MAINT: more 1.9.2 backports prep

[tylerjereddy]

Backport (with merge conflicts manually resolved, please check):

• FIX: milp: return feasible solutions if available on time out #16814
• MAINT: stats.pearson3: fix ppf for negative skew #17055

Uncomment the wheel upload code as suggested in #17132 (review). To produce a match to main for the upload block I did git checkout -p main -- .github/workflows/wheels.yml and applied the third hunk of the diff.

[andyfaff]

The wheels.yml changes look ok, in that it matches main. However, whether it works or not is another matter!

[mdhaber]

@mckib2 I think we need to remove that test in main while we search for a better test problem. That timeout is too long, and the time_limit is too long, too.
@tylerjereddy I forgot that gh-16814 also passed a new option to HiGHS. That might not be appropriate for a bug fix release. I can try to cut that out tomorrow.

[tylerjereddy]

@mdhaber so, I should revert one of the backports, or do both need some adjustment? If they aren't crucial it may be better to focus on final cibuildwheel checking pretty soon.

[mdhaber]

gh-17055 is fine; only gh-16814 needs adjustment. If you need to revert wholesale due to time crunch, I understand, but if I would have time to work on it this evening (not weekend, which I accidentally wrote originally), I can find just the parts of it that need to be removed.

[rgommers]

You can keep the fix but remove the timeout test completely. There's no real need for a test backport here. The fix itself seems fine still, just the test is not reliable.

[tylerjereddy]

@mdhaber happy on your end here?

[mdhaber]

Yes. It all looks good, and I've tested locally and it all seems to be working. (Did you want me to merge? Not sure if you typically squash these.)

[tylerjereddy]

I'll probably merge it today thanks. We don't usually squash, instead keeping the individual backports (which themselves are typically squashed on merge in their individual PRs).

[tylerjereddy]

Thanks for the help all, the delay here is just me now. I'll see what I can do this weekend. I'll probably wait until tomorrow (Saturday) to start the process since it will be slightly different and I'll want to be a little careful.

[andyfaff]

Check that all the macosx_arm64 wheels are tagged 12_0 not 11_0

[tylerjereddy]

Wheel builds now running after making small adjustment from gh-17171 : https://github.com/scipy/scipy/actions/runs/3211264776

[tylerjereddy]

The arm64 wheels will indeed need renaming. I won't open an issue for that because I was under the impression this may be smoother on main already, but if I'm wrong about that then please do open an issue.

-rw-rw-r-- 1 treddy treddy 28M Oct  8  2022 scipy-1.9.2-cp310-cp310-macosx_11_0_arm64.whl
-rw-rw-r-- 1 treddy treddy 28M Oct  8  2022 scipy-1.9.2-cp311-cp311-macosx_11_0_arm64.whl
-rw-rw-r-- 1 treddy treddy 28M Oct  8  2022 scipy-1.9.2-cp38-cp38-macosx_11_0_arm64.whl
-rw-rw-r-- 1 treddy treddy 28M Oct  8  2022 scipy-1.9.2-cp39-cp39-macosx_11_0_arm64.whl

I opened two smaller issues for convenience matters on the upload/workflow side, but so far things are probably still faster even with me manually unzipping artifacts, so that's good and my complaints look pretty easy to address I think.

[andyfaff]

The 11_0 --> 12_0 renaming is fixed on main.