Test on Ruby 3.1

[Bo98]

HOMEBREW_BOOTSNAP is not supported for now, as it doesn't work on anything not macOS system Ruby. I will investigate this further soon.

[MikeMcQuaid]

Looks good to far! Do we definitely need to test this on macOS and Ubuntu?

[Bo98]

macOS: yes, Cask had several issues
Ubuntu: don't think there was really anything flagged on this occasion, depends how much you value the extend/os/linux stuff.

[MikeMcQuaid]

@Bo98 Cool. My temptation would be that we consider adding these tests and then just defaulting (all?) Linux CI to use the newest Ruby we plan on using and have this split for macOS. Thoughts?

[Bo98]

Until the point portable Ruby moves to 3.1, it's worthwhile having at least one of Linux CI (not sure which one) running on 2.6 since adjustments for Ruby 3 can break 2.6.

[Bo98]

Going forward might be good to have a policy of how we handle this. If we're doing a switch every year, do we have it so that when we switch to 3.1 default we have one or two jobs testing 3.2 ahead of time (so some on "default" and a couple on "default + 1").

The backbone is in place now, so it's simply a case of tweaking the matrix, setting ruby: whatever wherever.

[MikeMcQuaid]

@Bo98 I think once we're back to supporting a specific major/minor version combination I think we just keep CI at the lowest version and let people use newer ones but flag it in brew doctor etc. and tell them to file PRs to fix. I don't want us to be actively supporting a too wide range of versions (which is also why newer versions are more appealing)

[Bo98]

Was more to prepare for when we migrate each year but yeah I'm guessing most Ruby versions don't have compatibility issues. In the scenario of a theoretical Ruby 4, it might be worth testing that ahead of time.

[Bo98]

I think we can move pretty quickly with the migration tbh. All the tests pass so it'll be a case of building a portable Ruby, adding an env to use it and most of the testing work will be areas we don't have CI coverage for (hopefully little) and individual formulae and casks (though I reckon kwarg hash usage there is minimal to none). I've not noticed any other issues using Ruby 3 so far with these fixes.

The larger changes, like dependency updates, will come after we drop 2.6 as it's signficantly easier to deal with if we wait until then. The TargetRubyVersion bump in RuboCop may or may not also bring up a lot of things.