[pre-commit.ci] pre-commit autoupdate

[pre-commit-ci]

updates:

• github.com/astral-sh/ruff-pre-commit: v0.11.10 → v0.11.11

[hameerabbasi]

cc @mtsokol @willow-ahrens Seems like we're ignoring keepdims here; assuming it's True. Was this related to the change discussed in the dev meeting two weeks ago?

[codspeed-hq]

CodSpeed Performance Report

Merging #865 will degrade performances by 24.75%

_{Comparing pre-commit-ci-update-config (a28db0e) with main (9988853)}

Summary

⚡ 1 improvements
❌ 1 regressions
✅ 338 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	test_gcxs_dot_ndarray[coo-m=200-n=200-p=200]	2.8 ms	1.9 ms	+43.55%
❌	test_index_fancy[side=100-rank=1-format='coo']	939.4 µs	1,248.3 µs	-24.75%

[willow-ahrens]

We never supported keepdims, a few days ago we merged a change which we thought would support keepdims, now it seems that this is broken, though only in the lazy case. I'll investigate.

[willow-ahrens]

By the way: Should these finch tests be moved to finch-tensor-python, or at least called from CI there?

[willow-ahrens]

I cannot reproduce the failing case locally. I am getting:

>>> A = finch.lazy(finch.Tensor(numpy.eye(5)))
>>> B = finch.lazy(finch.Tensor(sps.csr_matrix(numpy.eye(5))))
>>> finch.compute(finch.sum(finch.add(A, B), axis=0)).todense()
array([2., 2., 2., 2., 2.])

[willow-ahrens]

Is the problem just that we need to cut a new release of finch-tensor-python?

[mtsokol]

I see that the new release arrived. I cleared CI caches and re-triggered jobs.

[willow-ahrens]

was that the fix then? Perhaps I should have bumped the breaking version number with the last finch.jl release. It was a bugfix but a breaking one.

[mtsokol]

was that the fix then? Perhaps I should have bumped the breaking version number with the last finch.jl release. It was a bugfix but a breaking one.

Yes, new release of finch-tensor-python fixed it! The previous version of finch-tensor-python was installed previously but newest Finch.jl so Python layer was missing the keepdims handling (I'm surprised that it installed newest Finch.jl as we also hardcode Finch version).

Numba Array API error is unrelated IMO, for Finch Array API I need to adjust skips and xfails file.

[mtsokol]

Additional Numba and Finch Array API test failures are due to new hypothesis version IMO.

[mtsokol]

A few edge cases need to be resolved first to make Array API Finch job green (described in finch-tensor/Finch.jl#743)