if node_level == 0 {
             chunks.push(stored.len());
             stored.push(ChunkInfo {
-                chunk_size,
+                chunk_size: u64::from(chunk_size),
                 filter_mask,
                 offsets: keys[k..].to_vec(),
                 address,
@@ -1131,7 +1131,7 @@ pub fn generate_implicit_chunks_in_grid(
         }
 
         chunks.push(ChunkInfo {
-            chunk_size: chunk_byte_size as u32,
+            chunk_size: chunk_byte_size,
             filter_mask: 0,
             offsets,
             address: base_address.saturating_add(grid_idx.saturating_mul(chunk_byte_size)),
@@ -1190,9 +1190,15 @@ fn read_btree_v2_chunks<S: Storage + ?Sized>(
         }
         _ => return Err(bad("tree is not a chunk index")),
     };
-    let unfiltered_bytes = checked_chunk_byte_len(chunk_dims, elem_size)?;
-    let unfiltered_bytes =
-        u32::try_from(unfiltered_bytes).map_err(|_| bad("chunk larger than 4 GiB"))?;
+    // A u64 whatever the platform: an unfiltered chunk's size is only
+    // recorded here; a chunk this platform cannot address fails when read.
+    let unfiltered_bytes = u64::try_from(
+        chunk_dims
+            .iter()
+            .try_fold(elem_size as u128, |acc, &c| acc.checked_mul(c as u128))
+            .ok_or_else(|| bad("chunk size overflows"))?,
+    )
+    .map_err(|_| bad("chunk size overflows"))?;
 
     let records = collect_btree_v2_records_in(file_data, &header, offset_size, length_size)?;
     let mut chunks = Vec::with_capacity(records.len());
@@ -1213,10 +1219,7 @@ fn read_btree_v2_chunks<S: Storage + ?Sized>(
             pos += size_len;
             let mask = u32::from_le_bytes([data[pos], data[pos + 1], data[pos + 2], data[pos + 3]]);
             pos += 4;
-            (
-                u32::try_from(size).map_err(|_| bad("stored chunk larger than 4 GiB"))?,
-                mask,
-            )
+            (size, mask)
         };
         let mut offsets = Vec::with_capacity(rank);
         for &dim in chunk_dims {
@@ -1334,9 +1337,9 @@ pub fn list_chunks_in<S: Storage + ?Sized>(
             // Single chunk — one chunk covering the entire dataset
             let chunk_byte_size = checked_chunk_byte_len(&chunk_dims, elem_size)?;
             let (csize, fmask) = if let Some(fs) = single_filtered_size {
-                (fs as u32, single_filter_mask.unwrap_or(0))
+                (fs, single_filter_mask.unwrap_or(0))
             } else {
-                (chunk_byte_size as u32, 0)
+                (chunk_byte_size as u64, 0)
             };
             vec![ChunkInfo {
                 chunk_size: csize,
@@ -2124,7 +2127,7 @@ pub fn read_chunked_data_indexed_in<S: Storage + ?Sized>(
         .iter()
         .zip(&hits)
         .map(|(m, hit)| {
-            let len = m.file_size as usize;
+            let len = crate::addr::saturating_usize(m.file_size);
             ExtentReq {
                 addr: m.file_offset,
                 len,
@@ -2757,7 +2760,7 @@ mod tests {
             }
 
             chunk_infos.push(ChunkInfo {
-                chunk_size: chunk_bytes as u32,
+                chunk_size: chunk_bytes as u64,
                 filter_mask: 0,
                 offsets: vec![start as u64, 0],
                 address: data_offset as u64,
@@ -2869,7 +2872,7 @@ mod tests {
                     .collect();
                 let stored = crate::filters::compress_chunk(&chunk, &pipeline, 4).unwrap();
                 chunks.push(ChunkInfo {
-                    chunk_size: stored.len() as u32,
+                    chunk_size: stored.len() as u64,
                     filter_mask: 0,
                     offsets: vec![r0 as u64, c0 as u64, 0],
                     address: file.len() as u64,
@@ -2943,7 +2946,7 @@ mod tests {
         let short = crate::filters::compress_chunk(&[1u8; 64], &pipeline, 4).unwrap();
         for bad in [5usize, 11, 40] {
             chunks[bad].address = file.len() as u64;
-            chunks[bad].chunk_size = short.len() as u32;
+            chunks[bad].chunk_size = short.len() as u64;
             file.extend_from_slice(&short);
         }
         for _ in 0..20 {
@@ -3133,7 +3136,7 @@ mod tests {
             file_data[data_offset..data_offset + compressed.len()].copy_from_slice(&compressed);
 
             chunk_infos.push(ChunkInfo {
-                chunk_size: compressed.len() as u32,
+                chunk_size: compressed.len() as u64,
                 filter_mask: 0,
                 offsets: vec![start as u64, 0],
                 address: data_offset as u64,
@@ -3215,7 +3218,7 @@ mod tests {
                 file_data[data_offset..data_offset + chunk_size].copy_from_slice(&chunk_bytes);
 
                 chunk_infos.push(ChunkInfo {
-                    chunk_size: chunk_size as u32,
+                    chunk_size: chunk_size as u64,
                     filter_mask: 0,
                     offsets: vec![row_start as u64, col_start as u64, 0],
                     address: data_offset as u64,
@@ -3340,7 +3343,7 @@ mod tests {
             assert_eq!(c.address, 0x1000 + i as u64 * chunk_byte_size as u64);
             assert_eq!(c.offsets, vec![i as u64 * 20]);
             assert_eq!(c.filter_mask, 0);
-            assert_eq!(c.chunk_size, chunk_byte_size as u32);
+            assert_eq!(c.chunk_size, chunk_byte_size as u64);
         }
     }
 

crates/clawhdf5-format/src/chunked_write.rs

@@ -2125,7 +2125,7 @@ mod tests {
             for info in &infos {
                 // Skipped chunks are stored at the chunk's size (shuffled).
                 assert_eq!(
-                    info.chunk_size == (c * 8) as u32,
+                    info.chunk_size == (c * 8) as u64,
                     info.filter_mask != 0,
                     "{info:?}"
                 );

crates/clawhdf5-format/src/extensible_array.rs

@@ -224,7 +224,7 @@ fn read_element(
         };
         Ok((
             Some(ChunkInfo {
-                chunk_size: chunk_byte_size as u32,
+                chunk_size: chunk_byte_size,
                 filter_mask: 0,
                 offsets,
                 address,
@@ -259,7 +259,7 @@ fn read_element(
         };
         Ok((
             Some(ChunkInfo {
-                chunk_size: chunk_size as u32,
+                chunk_size,
                 filter_mask,
                 offsets,
                 address,
@@ -946,7 +946,7 @@ mod tests {
         assert_eq!(chunks.len(), 2);
         assert_eq!(chunks[0].address, base_addr);
         assert_eq!(chunks[0].offsets, vec![0]);
-        assert_eq!(chunks[0].chunk_size, chunk_byte_size as u32);
+        assert_eq!(chunks[0].chunk_size, chunk_byte_size as u64);
         assert_eq!(chunks[1].address, base_addr + chunk_byte_size);
         assert_eq!(chunks[1].offsets, vec![20]);
     }

crates/clawhdf5-format/src/filters.rs

@@ -340,10 +340,23 @@ pub fn decompress_chunk_exact_with<'s>(
                 } else {
                     MAX_DECOMPRESS_SIZE
                 };
-                let size_hint = if ctx.max_output != 0 {
+                // The bound is the output's size when only size-preserving
+                // filters (shuffle, Fletcher32) remain to be undone; before
+                // any other filter (a second deflate, N-Bit, ...) it is only
+                // a ceiling, and the output starts smaller and grows: the
+                // inflater writes every byte it reserves, so reserving a
+                // 4 GiB chunk's bound for a stage a few MiB long would hold
+                // twice the chunk's memory.
+                let exact = pipeline.filters[..i].iter().enumerate().all(|(j, f)| {
+                    filter_skipped(filter_mask, j)
+                        || matches!(f.filter_id, FILTER_SHUFFLE | FILTER_FLETCHER32)
+                });
+                let size_hint = if ctx.max_output == 0 {
+                    input.len().saturating_mul(4).min(1 << 20)
+                } else if exact {
                     ctx.max_output
                 } else {
-                    input.len().saturating_mul(4).min(1 << 20)
+                    input.len().saturating_mul(4).min(ctx.max_output)
                 };
                 let inflater = scratch
                     .inflater

crates/clawhdf5-format/src/fixed_array.rs

@@ -383,7 +383,7 @@ fn parse_fa_element(
     offset_size: u8,
     element_size: u8,
     chunk_byte_size: u64,
-) -> Result<Option<(u64, u32, u32)>, FormatError> {
+) -> Result<Option<(u64, u64, u32)>, FormatError> {
     let os = offset_size as usize;
     if client_id == 0 {
         // Non-filtered: element is just the chunk address.
@@ -393,7 +393,7 @@ fn parse_fa_element(
             return Ok(None);
         }
         let address = read_offset(file_data, abs, offset_size)?;
-        Ok(Some((address, chunk_byte_size as u32, 0)))
+        Ok(Some((address, chunk_byte_size, 0)))
     } else {
         // Filtered: address(offset_size) + chunk_size(variable) + filter_mask(4)
         let es = element_size as usize;
@@ -418,7 +418,7 @@ fn parse_fa_element(
             file_data[fm_off + 2],
             file_data[fm_off + 3],
         ]);
-        Ok(Some((address, chunk_size as u32, filter_mask)))
+        Ok(Some((address, chunk_size, filter_mask)))
     }
 }
 
@@ -688,7 +688,7 @@ mod tests {
             assert_eq!(c.address, base_addr + i as u64 * chunk_byte_size as u64);
             assert_eq!(c.offsets, vec![i as u64 * 20]);
             assert_eq!(c.filter_mask, 0);
-            assert_eq!(c.chunk_size, chunk_byte_size as u32);
+            assert_eq!(c.chunk_size, chunk_byte_size as u64);
         }
     }
 

crates/clawhdf5-format/src/parallel_read.rs

@@ -426,7 +426,7 @@ mod tests {
         for i in 0..8u64 {
             let len = if short && i == 5 { 16 } else { 32 };
             infos.push(ChunkInfo {
-                chunk_size: len as u32,
+                chunk_size: len as u64,
                 filter_mask: 0,
                 offsets: vec![i * 8],
                 address: file.len() as u64,

crates/clawhdf5-format/src/partial_read.rs

@@ -243,6 +243,58 @@ fn copy_overlap(
     }
 }
 
+/// Copy the part of unfiltered chunk `chunk` (`chunk_bytes` long, shape
+/// `chunk_shape`) that overlaps the box into `out`, reading only the runs of
+/// the overlap from the file. The whole chunk must still lie inside the
+/// file, as it must when it is fetched whole.
+#[allow(clippy::too_many_arguments)]
+fn read_unfiltered_overlap<S: Storage + ?Sized>(
+    file_data: &S,
+    chunk: &crate::chunked_read::ChunkInfo,
+    chunk_bytes: usize,
+    chunk_shape: &[u64],
+    elem_size: usize,
+    out: &mut [u8],
+    box_start: &[u64],
+    box_extent: &[u64],
+) -> Result<(), FormatError> {
+    let rank = chunk_shape.len();
+    let origin = &chunk.offsets[..rank];
+    let (lo, extent): (Vec<u64>, Vec<u64>) = (0..rank)
+        .map(|d| {
+            let lo = origin[d].max(box_start[d]);
+            let hi = origin[d]
+                .saturating_add(chunk_shape[d])
+                .min(box_start[d] + box_extent[d]);
+            (lo, hi.saturating_sub(lo))
+        })
+        .unzip();
+    let file_len = crate::storage::len_usize(file_data);
+    let base = crate::addr::to_usize(chunk.address)?;
+    if base > file_len || chunk_bytes > file_len - base {
+        return Err(FormatError::UnexpectedEof {
+            expected: base.saturating_add(chunk_bytes),
+            available: file_len,
+        });
+    }
+    let overlap = Selection::Hyperslab {
+        start: lo.iter().zip(origin).map(|(l, o)| l - o).collect(),
+        stride: vec![1; rank],
+        count: extent.clone(),
+        block: vec![1; rank],
+    };
+    let rows = crate::gather::gather_storage(
+        file_data,
+        chunk.address,
+        chunk_bytes,
+        chunk_shape,
+        elem_size,
+        &overlap,
+    )?;
+    copy_overlap(&rows, &lo, &extent, out, box_start, box_extent, elem_size);
+    Ok(())
+}
+
 /// Read `selection` without materialising the whole dataset, when that is
 /// possible and worthwhile. `Ok(None)` means "use the full-read path": an
 /// `All`/`None`/invalid selection, a layout this doesn't handle (compact,
@@ -281,6 +333,39 @@ pub fn read_selection_in<S: Storage + ?Sized>(
     offset_size: u8,
     length_size: u8,
     selection: &Selection,
+) -> Result<Option<Vec<u8>>, FormatError> {
+    read_selection_filled_in(
+        file_data,
+        layout,
+        dataspace,
+        elem_size,
+        pipeline,
+        offset_size,
+        length_size,
+        selection,
+        None,
+    )
+}
+
+/// [`read_selection_in`] for a dataset whose fill value is `fill` (one
+/// element's bytes; `None` or all zeros is the default fill): the elements
+/// of a chunked dataset's selection that lie in chunks never written read as
+/// `fill`, as they do in a full read. Only the chunks the selection's
+/// bounding box overlaps are read, so a selection of a few elements of a
+/// dataset whose chunks are 4 GiB or more costs one decoded chunk (a
+/// filtered chunk has to be decoded whole) or, unfiltered, only the bytes
+/// it selects.
+#[allow(clippy::too_many_arguments)]
+pub fn read_selection_filled_in<S: Storage + ?Sized>(
+    file_data: &S,
+    layout: &DataLayout,
+    dataspace: &Dataspace,
+    elem_size: usize,
+    pipeline: Option<&FilterPipeline>,
+    offset_size: u8,
+    length_size: u8,
+    selection: &Selection,
+    fill: Option<&[u8]>,
 ) -> Result<Option<Vec<u8>>, FormatError> {
     let dims = &dataspace.dimensions;
     if dims.is_empty() || elem_size == 0 {
@@ -341,8 +426,18 @@ pub fn read_selection_in<S: Storage + ?Sized>(
         return Ok(None);
     }
     let mut boxed = alloc_output(checked_byte_len(box_elements, elem_size)?)?;
+    if let Some(fill) = fill.filter(|f| <[u8]>::len(f) == elem_size && f.iter().any(|&b| b != 0)) {
+        for element in boxed.chunks_exact_mut(elem_size) {
+            element.copy_from_slice(fill);
+        }
+    }
 
     match layout {
+        // No chunk was ever written: every element is the fill value.
+        DataLayout::Chunked {
+            btree_address: None,
+            ..
+        } if fill.is_some() => {}
         DataLayout::Chunked {
             btree_address: Some(_),
             ..
@@ -373,6 +468,25 @@ pub fn read_selection_in<S: Storage + ?Sized>(
                     })
                 })
                 .collect();
+            // An unfiltered chunk of a file that is not in memory: fetch
+            // only the rows the box needs, not the whole chunk (which may be
+            // 4 GiB or more).
+            let (direct, wanted): (Vec<_>, Vec<_>) = wanted.into_iter().partition(|c| {
+                file_data.as_contiguous().is_none()
+                    && pipeline.is_none_or(|pl| all_filters_skipped(pl, c.filter_mask))
+            });
+            for chunk in direct {
+                read_unfiltered_overlap(
+                    file_data,
+                    chunk,
+                    chunk_bytes,
+                    &chunk_shape,
+                    elem_size,
+                    &mut boxed,
+                    &box_start,
+                    &box_extent,
+                )?;
+            }
             // Their stored bytes, batch by batch when the file is not in
             // memory; each batch's chunks are decoded into this thread's
             // reusable buffers before the next batch is fetched.

crates/clawhdf5-format/tests/raw_fetch_bounds.rs

@@ -173,7 +173,7 @@ fn crafted() -> (Vec<u8>, Chunked, Vec<ChunkInfo>) {
     assert!(
         chunks
             .iter()
-            .all(|c| c.chunk_size == HUGE && c.address == blob)
+            .all(|c| c.chunk_size == u64::from(HUGE) && c.address == blob)
     );
     (bytes, ds, chunks)
 }
@@ -313,7 +313,7 @@ fn large_reads_are_fetched_in_batches() {
     let data: Vec<u8> = (0..2 * CHUNK).map(|i| (i % 251) as u8).collect();
     let chunks: Vec<ChunkInfo> = (0..40u64)
         .map(|i| ChunkInfo {
-            chunk_size: CHUNK as u32,
+            chunk_size: CHUNK as u64,
             filter_mask: 0,
             offsets: vec![i * CHUNK as u64],
             address: (i % 2) * CHUNK as u64,

crates/clawhdf5/src/reader.rs

@@ -1528,11 +1528,11 @@ impl<'f> Dataset<'f> {
         let ds = self.dataspace()?;
         let dl = self.data_layout()?;
         let pipeline = self.filter_pipeline()?;
-        // The selection reader knows nothing about fill values. When they
-        // matter — no storage at all, or a non-zero fill on a chunked (possibly
-        // sparse) dataset — select from a fill-aware full read instead. (The
-        // selection reader currently decodes the full dataset too, so this
-        // costs nothing extra.)
+        // Fill values matter when there is no storage at all, or a non-zero
+        // fill on a chunked (possibly sparse) dataset: a chunked dataset's
+        // selection is then read over a box of fill values; anything else
+        // (and a selection whose box is most of the dataset) is selected
+        // from a fill-aware full read.
         let fill = clawhdf5_format::fill_value::dataset_fill_value_from_storage(
             &self.file.data,
             &self.header.messages,
@@ -1545,6 +1545,29 @@ impl<'f> Dataset<'f> {
                 && !clawhdf5_format::fill_value::is_default(fill.as_deref()));
         if fill_matters {
             clawhdf5_format::partial_read::validate(selection, &ds.dimensions)?;
+            // A chunked dataset: read the chunks the selection touches over
+            // a box of fill values, not the whole dataset (which may be many
+            // GiB when its chunks are large).
+            if matches!(dl, DataLayout::Chunked { .. }) {
+                clawhdf5_format::chunked_read::check_chunk_element_size(
+                    &dl,
+                    &dt,
+                    self.file.offset_size(),
+                )?;
+                if let Some(selected) = clawhdf5_format::partial_read::read_selection_filled_in(
+                    &self.file.data,
+                    &dl,
+                    &ds,
+                    dt.type_size() as usize,
+                    pipeline.as_ref(),
+                    self.file.offset_size(),
+                    self.file.length_size(),
+                    selection,
+                    Some(fill.as_deref().unwrap_or(&[])),
+                )? {
+                    return Ok(selected);
+                }
+            }
             let full = self.read_raw()?;
             return Ok(data_read::extract_selection_from_buffer(
                 &full,

crates/clawhdf5/tests/fixtures/gen_huge_chunks.py

@@ -0,0 +1,109 @@
+"""Write HDF5 files whose chunks are 4 GiB or more, one dataset per chunk index.
+
+    python gen_huge_chunks.py filtered OUT.h5     # huge_chunks_filtered.h5
+    python gen_huge_chunks.py unfiltered OUT.h5   # generated at test time
+
+libhdf5 2.x writes a chunk of more than 0xFFFFFFFF bytes with layout message
+version 5 (`H5D__chunk_construct`: "chunk size > 4GB requires
+H5F_LIBVER_V200"), so never with a version-1 B-tree; a filtered chunk index
+element of a version-5 layout stores the chunk's size in "size of lengths"
+bytes (8). Every dataset is `<f8` with chunks of N = 2**29 + 1 elements
+(4 GiB + 8 bytes):
+
+    single     shape (N,), chunks (N,)                       Single Chunk
+    implicit   shape (2N,), early allocation (unfiltered)    Implicit
+    farray     shape (N+10,)                                 Fixed Array
+    earray     shape (N+10,), maxshape (None,)               Extensible Array
+    btree2     shape (2, N+10), chunks (1, N),
+               maxshape (None, None)                         v2 B-tree
+
+Only a few elements are written: `d[0:10] = 0..9` and, where there is a
+second chunk along the axis, `d[N:N+10] = 100..109` (btree2: row 0 as that,
+row 1 `200..209` and `300..309`; implicit: `d[N:N+10]` and
+`d[2N-10:2N] = 500..509`). Everything else reads as the fill value.
+
+`filtered` (the committed fixture, no `implicit`: that index is never
+filtered): deflate level 9 twice in the pipeline, fill value -1.0. One
+deflate leaves a 4 GiB chunk of a repeated 8-byte pattern at about 6 MiB;
+the second pass takes that to about 14 KiB, so the file is small while each
+chunk still inflates to 4 GiB + 8 bytes.
+
+`unfiltered`: fill time "never" and the default fill value, so libhdf5
+writes only the elements written (an unfiltered chunk larger than the chunk
+cache is written in place) and the file is sparse: tens of GiB long but a few
+blocks on disk. Unwritten elements read as whatever the file holds there:
+zeros. Do not put it on tmpfs, which is memory.
+
+libhdf5 holds a whole filtered chunk in memory while it writes it, so the
+`filtered` run needs about 4.5 GiB of memory (and about 3 minutes).
+
+Generated 2026-09-28 with h5py 3.16.0 (libhdf5 2.0.0).
+"""
+import sys
+
+import h5py
+import numpy as np
+
+N = 2**29 + 1
+UNLIM = h5py.h5s.UNLIMITED
+
+
+def make(fid, name, index, filtered):
+    dcpl = h5py.h5p.create(h5py.h5p.DATASET_CREATE)
+    if index == "single":
+        shape, maxshape, chunk = (N,), (N,), (N,)
+    elif index == "implicit":
+        shape, maxshape, chunk = (2 * N,), (2 * N,), (N,)
+        dcpl.set_alloc_time(h5py.h5d.ALLOC_TIME_EARLY)
+    elif index == "farray":
+        shape, maxshape, chunk = (N + 10,), (N + 10,), (N,)
+    elif index == "earray":
+        shape, maxshape, chunk = (N + 10,), (UNLIM,), (N,)
+    elif index == "btree2":
+        shape, maxshape, chunk = (2, N + 10), (UNLIM, UNLIM), (1, N)
+    dcpl.set_chunk(chunk)
+    if filtered:
+        dcpl.set_deflate(9)
+        dcpl.set_deflate(9)
+        dcpl.set_fill_value(np.array(-1.0, dtype="<f8"))
+    else:
+        dcpl.set_fill_time(h5py.h5d.FILL_TIME_NEVER)
+        if index == "single":
+            # libhdf5 2.0.0 drops a write to an unallocated unfiltered
+            # Single Chunk this large when the fill time is "never" (the
+            # dataset stays unallocated and reads zeros); allocated at
+            # creation, the write lands in place.
+            dcpl.set_alloc_time(h5py.h5d.ALLOC_TIME_EARLY)
+    space = h5py.h5s.create_simple(shape, maxshape)
+    dsid = h5py.h5d.create(fid, name.encode(), h5py.h5t.IEEE_F64LE, space, dcpl=dcpl)
+    ds = h5py.Dataset(dsid)
+    if len(shape) == 2:
+        ds[0, 0:10] = np.arange(10.0)
+        ds[0, N : N + 10] = np.arange(100.0, 110.0)
+        ds[1, 0:10] = np.arange(200.0, 210.0)
+        ds[1, N : N + 10] = np.arange(300.0, 310.0)
+    else:
+        ds[0:10] = np.arange(10.0)
+        if shape[0] > N:
+            ds[N : N + 10] = np.arange(100.0, 110.0)
+        if index == "implicit":
+            ds[2 * N - 10 : 2 * N] = np.arange(500.0, 510.0)
+    dsid.close()
+
+
+def main():
+    mode, out = sys.argv[1], sys.argv[2]
+    filtered = {"filtered": True, "unfiltered": False}[mode]
+    fapl = h5py.h5p.create(h5py.h5p.FILE_ACCESS)
+    fapl.set_libver_bounds(h5py.h5f.LIBVER_EARLIEST, h5py.h5f.LIBVER_V200)
+    fid = h5py.h5f.create(out.encode(), h5py.h5f.ACC_TRUNC, fapl=fapl)
+    indexes = ["single", "farray", "earray", "btree2"]
+    if not filtered:
+        indexes.insert(1, "implicit")
+    for index in indexes:
+        make(fid, index, index, filtered)
+    fid.close()
+
+
+if __name__ == "__main__":
+    main()

crates/clawhdf5/tests/huge_chunks_interop.rs

@@ -0,0 +1,296 @@
+//! Chunks of 4 GiB or more, which HDF5 2.0 writes (layout message version 5,
+//! `H5F_LIBVER_V200`), in every chunk index libhdf5 uses for them.
+//!
+//! `fixtures/huge_chunks_filtered.h5` (written by libhdf5 2.0.0 through h5py
+//! 3.16, `fixtures/gen_huge_chunks.py filtered`) holds one dataset per
+//! filtered index — Single Chunk, Fixed Array, Extensible Array, v2 B-tree —
+//! whose chunks are 2^29 + 1 `f64` (4 GiB + 8 bytes), stored deflated twice
+//! so each takes about 20 KiB. Listing their chunks is cheap and always runs;
+//! decoding one inflates 4 GiB, so those tests are opt-in:
+//! `CLAWHDF5_HUGE_CHUNKS=1` (run them one at a time, `--test-threads=1`:
+//! each needs about 4.5 GiB of memory). The same variable enables the
+//! unfiltered tests, which have h5py write a sparse file (about 44 GiB long,
+//! a few blocks on disk) under `tests/scratch/`, and the writer tests.
+//!
+//! libhdf5 never writes such a chunk with a version-1 B-tree (a chunk of more
+//! than 0xFFFFFFFF bytes forces layout version 5 and with it the newer
+//! indexes) and refuses to open one; `chunked_read`'s unit tests cover that.
+
+use std::path::{Path, PathBuf};
+use std::process::Command;
+
+use clawhdf5::File;
+use clawhdf5_format::chunked_read::list_chunks;
+use clawhdf5_format::data_layout::DataLayout;
+use clawhdf5_format::dataspace::Dataspace;
+use clawhdf5_format::message_type::MessageType;
+use clawhdf5_format::object_header::ObjectHeader;
+use clawhdf5_format::selection::Selection;
+use clawhdf5_format::superblock::Superblock;
+
+/// Elements per chunk along the chunked axis: 4 GiB + 8 bytes of `f64`.
+const N: u64 = (1 << 29) + 1;
+
+fn fixture() -> PathBuf {
+    Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/fixtures/huge_chunks_filtered.h5")
+}
+
+fn heavy() -> bool {
+    if std::env::var("CLAWHDF5_HUGE_CHUNKS").is_ok_and(|v| v == "1") {
+        return true;
+    }
+    eprintln!("SKIP: set CLAWHDF5_HUGE_CHUNKS=1 to decode 4 GiB chunks");
+    false
+}
+
+fn python() -> String {
+    std::env::var("CLAWHDF5_PYTHON").unwrap_or_else(|_| "python3".to_string())
+}
+
+fn python_available() -> bool {
+    Command::new(python())
+        .args(["-c", "import h5py"])
+        .output()
+        .is_ok_and(|o| o.status.success())
+}
+
+fn interop_required() -> bool {
+    std::env::var("CLAWHDF5_REQUIRE_INTEROP").is_ok_and(|v| v == "1")
+}
+
+/// The raw layout message version, the parsed layout, the dataspace and the
+/// chunks of `name` in the in-memory file `data`.
+fn layout_of(
+    data: &[u8],
+    name: &str,
+) -> (
+    u8,
+    DataLayout,
+    Dataspace,
+    Vec<clawhdf5_format::chunked_read::ChunkInfo>,
+) {
+    let sb = Superblock::parse(data, 0).unwrap();
+    let addr = clawhdf5_format::group_v2::resolve_path_any(data, &sb, name).unwrap();
+    let hdr = ObjectHeader::parse(data, addr as usize, sb.offset_size, sb.length_size).unwrap();
+    let msg = |t| {
+        hdr.messages
+            .iter()
+            .find(|m| m.msg_type == t)
+            .unwrap_or_else(|| panic!("{name}: no {t:?} message"))
+    };
+    let lm = msg(MessageType::DataLayout);
+    let layout = DataLayout::parse(&lm.data, sb.offset_size, sb.length_size).unwrap();
+    let space = Dataspace::parse(&msg(MessageType::Dataspace).data, sb.length_size).unwrap();
+    let (chunks, _) =
+        list_chunks(data, &layout, &space, 8, sb.offset_size, sb.length_size).unwrap();
+    (lm.data[0], layout, space, chunks)
+}
+
+/// The fixture's datasets: name, chunk index type, shape, and the scaled
+/// origins of the chunks libhdf5 wrote.
+const FILTERED: &[(&str, u8, &[u64], &[&[u64]])] = &[
+    ("single", 1, &[N], &[&[0]]),
+    ("farray", 3, &[N + 10], &[&[0], &[N]]),
+    ("earray", 4, &[N + 10], &[&[0], &[N]]),
+    (
+        "btree2",
+        5,
+        &[2, N + 10],
+        &[&[0, 0], &[0, N], &[1, 0], &[1, N]],
+    ),
+];
+
+/// Every index of the fixture: layout version 5, its chunks listed at the
+/// right origins with their stored (deflated) sizes. The index elements
+/// store a size in 8 bytes (libhdf5's `H5F_SIZEOF_SIZE` under layout
+/// version 5), where version 4 would use 6 for a chunk this size.
+#[test]
+fn filtered_huge_chunk_indexes_list() {
+    let data = std::fs::read(fixture()).unwrap();
+    for &(name, index, shape, origins) in FILTERED {
+        let (version, layout, space, mut chunks) = layout_of(&data, name);
+        assert_eq!(version, 5, "{name}: layout message version");
+        let DataLayout::Chunked {
+            chunk_dimensions,
+            chunk_index_type,
+            ..
+        } = &layout
+        else {
+            panic!("{name}: not chunked: {layout:?}");
+        };
+        assert_eq!(*chunk_index_type, Some(index), "{name}");
+        assert_eq!(chunk_dimensions.last(), Some(&8), "{name}");
+        assert_eq!(space.dimensions, shape, "{name}");
+        chunks.sort_by(|a, b| a.offsets.cmp(&b.offsets));
+        let got: Vec<&[u64]> = chunks.iter().map(|c| &c.offsets[..shape.len()]).collect();
+        assert_eq!(got, origins, "{name}");
+        for c in &chunks {
+            assert!(
+                (10_000..40_000).contains(&c.chunk_size),
+                "{name}: stored size {} of chunk {:?}",
+                c.chunk_size,
+                c.offsets
+            );
+            assert_eq!(c.filter_mask, 0);
+        }
+    }
+}
+
+/// `f64` values of `sel` in dataset `name` of `file`.
+fn sel(file: &File, name: &str, start: &[u64], count: &[u64]) -> Vec<f64> {
+    let ds = file.dataset(name).unwrap();
+    let s = Selection::Hyperslab {
+        start: start.to_vec(),
+        stride: vec![1; start.len()],
+        count: count.to_vec(),
+        block: vec![1; start.len()],
+    };
+    ds.read_f64_selection(&s).unwrap()
+}
+
+fn f(range: std::ops::Range<i32>) -> Vec<f64> {
+    range.map(f64::from).collect()
+}
+
+/// Reads that touch a few elements of each 4 GiB chunk: written values, the
+/// fill value (-1) next to them, and the edge of the dataset.
+#[test]
+fn filtered_huge_chunks_read() {
+    if !heavy() {
+        return;
+    }
+    let file = File::open(fixture()).unwrap();
+    let mut first = f(0..10);
+    first.extend([-1.0; 2]);
+    for name in ["single", "farray", "earray"] {
+        assert_eq!(sel(&file, name, &[0], &[12]), first, "{name}");
+    }
+    assert_eq!(sel(&file, "single", &[N - 2], &[2]), [-1.0; 2]);
+    let mut edge = vec![-1.0; 2];
+    edge.extend(f(100..110));
+    for name in ["farray", "earray"] {
+        assert_eq!(sel(&file, name, &[N - 2], &[12]), edge, "{name}");
+    }
+    assert_eq!(sel(&file, "btree2", &[0, 0], &[1, 10]), f(0..10));
+    assert_eq!(sel(&file, "btree2", &[1, N], &[1, 10]), f(300..310));
+    assert_eq!(
+        sel(&file, "btree2", &[0, N - 1], &[2, 2]),
+        [-1.0, 100.0, -1.0, 300.0]
+    );
+}
+
+/// A directory under `tests/scratch/` (on disk: the sparse files must not
+/// land on a tmpfs `/tmp`), removed when dropped.
+fn scratch() -> tempfile::TempDir {
+    let root = Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/scratch");
+    std::fs::create_dir_all(&root).unwrap();
+    tempfile::tempdir_in(root).unwrap()
+}
+
+/// Have h5py (libhdf5 2.x) write `fixtures/gen_huge_chunks.py`'s file for
+/// `mode` into `dir`; `None` when there is no h5py (a failure under
+/// `CLAWHDF5_REQUIRE_INTEROP=1`).
+fn generate(dir: &Path, mode: &str) -> Option<PathBuf> {
+    if !python_available() {
+        assert!(
+            !interop_required(),
+            "CLAWHDF5_REQUIRE_INTEROP=1 but python3 with h5py is not available"
+        );
+        eprintln!("SKIP: python3 with h5py not available");
+        return None;
+    }
+    let script = Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/fixtures/gen_huge_chunks.py");
+    let path = dir.join(format!("{mode}.h5"));
+    let out = Command::new(python())
+        .arg(&script)
+        .arg(mode)
+        .arg(&path)
+        .output()
+        .unwrap();
+    assert!(
+        out.status.success(),
+        "gen_huge_chunks.py {mode} failed:\n{}",
+        String::from_utf8_lossy(&out.stderr)
+    );
+    Some(path)
+}
+
+/// Positioned reads of a file (no mmap), counting the bytes read.
+struct Counting {
+    file: std::fs::File,
+    len: u64,
+    read: std::sync::atomic::AtomicU64,
+}
+
+impl clawhdf5_format::storage::Storage for Counting {
+    fn read_at(
+        &self,
+        offset: u64,
+        len: usize,
+    ) -> Result<std::borrow::Cow<'_, [u8]>, clawhdf5_format::error::FormatError> {
+        use std::os::unix::fs::FileExt;
+        let len = len.min(usize::try_from(self.len.saturating_sub(offset)).unwrap_or(usize::MAX));
+        let mut buf = vec![0; len];
+        self.file.read_exact_at(&mut buf, offset).unwrap();
+        self.read
+            .fetch_add(len as u64, std::sync::atomic::Ordering::Relaxed);
+        Ok(buf.into())
+    }
+
+    fn len(&self) -> u64 {
+        self.len
+    }
+}
+
+fn check_unfiltered(file: &File) {
+    let mut first = f(0..10);
+    first.extend([0.0; 2]);
+    for name in ["single", "implicit", "farray", "earray"] {
+        assert_eq!(sel(file, name, &[0], &[12]), first, "{name}");
+    }
+    assert_eq!(sel(file, "single", &[N - 2], &[2]), [0.0; 2]);
+    let mut edge = vec![0.0; 2];
+    edge.extend(f(100..110));
+    for name in ["implicit", "farray", "earray"] {
+        assert_eq!(sel(file, name, &[N - 2], &[12]), edge, "{name}");
+    }
+    let mut last = vec![0.0; 2];
+    last.extend(f(500..510));
+    assert_eq!(sel(file, "implicit", &[2 * N - 12], &[12]), last);
+    assert_eq!(sel(file, "btree2", &[0, 0], &[1, 10]), f(0..10));
+    assert_eq!(sel(file, "btree2", &[1, N], &[1, 10]), f(300..310));
+    assert_eq!(
+        sel(file, "btree2", &[0, N - 1], &[2, 2]),
+        [0.0, 100.0, 0.0, 300.0]
+    );
+}
+
+/// Unfiltered chunks of 4 GiB + 8 bytes in every index libhdf5 gives them
+/// (Single Chunk, Implicit, Fixed Array, Extensible Array, v2 B-tree), in a
+/// sparse file h5py writes: read through a memory map, and through
+/// positioned reads, where a selection reads only the rows it needs.
+#[test]
+fn unfiltered_huge_chunks_read() {
+    if !heavy() {
+        return;
+    }
+    let dir = scratch();
+    let Some(path) = generate(dir.path(), "unfiltered") else {
+        return;
+    };
+    check_unfiltered(&File::open(&path).unwrap());
+
+    let file = std::fs::File::open(&path).unwrap();
+    let len = file.metadata().unwrap().len();
+    assert!(len > 40 << 30, "{len}");
+    let storage = std::sync::Arc::new(Counting {
+        file,
+        len,
+        read: 0.into(),
+    });
+    let positioned = File::open_storage(storage.clone()).unwrap();
+    check_unfiltered(&positioned);
+    // Every read above together: metadata and a few rows, not 4 GiB chunks.
+    let read = storage.read.load(std::sync::atomic::Ordering::Relaxed);
+    assert!(read < 1 << 20, "{read} bytes read");
+}

crates/clawhdf5/tests/parallel_integration.rs

@@ -283,7 +283,7 @@ mod parallel_tests {
         for (i, chunk) in compressed_chunks.iter().enumerate() {
             file_data[offset..offset + chunk.len()].copy_from_slice(chunk);
             chunk_infos.push(ChunkInfo {
-                chunk_size: chunk.len() as u32,
+                chunk_size: chunk.len() as u64,
                 filter_mask: 0,
                 offsets: vec![(i * chunk_elems) as u64, 0],
                 address: offset as u64,