chore(ci): use macos-15-intel for x86 wheel building and testing

[ariostas]

The wheels that are built in macos-14 and macos-15 are identical (since they are both arm64), so I just switched to using `macos-latest.

This closes #3718 which was failing since there were duplicate wheels.

Also, I guess we quietly dropped support for x86 macs, but no one has complained, so I guess it's fine.

[ikrommyd]

Also, I guess we quietly dropped support for x86 macs, but no one has complained, so I guess it's fine.

I see x86 macos wheels under https://pypi.org/project/awkward-cpp/50/#files

[ariostas]

Oh right, there hasn't been an awkward-cpp release since #3666, so I guess we still have time to properly announce it and/or bump minor version when there is a new awkward-cpp release.

[github-actions]

The documentation preview is ready to be viewed at http://preview.awkward-array.org.s3-website.us-east-1.amazonaws.com/PR3720

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.72%. Comparing base (b749e49) to head (12f27a9).
⚠️ Report is 487 commits behind head on main.

Additional details and impacted files

see 200 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ariostas]

Thanks to Henry for pointing out that macos-15-intel was an option!

[ianna]

@ariostas - it looks like the ci runner does not come with llvm?

[ariostas]

@ianna I think we'll have to restrict to numba 0.62 for macOS x86 because both numba and llvmlite stopped shipping wheels for it.

[ianna]

@ianna I think we'll have to restrict to numba 0.62 for macOS x86 because both numba and llvmlite stopped shipping wheels for it.

Yes. Or maybe it is time to drop macOS x86?

[ariostas]

Hmm yeah, I guess we should do that soon. Maybe for now we can give up on doing the full testing and just test with the minimal requirements. Is there a plan for a minor version bump soon? We probably should schedule the drop of support to go with that

[ikrommyd]

@ianna I think we'll have to restrict to numba 0.62 for macOS x86 because both numba and llvmlite stopped shipping wheels for it.

I'm a bit confused, shouldn't leaving it unpinned just grab the latest possible version (0.62) if no wheels exist for 0.63?

Edit: No I'm stupid. It would try to compile from source.

[ikrommyd]

Or maybe it is time to drop macOS x86?

I think for that, it's probably good to see what the scientific python community is doing in general.

[ianna]

Hi @ariostas - I think, just tensorflow-cpu would be ok for ml on ubuntu. Thanks.

[ariostas]

@ianna this is ready now. The GPU tests are failing, but that's due to the cudf issue. Also, you'll need to mark some of the tests are not required since it's not longer going to test on all versions of Python.

[ianna]

@ariostas - Great! Thanks for making it happen. Please go ahead and merge it. Thanks!