fix: read multi-direct-block fractal heaps (root indirect block)

Browse Source

The fractal-heap reader split direct vs indirect block rows using the FRHP
"Starting # of Rows in Root Indirect Block" field (a constant, typically 1),
mislabeled as starting_row_of_indirect_blocks. For any heap whose data spans
more than one direct block — common in libhdf5 files with a large group or
many dense attributes — this treated direct blocks as indirect and walked into
garbage, failing with InvalidFractalHeapSignature.

Derive the split from the heap geometry instead: max_direct_rows =
log2(max_direct_block_size / starting_block_size) + 2. Rows below it hold
direct blocks; rows at/above hold child indirect blocks.

Validated against an h5py-written group with 400 dense attributes (root
indirect block, 4 rows, 13 direct blocks): all values now read correctly.
Regression fixture covers an 80-attribute multi-block heap.

Co-Authored-By: Claude Opus 4.8 <[email protected]>

...

This commit is contained in:

osobh

2026-06-04 01:27:45 +00:00

co-authored by Claude Opus 4.8

parent 20ad16ab69

commit 0aab49f2f0

3 changed files with 40 additions and 2 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

crates/clawhdf5-format/tests/fixtures/fractal_heap_multiblock.h5

Vendored

BIN

Binary file not shown.