Latest Results
fix: apply drop_axes squeeze in partial decode path for sharding (#3691) (#3763)
* fix: apply drop_axes squeeze in partial decode path for sharding
When reading sharded arrays with mixed integer/list indexing (e.g.
arr[0:10, 0, [0, 1]]), the outer OrthogonalIndexer produces chunk
selections that have been ix_()-transformed for orthogonal advanced
indexing. Integer indices become single-element ranges (size-1 dims)
via ix_() to enable NumPy orthogonal indexing.
In CodecPipeline.read_batch(), the non-partial path correctly applies
drop_axes.squeeze() to remove those size-1 integer dimensions before
writing to the output buffer. However, the partial decode path (used
by ShardingCodec) was missing this squeeze step.
Fixes #3691
Also: Fix line length violation in test error message to comply with
100 character linting limit.
* fix(mypy): add type ignore comments for dynamic array indexing in sharding test
The test uses complex indexing patterns (mixed integer/list indices) that
mypy's zarr.Array stubs don't recognize as valid. Add specific type ignore
comments for [index] and [union-attr] errors to suppress false positives.
* fix(mypy): correct type-ignore codes for union attribute access in sharding test
- Line 542: Fix assert accessing .shape by changing from [index] to [union-attr]
- Line 544: Add missing type-ignore[union-attr] for f-string .shape access
- Lines 554-555: Remove unused type-ignore[index] comments on assignments
The mypy errors were caused by indexing operations returning union types that
include scalar types (int, float, etc.), which don't have a .shape attribute.
The proper fix uses type-ignore[union-attr] for attribute access, not [index].
---------
Co-authored-by: Cipher <cipher@openclaw.ai>
Co-authored-by: Davis Bennett <davis.v.bennett@gmail.com> Latest Branches
-2%
williamsnell:fill-missing-chunks -1%
maxrjones:variable-chunks-test ×260
d-v-b:perf/smarter-codecs © 2026 CodSpeed Technology