Conversation
e7f3dcd to
6cc41dd
Compare
|
I'm going to ignore
|
heh, the release is out on PyPI but these flaky/irrelevant failures block finishing the release process: this PR cannot be merged because branch protection rules :-)). So PSA: main is blocked for merging until this mess is dealt with :-). |
NumPy < 2 fails to promote an empty f32 array with a scalar, returns an empty f64 array
How about we turn off that part of the branch protection? It's been a repeated mess, and I can't think of any other project I work on that doesn't allow committers to ignore flaky failures or known issues. |
|
I've done that now, because a release from an unmerged PR is quite bad. @ev-br this is mergeable now. I didn't review, but trust you can make the right decision here. |
|
Thanks Ralf! This is really helpful. A few hours later, and with a couple of fixes to the test suite, we're down to: one failure on numpy 1.21, and a new failure from ruff (which made a release several hours ago), telling us that array-api-compat code is using type annotations in a way incompatible with python 3.9. |
No description provided.