Use ghc+integer-gmp for alpine linux build release

[jneira]

• After diagnose the relese build was failing for alpine+ghc with integer-simple due to a new package requiring integer-gmp (https://hackage.haskell.org/package/mod)
• Pinged @maerwald and kindly added in ghcup the missinf ghc versions for alpine with integer-gmp (Use integer-gmp bindists for alpine x86_64, fixes #14 ghcup-metadata#15)
• So we can remove the existing workaround here and let haskell/action download the new bindists

[jneira]

• Checking the release build here: https://github.com/jneira/haskell-language-server/actions/runs/1725135759

[jneira]

The build has progressed but it has failed for linux and ghc-9.2.1 with

src/Ide/Version.hs:24:21: error:
Error:     • Exception when trying to run compile-time code:
        PATH: getEnv: does not exist (no environment variable)
      Code: gitCommitCount
    • In the untyped splice: $gitCommitCount
   |
24 |   let commitCount = $gitCommitCount
   |                     ^^^^^^^^^^^^^^^
cabal: Failed to build hls-1.5.1.0 (which is required by exe:hls from
hls-1.5.1.0).

🤔

[hasufell]

The build has progressed but it has failed for linux and ghc-9.2.1 with

src/Ide/Version.hs:24:21: error:
Error:     • Exception when trying to run compile-time code:
        PATH: getEnv: does not exist (no environment variable)
      Code: gitCommitCount
    • In the untyped splice: $gitCommitCount
   |
24 |   let commitCount = $gitCommitCount
   |                     ^^^^^^^^^^^^^^^
cabal: Failed to build hls-1.5.1.0 (which is required by exe:hls from
hls-1.5.1.0).

thinking

Sounds like this part is failing: https://github.com/acfoltzer/gitrev/blob/450a7abf444260e95791649489705e54650c0232/src/Development/GitRev.hs#L65

The environment is borked (doesn't contain the PATH env var)?

[jneira]

it does (https://github.com/jneira/haskell-language-server/runs/4889676070?check_suite_focus=true)

Path: /__t/stack/2.7.3/x64:/github/home/.ghcup/bin:/github/home/.ghcup/ghc/9.2.1/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

the build is successful for all ghcs < 9.2.1, using the same docker image for all, and the build for ghc-9.2.1 is successful in Ubuntu so i would say it is a weird interaction between both

[jneira]

well the workflow between the Ubuntu build and the alpine one 8s not exactly the same but I don't see any change which could cause the bug
will try to use Ubuntu in the same workflow to check it though (and as possible workaround)

also we could trace the value of the PATH env bar in our code but getExecutable is pretty standard

[jneira]

the build for ghc-9.2.1 is successful in Ubuntu so I would say it is a weird interaction between both

well the ghc version for alpine is not exactly the same than the Ubuntu one

anyone has an alpine machine to test the build, try to reproduce and trace the cause easier?

[jneira]

As first sanity check i am gonna invalidate the cache created in ubuntu

[jneira]

A fresh cache did not fix it: https://github.com/jneira/haskell-language-server/runs/4893126248?check_suite_focus=true

[jneira]

As expected changing alpine for ubuntu in the ghc-9.2.1 build has been succesful so the workflow is not the culprit: https://github.com/jneira/haskell-language-server/runs/4893355328?check_suite_focus=true

At this point i think we could consider use ubuntu for ghc-9.2.1 (it is experimental anyways) until we fix the bug for alpine + ghc-9.2.1. That way we will unblock the incoming release.

//cc @hasufell @pepeiborra @michaelpj

[michaelpj]

@Bodigrim made a release of mod with integer-simple (Bodigrim/mod#13 (comment)), so if we bump the index-state we may find that cabal can now get a build plan when using integer-simple.

[michaelpj]

Personally, the commit count suffix on the version string doesn't seem that useful to me, we could consider just binning that.

[hasufell]

@Bodigrim made a release of mod with integer-simple (Bodigrim/mod#13 (comment)), so if we bump the index-state we may find that cabal can now get a build plan when using integer-simple.

Probably won't, because I already updated ghcup alpine bindists to integer-gmp.

[pepeiborra]

I support dropping the dependency if it helps.

[jneira]

All gitrev definitions needs find git in the env var PATH throught findExecutable "git" . We are actually using it to get the git commit here:

haskell-language-server/src/Ide/Arguments.hs

Lines 141 to 150 in b937baa

	haskellLanguageServerVersion :: IO String
	haskellLanguageServerVersion = do
	path <- getExecutablePath
	let gitHashSection = case $(gitHash) of
	x | x == "UNKNOWN" -> ""
	x -> " (GIT hash: " <> x <> ")"
	return $ "haskell-language-server version: " <> haskellLanguageServerNumericVersion
	<> " (GHC: " <> VERSION_ghc
	<> ") (PATH: " <> path <> ")"
	<> gitHashSection

and i would like to keep the commit when doing hls --version

The alternative githash dont use findExecutable though but does a raw (proc "git" args) { cwd = Just root } so maybe it would work 🤷

[jneira]

Meanwhile i've reproduced the ghc+alpine bug in a fresh alpine-3.12 vm installing ghc with ghcup:

wonderful

System.Environment.getEnv "PATH" also fails (and with whatever env var known to be set like $HOME, $USER, etc)
Opened https://gitlab.haskell.org/ghc/ghc/-/issues/20977

[jneira]

It worths to note that ghc-9.2.1 in alpine seems to be fragile:

• "Alpine segfaults many tests in 9.2.1" https://gitlab.haskell.org/ghc/ghc/-/issues/20266
• "Alpine ci job is broken" https://gitlab.haskell.org/ghc/ghc/-/issues/20574

[jneira]

Trying githash in alpine+9.2.1 here: https://github.com/jneira/haskell-language-server/runs/4896130017?check_suite_focus=true

[jneira]

Ok, githash worked, yay! thanks @hasufell for the suggestion
I stand a little bit disturbed by the bug but whatever

[hasufell]

@jneira I've updated the ghc 9.2.1 alpine bindist in ghcup. getEnv should work now.

[jneira]

@hasufell many thanks! i've reran the build workflow and it should use the new ghc bindist: https://github.com/haskell/haskell-language-server/actions/runs/1728847536

EDIT: it did:

[ Info  ] downloading: https://downloads.haskell.org/~ghcup/unofficial-bindists/ghc/9.2.1/ghc-9.2.1-x86_64-alpine-linux-integer-gmp.tar.xz as file /tmp/ghcup-b0002d8433a56e77/ghc-9.2.1-x86_64-alpine-linux-integer-gmp.tar.xz

[jneira]

@jneira I've updated the ghc 9.2.1 alpine bindist in ghcup. getEnv should work now.

i've double checked in my vm and it works as intended 👍

[jneira]

Well i think we can keep githash cause as commented by @hasufell it is being actively maintained so the pr could be merged if you agree

[pepeiborra]

The githash change looks fine to me, but I haven't reviewed the CI changes. Maybe @Anton-Latukha can take a look.

[Anton-Latukha]

Well, as I came around to look at it... To review it, I needed/need to fully learn & understand the build workflow, so time.

Overall, glad to see build being ported to a formalized pattern.

Seems good. Would submit PRs if would find something.

[pepeiborra]

does this need to be $CABAL_ARGS ?

[jneira]

omg 😧

[jneira]

that should have failed, as @hasufell suggested, setting the appropriate bash flag