GitHub Actions: use macos-13 or macos-14 instead of macos-latest

[hugovk]

GitHub is migrating macos-latest to point to macos-14 which is (a) M1 and (b) does not have Python 3.8 or 3.9:

• https://github.blog/changelog/2024-04-01-macos-14-sonoma-is-generally-available-and-the-latest-macos-runner-image/
• Python3.9 on aarch64 for macOS 13.4.1. actions/setup-python#696 (comment)

This can cause failed builds for migrated repos.

At least during this migration, let's explicitly use macos-13 or macos-14 instead of macos-latest.

[hugovk]

GitHub has now added Python 3.8 and 3.9 to macos-14/Apple Silicon:

actions/setup-python#808 (comment)

We can save about 45s of CI time if we switch them all to macos-14:

macos-13 / Python 3.9 / 3m 50s	
macos-13 / Python 3.8 / 3m 37s

https://github.com/python-pillow/Pillow/actions/runs/8840522550/usage

macos-14 / Python 3.9 / 3m 15s
macos-14 / Python 3.8 / 3m 27s

https://github.com/hugovk/Pillow/actions/runs/8843777263/usage

Which is a nice little saving, but not huge in the scale of our CI.

The question is, shall we keep testing at least one job on macos-13/Intel to increase our test surface?

Might help find some bugs.

[radarhere]

Python 3.9.1 was the first version of Python to support Silicon, but it looks like it was only backported to Python 3.8 in 3.8.10.

It's not really a strong argument, but that makes me think we should keep macos-13 until Python 3.8 is EOL at least?

[hugovk]

Sure, although we'll still have Intel users for a while longer.

The internet speculates Apple might support Intel until 2025 or 2026 or 2027, and that we'd have to wait to see what Apple actually decides because they've not announced anything.

[radarhere]

Regarding when users will actually stop using Intel, it's worth remembering that last month we received an issue about macOS 10.10, which is almost 10 years old at this point.

[hugovk]

Indeed, and we don't have to support such old versions.

Anyway, I suggest we leave 3.8 and 3.9 on Intel for now.

[aclark4life]

Also depends on user demand. If you take away something and people notice it's gone, then we can always put it back. In general, I would drag along as broad a support as you can easily maintain. As the maintenance burden increases for older tech, you can base the decision to drop it on predicted user demand and act accordingly. If you need to support something you can't easily support, then it gets harder. 🤔

[hugovk]

Yeah, it all adds up. I think a fair amount of effort has gone into maintaining old macOS versions, all of which is time away from everything else.

Having clear policies helps both our users and us. We have:

In general, we aim to support all current versions of Linux, macOS, and Windows.

Which means macOS 12, 13 and 14, not 10.10, 10.11, 10.12, 10.13, 10.14, 10.15 or 11 - 10.10 is very much no longer receiving security updates.

For the time-being, I don't mind keeping 10.10 along.