FIX add long long for int32/int64 windows compat in NumPy 2.0

[glemaitre]

closes #29028

Since NumPy 2.0 will switch from long to long long for Windows machine, then it seems that we have trigger a non-matching type as shown here: conda-forge/scikit-learn-feedstock#259 (comment)

This should fix it.

[glemaitre]

@jeremiedbb since you have a windows machine where you can build scikit-learn, could you give a try and run the test test_KNeighborsClassifier_raise_on_all_zero_weights to be sure that we don't raise the TypeError?

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: a8dff48. Link to the linter CI: here}

[glemaitre]

So from the discussion on the feedstock, it seems that this is a rather urgent fix to unblock the migration to NumPy 2.0 on the conda-forge side. I think that it makes sense to make a proper bug fix release (1.4.3) and I'll update the changelog accordingly.

[jeremiedbb]

I find that it's a huge waste of time to make a 1.4.3 today or even maybe tomorrow while we'll release 1.5.0 on monday

[jeremiedbb]

I tested locally on my windows machine and confirm that it fails without the fix and works with this fix

[jeremiedbb]

Please add np.int64 to the list of dtypes in the parametrization of test_all_with_any_reduction_axis_1

[glemaitre]

I find that it's a huge waste of time to make a 1.4.3 today or even maybe tomorrow while we'll release 1.5.0 on monday

I did not exactly the timeline for the major. But if we release on Monday, I agree.

[glemaitre]

We can also argue that we should avoid releasing on Friday.

[jeremiedbb]

We can also argue that we should avoid releasing on Friday.

That's why I said monday 😄. I aimed to create the release PR tomorrow and try to release on monday

[glemaitre]

@jeremiedbb I think this is fine to skip adding an entry in the changelog?

[jeremiedbb]

LGTM

[thomasjpfan]

LGTM