github: use macos-14, add new aarch64-apple-darwin release binary

[thoughtpolice]

GitHub announced these new Apple Silicon based runners today. Let's take them for a spin.

Let's also add an entry in the release matrix to build and publish aarch64- apple-darwin binaries, too. This doesn't migrate the old release matrix entry; it still uses a macos-11 runner. This means the x86 binaries should work on a few older macOS versions if users need it.

This should also close #1887, since these runners are no longer considered to be in beta.

Checklist

If applicable:

• I have updated CHANGELOG.md
• I have updated the documentation (README.md, docs/, demos/)
• I have updated the config schema (cli/src/config-schema.json)
• I have added tests to cover my changes

[ilyagr]

Thanks!

Have you double-checked that the release workflow works? You can e.g. create a release in your fork. If not, I can try to check it.

The x86 version of jj does work with Rosetta, but an aarch64 binary is nicer. For example, cargo binstall should then work even if Rosetta is not installed.

For future: Is there a way to run tests on aarch64 as well? We could also consider a universal binary, but

Aside: cargo-binstall uses https://github.com/cross-rs/cross for cross-compilation. I was wondering whether to use that for an aarch64-linux binary, but I would guess it's slow (it runs cargo in a VM, I think).

[thoughtpolice]

Have you double-checked that the release workflow works? You can e.g. create a release in your fork. If not, I can try to check it.

No, I was going to take it for a spin later before bugging Martin to approve this, but if you can kick the tires on it for me, I'd appreciate it!

For future: Is there a way to run tests on aarch64 as well? We could also consider a universal binary, but

You mean macOS x86_64? This PR migrates the actual macOS CI build (used for every PR) to use aarch64, I believe macos-14 is only aarch64, so it is running the tests on aarch64 by definition from now on. Which I think is fine.

What we might be able to do is kill two birds with one stone and run a parallel build job that runs the job's shell under Rosetta and thus builds an x86_64 binary and tests it. That would give us pretty good confidence that we have all bases covered. EDIT: But it would probably be really annoyingly slow, if I had to guess, so maybe not worth it.

[ilyagr]

I think the two "build"s in release.yml need to have different names. Only aarch64 seems to currently be built.

https://github.com/ilyagr/jj/actions/runs/7717709786/job/21037387946
https://github.com/ilyagr/jj/releases/tag/0.13.1-aarch64test

(The builds are not quite finished as I'm posting this)

[thoughtpolice]

Good catch, pushed a fix.

[ilyagr]

Looks pretty good. The results should appear here eventually:

https://github.com/ilyagr/jj/releases/tag/0.13.3-aarch64test
https://github.com/ilyagr/jj/actions/runs/7717825075

[ilyagr]

You could also put this in the changelog ("Official darwin-aarch64 binaries are now available")

[ilyagr]

The binaries seem to work, except for a new bug I discovered: #2896. That bug is already present in jj v0.13

You mean macOS x86_64? This PR migrates the actual macOS CI build (used for every PR) to use aarch64, I believe macos-14 is only aarch64, so it is running the tests on aarch64 by definition from now on. Which I think is fine.

Indeed, the tests run on aarch64 now. I'm not sure whether it's good enough to only run aarch64 tests at this point, but it might be OK. We could also add a macos-13 runner.

I'll approve this, but we'll need @martinvonz to look at this regardless. This needs changes to required CI checks (one got renamed).

[martinvonz]

I'll approve this, but we'll need @martinvonz to look at this regardless. This needs changes to required CI checks (one got renamed).

Thanks for reviewing and testing!

[martinvonz]

And thanks for the PR, @thoughtpolice! As you can see, I've updated the branch protections so you can merge it.