BUG: Fix np.dot to allow user dtypes#30931
Merged
seberg merged 13 commits intonumpy:mainfrom Mar 6, 2026
Merged
Conversation
MaanasArora
force-pushed
the
bug/allow-dotfunc-user-dtypes
branch
from
01363f7 to
f057091
Compare
MaanasArora
changed the title
PyArray_InnerProduct to allow user dtypesnp.dot to allow user dtypes
MaanasArora
force-pushed
the
bug/allow-dotfunc-user-dtypes
branch
from
f057091 to
9481f81
Compare
seberg
reviewed
Mar 4, 2026
Member
|
ping @SwayamInSync this seems like something you might be interested in |
…ypes in PyArray_MatrixProduct2
seberg
reviewed
Mar 5, 2026
Member
seberg
left a comment
seberg
left a comment
There was a problem hiding this comment.
Thanks, I am very happy to do this PyArray_DTypeFromObject but I think it makes sense here to apply it accross all functions that end up using new_array_for_sum.
seberg
reviewed
Mar 5, 2026
Member
seberg
left a comment
seberg
left a comment
There was a problem hiding this comment.
There is still a segfault, hopefully something small. But I also noticed 2 smaller things.
Thanks for looking at all of it now!
Co-authored-by: Sebastian Berg <sebastian@sipsolutions.net>
MaanasArora
force-pushed
the
bug/allow-dotfunc-user-dtypes
branch
from
970355f to
0f8f781
Compare
seberg
reviewed
Mar 5, 2026
seberg
reviewed
Mar 5, 2026
seberg
approved these changes
Mar 5, 2026
SwayamInSync
approved these changes
Mar 6, 2026
Member
SwayamInSync
left a comment
SwayamInSync
left a comment
There was a problem hiding this comment.
I'm sorry coming late, this looks good to me.
Thanks @MaanasArora for working on this
seberg
approved these changes
Mar 6, 2026
Member
seberg
left a comment
seberg
left a comment
There was a problem hiding this comment.
Thanks for the final fixes, let's get this in. (It wouldn't hurt to add more byte-swap related tests, but I don't want to block on that -- all modified code paths include NPY_DT_CALL_ensure_canonical.)
honi-at-simspace
added a commit
to HoniSanders/numpy
that referenced
this pull request
Apr 15, 2026
Brings the maxlag2025 branch up to date with current numpy/numpy main (843 commits since branch base). Conflict resolution in numpy/_core/src/multiarray/multiarraymodule.c: - Adopt upstream's _pyarray_correlate signature change from `int typenum` to `PyArray_Descr *typec` (PR numpy#30931). - Keep our refactor: the function still takes (minlag, maxlag, lagstep) instead of (mode, *inverted), and handles array swap, negative-step normalization, and output reversal internally. - Update PyArray_Correlate, PyArray_Correlate2, and PyArray_CorrelateLags to use upstream's PyArray_DTypeFromObject + NPY_DT_CALL_ensure_canonical pattern for type resolution, with proper Py_DECREF(typec) cleanup. - Update the new array_correlatelags argument parser to use upstream's brace-syntax for npy_parse_arguments. Add NPY_2_6_API_VERSION 0x00000016 and matching version-string clause in numpy/_core/include/numpy/numpyconfig.h to support the bumped C API version that registers PyArray_CorrelateLags at slot 369. Tests: numpy/_core/tests/test_numeric.py: 512 passed, 1 skipped.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This is a start to fix
PyArray_MatrixProduct2for user-defined dtypes, as a bug was reported in jax-ml/ml_dtypes#360. This is a draft as I've not tested this outside of a dummy dtype. I will try to use the linked PR with this, but wanted to get this out if anyone has comments. ping @seberg - thanks!There also seem to be other methods with the same issue, so we might need a helper function long run. Hopefully this stays minimal on more testing, though not sure if even this is small enough to backport.