CI: GHC 9.2.7 Support

[andys8]

Draft for now to check with CI what needs to be changed.

Solves #3511

[andys8]

Looks like GHC 9.2.7 can't be installed with chocolatey on Windows, but that's also the case for 9.2.6 🤔

[andys8]

• ❓ CI had exceptions for 9.2.5. Not sure if this was due to issues with 9.2.5 or 9.2 in general.
• ❓ Nix failing due to cachix. Related or flaky?
• ❓ Can't be installed on Windows? Expected or does another place need to be updated?
• ❓ What else needs to be changed?
• ℹ️ cabal build and ghcup compile hls seem to work with 9.2.7 on my machine

[michaelpj]

Looks like GHC 9.2.7 can't be installed with chocolatey on Windows, but that's also the case for 9.2.6

Right, but we never got to 9.2.6 in our CI, we were on 9.2.5. Looks like we're blocked on haskell/actions#190

CI had exceptions for 9.2.5. Not sure if this was due to issues with 9.2.5 or 9.2 in general.

Almost always major-version related so updating them is right.

Nix failing due to cachix. Related or flaky?

It's flaky, that's why it's not required.

So overall I think this PR looks right, we just need the setup action to support it.

[andys8]

Cool, maybe the change we're waiting for isn't far away.
Mistuke/GhcChoco@bcbf15a

[michaelpj]

Hmm, I don't understand how github action updates work. They released a minor version that supports the new GHCs, so shouldn't this now work?

[andys8]

I don't quite understand how HLS CI works and which of the jobs are known to be flaky 😅

The latest run seems to have failed for 9.0.2 which should be unrelated to the changes in this PR. Still not sure if my changes are complete and correct, though.

[fendor]

Haskell setup actions are broken for windows haskell/actions#202 (comment)

[July541]

Just a kindly reminder, do you still have the plan to continue?

[andys8]

Resolved the merge conflicts

[July541]

The test failed four times, are we so unlucky? 🤔

[andys8]

Could somebody have a look at CI and this PR? Otherwise should I consider closing it?

[fendor]

@andys8 Yep, will do!

[fendor]

As @July541 correctly identified, there are indeed three test-cases that were incorrectly marked as broken because the distribution issue for GHC 9.2.7 has been fixed.
I removed the special casing, since our CI doesn't support more than one GHC 9.2 version and adding extra CPP seemed like a bad idea.

[fendor]

@andys8 thank you for the PR, sorry it took us so long to get it finally merged!

[michaelpj]

The shame of not getting it merged was motivational :D 🙃

[andys8]

No need to apologize. Thanks for your work.