Support GHC nightly

[brandonchinn178]

Resolves #19

[andreasabel]

Great work, thanks!

• Please add a test case for this feature to the CI.

• Is the bump to node 20 essential? I don't know what the best practice for actions is atm, whether node 16 is still recommended. If there are no strong reasons to move away from the default (which seems 16 still), I'd like to stick to it. The setup-node action does not know about 20, it seems from the docs, the newest there is 18: https://github.com/actions/setup-node

[brandonchinn178]

No, the bump to Node 20 isn't essential, I'll remove it. I think I was getting confused about why dist/index.js was making additional changes, but I found the actual issue. See commit message: 9e4f89a

As an aside, why does CI not check that dist/index.js is properly regenerated? I believe that would've caught this issue. People could always bypass pre-commit hooks locally, so CI should always revalidate whatever pre-commit hooks are running

[andreasabel]

Thanks for the update!

Since I am not the architect of this action but just stepped in for a burned-out maintainer I don't know how the .js generation works and what could go wrong with it. I am happy to accept additions to CI that test that it is generated correctly.

[brandonchinn178]

@andreasabel I pushed a tentative implementation, let me know what you think.

I'm also unsure why the windows job is failing now; it passed last time: https://github.com/haskell-actions/setup/actions/runs/5527334962/jobs/10082976242

[andreasabel]

@andreasabel I pushed a tentative implementation, let me know what you think.

Great, thanks for the update!

On a cursory look, I noticed that the ghcup-release-channels is implemented as a comma-separated list. Is there a specific reason for this?
A priori, I would expect that a YAML list is used here.

I'm also unsure why the windows job is failing now; it passed last time: haskell-actions/setup/actions/runs/5527334962/jobs/10082976242

This could be an upstream problem. I noticed that in the failing run, you got an older version of the Windows runner than in the successful run:

• fails with 2023-05-17: https://github.com/haskell-actions/setup/actions/runs/5539225865/jobs/10109961211?pr=20#step:1:9
• succeeds with 2023-06-30: https://github.com/haskell-actions/setup/actions/runs/5527334962/jobs/10082976242#step:1:9

In particular, in the older image ghcup additionally prints New ghc version available. If you want to install this latest version, run 'ghcup install ghc 9.6.2'. This indicates that the older image does not have the most recent GHC, so it could also have a version of ghcup that is too old (mere speculation).

[brandonchinn178]

A priori, I would expect that a YAML list is used here.

We can't use YAML lists 😢 Workflow inputs are converted into env vars. But actually, I do see @actions/core has a helper for parsing a multiline string input, so let me switch to that instead of a comma separated string

Also, I thought the action upgrades ghcup to the version in the versions file? Does it not for windows?

EDIT: oh I see it aborts before installing ghcup:

setup/src/installer.ts

Line 315 in 1ffbf9f

return 'ghcup';

It also seems like ghcup doesn't provide windows 32 binaries.

[andreasabel]

@brandonchinn178 wrote:

But I could switch to a multiline string instead of a comma separated string if you want

Yes, I think that would be more idiomatic. I know it e.g. from actions/cache inputs path and restore-keys and many other actions.

Also, I thought the action upgrades ghcup to the version in the versions file? Does it not for windows?

I would expect the same. Apparently something is not working correctly here.
The code you cite originally comes from this PR, which added the ghcup install method to Windows:

• use ghcup on windows (clone of #192) haskell/actions#206

[brandonchinn178]

I tried removing the windows early exit, and it didnt work. ghcup has windows 64 binaries but github actions uses windows 32. similarly, i tried ghcup upgrade, but that also failed

[brandonchinn178]

@andreasabel I saw you merged #27; what should I do with this PR? Should I break out the various refactorings?

• Remove broken HEAD logic
• Add new ghc-release-channels input
• Typescript improvements

[ulysses4ever]

Sad to see this lingering...