WHL: what should we do about mac-intel wheels ?

[neutrinoceros]

GitHub is about to discontinue the macos-13 runner image, which is the last image which we can test x86_64 wheels for macOS on. We might still be able to cross-compile for this target via more recent image, which all run on mac-Silicon archs, but testing will be impossible on GHA in a couple months.
The newest Macs running on Intel processors were issued in late 2020, which is not that long ago, so it is very likely that a fraction of our users are still running this hardware (FWIW, I'm actually typing these lines on a mac-intel).
That said, this fraction is bound to shrink if it isn't negligible already, so I assume it is indeed as good a time as any to discontinue tier 1 support for this platform. Here are our options:

1. simply stop supporting mac-intel when testing becomes impossible
2. continue builds and distributions (using cross compilation) for a time, but give users explicit notice that these wheels are not tested anymore
3. move this target to another CI service (which one ?). This would only solve the problem for a time, of course.
4. do whatever numpy ends up doing

All of these seem reasonable in my own opinion, but we have a little time ahead of us to build consensus, so I'd like others to weigh in before I open a PR with whatever solution is prefered by the community (and in particular, our release team).

[neutrinoceros]

(tentatively putting this in the 7.1.1 milestone, because it might turn into a "🔥" if the image goes away before we cut the release, but otherwise, this can safely be moved to the next milestone)

[pllim]

Thanks! I also posted at https://groups.google.com/g/astropy-dev/c/UZcY3usZHRE

[pllim]

I also asked Numpy (couldn't find an issue on their GH) and will report back if I hear anything.

[neutrinoceros]

did you ask on a public channel ? I cannot find it

[tepickering]

i'd say 4 makes the most sense. otherwise 2 as a way to continue, but phase out support. i don't think 3 is worthwhile for a deprecated platform. 1 seems a bit hasty and harsh. there are enough mac-intel machines out there that i don't think we want to pull the rug out too quickly.

[pllim]

did you ask on a public channel ?

Scientific Python discord, numpy channel, yes.

https://discord.com/channels/786703927705862175/1020054688818806824/1397996350913773568

[pllim]

Also have to get rid of this:

astropy/.github/workflows/ci_cron_weekly.yml

Line 67 in 8599d49

os: macos-13

UPDATE: Done in #18460

[neutrinoceros]

Oh actually, it looks like we are already cross-compiling and testing (which I didn't realize was possible) these wheels: OpenAstronomy/github-actions-workflows#257
so the issue is definitely less urgent than I previously thought. We'll still need to decide when to discontinue support for mac-intel, but we don't have to act now.

[pllim]

Reply from Numpy: the plan in cibuildwheel is to cross-compile them just like they did for ARM before there were ARM runners. With rosetta it’s way easier to test the resulting wheels locally. I think the plan is to just handle all this in cibuildwheel, which that team will focus on in due time.

[pllim]

So, if we go with Option 4 (follow numpy) and there is already #18460 to remove explicit call for macos-13 runners, we can just close this issue now? What else need to be done?