Support GHC 9.4

[ulysses4ever]

Current status: slowly working through test failures... See discussion below.

---

• Patches conform to the coding conventions.
• Any changes that could be relevant to users have been recorded in the changelog.
• The documentation has been updated, if necessary.

Please also shortly describe how you tested your change. Bonus points for added tests!

[ulysses4ever]

Only the last commit in this PR is actually the contribution.

[ulysses4ever]

Windows is failing because cabal 3.8 is required there for GHC 9.4, but haskell/actions/setup still thinks that latest is 3.6?

[ulysses4ever]

Right... haskell/actions#113

Maybe it's about the time to move to pure ghcup workflow…

[ulysses4ever]

Tests are not ready yet: opened haskell-hvr/windns#3

[ulysses4ever]

Need to add TypeOperators extension to make it compile with 9.4. This was brought up earlier in #8268.

[ulysses4ever]

Hmm, an actual test failure… Backpack.

[Mikolaj]

Does it look like a GHC bug or a cabal bug that manifests only in the configuration we happened to set for GHC 9.4?

[Mikolaj]

That's neither:

haddock: internal error: haddock:iface

Not very surprising. Could we mark it as expected fail for 9.4?

Edit: unless it doesn't fail locally, but I hope it's at least consistent...

[ulysses4ever]

@Mikolaj I'm sorry, I don't see "haddock: internal error" on the CI. When I click on Details for the 9.4.1 job, I see

UNEXPECTED FAIL: PackageTests/Backpack/Includes2/setup-per-component.test.hs PackageTests/Backpack/Includes2/setup-external.test.hs PackageTests/Backpack/Includes3/setup-external-ok.test.hs

and Ctrl+F doesn't help either. What am I missing?

[Mikolaj]

Yes, C-f doesn't work (perhaps in a raw log it does). Perhaps scroll, in https://github.com/haskell/cabal/runs/7984223230?check_suite_focus=true, up to

Warning: The documentation for the following packages are not installed. No links will be generated to these packages: base-4.17.0.0, ghc-bignum-1.3, ghc-prim-0.9.0, Includes2-0.1.0.0
creating setup-per-component.dist/work/dist/doc/html/Includes2
Running: /usr/local/.ghcup/ghc/9.4.1/bin/haddock-ghc-9.4.1 '@setup-per-component.dist/work/dist/doc/html/Includes2/haddock-response35120-1.txt'
   0% (  0 /  3) in 'Mine'
  Missing documentation for:
    Module header
    Mine (mylib/Mine.hs:3)
    mine (mylib/Mine.hs:4)
haddock: internal error: haddock:iface
CallStack (from HasCallStack):
  error, called at utils/haddock/haddock-api/src/Haddock/Interface.hs:217:16 in main:Haddock.Interface


*** unexpected failure for PackageTests/Backpack/Includes2/setup-per-component.test.hs

[ulysses4ever]

ah, now I see it, thanks!

[ulysses4ever]

Windows 🤯

[Mikolaj]

Joy. This is something to do with foreign libraries, which has just been changed in GHC and cabal. @bgamari: does it look there is a problem in the new machinery or is the test outdated somehow? It's this one: https://github.com/haskell/cabal/tree/master/cabal-testsuite/PackageTests/ForeignLibs

[Mikolaj]

@Mistuke: did you see anything like that in you 9.4 tests or does it ring a bell otherwise?

+ "C:\tools\ghc-9.4.1\lib\../mingw/bin\clang.exe" "-std=c11" "-Wall" "-o" "uselib" "UseLib.c" "-l" "myforeignlib" "-L" "D:\a\cabal\cabal\cabal-testsuite\PackageTests\ForeignLibs\setup.dist\usr"
ld.lld: error: undefined symbol: sayHi

>>> referenced by C:/Users/RUNNER~1/AppData/Local/Temp/UseLib-ef4d96.o:(main)

clang: error: linker command failed with exit code 1 (use -v to see invocation)


stderr:
*** Exception: Command "C:\tools\ghc-9.4.1\lib\../mingw/bin\clang.exe" "-std=c11" "-Wall" "-o" "uselib" "UseLib.c" "-l" "myforeignlib" "-L" "D:\a\cabal\cabal\cabal-testsuite\PackageTests\ForeignLibs\setup.dist\usr" failed.
Output:
ld.lld: error: undefined symbol: sayHi

The log is from https://github.com/haskell/cabal/runs/8028896326?check_suite_focus=true

The test is https://github.com/haskell/cabal/tree/master/cabal-testsuite/PackageTests/ForeignLibs

[ulysses4ever]

@Mikolaj incidentally: this is very close to what I saw on hackage-security CI when updating to GHC 9.4 but forgetting to update to cabal 3.8 at the same time. Example

[Mistuke]

@Mistuke: did you see anything like that in you 9.4 tests or does it ring a bell otherwise?

+ "C:\tools\ghc-9.4.1\lib\../mingw/bin\clang.exe" "-std=c11" "-Wall" "-o" "uselib" "UseLib.c" "-l" "myforeignlib" "-L" "D:\a\cabal\cabal\cabal-testsuite\PackageTests\ForeignLibs\setup.dist\usr"
ld.lld: error: undefined symbol: sayHi

@Mikolaj in GHC 9.4 the foreign wrappers were rewritten, so might be a bug there, but the test itself is a bit suspicious, the call to sayHi isn't marked extern so it's visibility is local by default..

Does it work if you add extern here? https://github.com/haskell/cabal/blob/master/cabal-testsuite/PackageTests/ForeignLibs/UseLib.c#L7

[ulysses4ever]

@Mistuke adding extern doesn't seem to make a difference.

[Mistuke]

Ah, I'll take a look tomorrow morning then.

…

Sent from my Mobile

On Fri, Aug 26, 2022, 23:40 Artem Pelenitsyn ***@***.***> wrote: @Mistuke <https://github.com/Mistuke> adding extern doesn't seem to make a difference. — Reply to this email directly, view it on GitHub <#8420 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAI7OKOYWJFMUHFDPACJIUDV3FBVLANCNFSM57IZ3M3A> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[andreasabel]

I needed not-yet-merged #8398, so it may look a bit messy. Hopefully, after that one is merged, we rebase and all is good.

I hand-merged it, so maybe you can rebase this on latest master now?

[ulysses4ever]

@andreasabel

I hand-merged it, so maybe you can rebase this on latest master now?

Thanks. Done.

[Mistuke]

@ulysses4ever I'm having some trouble reproducing the issue, what's the easiest way to?

[Mistuke]

alternatively, is it possible to get a GHC -v3 log from the test compilation?

[ulysses4ever]

@Mistuke I'm sorry but I don't have a Windows machine suitable for reproduction of the failure at hand. We only see the failure in CI currently. I can try to get such a machine but it may take time. If you have one, the way to reproduce it is:

❯ git clone https://github.com/ulysses4ever/cabal && cd cabal && git checkout adopt-ghc-9.4.1-win-fail
...
❯ cabal build cabal cabal-tests
... 
❯ cabal run cabal-tests -- --with-cabal="$(cabal list-bin cabal)" cabal-testsuite/PackageTests/ForeignLibs/setup.test.hs

(forward slashes should work on Windows, right? otherwise, change for backslashes in the test path...)

[ulysses4ever]

Oh my goodness... I think I solved it -- by updating cabal in the container with old GHC. Ready for review!

[ysangkok]

Why are some of the tests guarded by == 9.4.* and others by == 9.4.2? Seems weird to guess on which of these will be fixed in the next release.

[ulysses4ever]

@ysangkok no particular reason, I guess. What do you propose, 9.4.2 or 9.4.*? I can set whatever you think is better (I have no preference, personally).

[ysangkok]

Probably better to use 9.4.* and then revisit when the last 9.4 release comes out.

[ulysses4ever]

@ysangkok done!

[ulysses4ever]

@ysangkok since you took a look into it, maybe you'd consider approving it.

[ulysses4ever]

Thanks @Kleidukos!

---

I'm using merge-me instead of squash because I specifically tried to organize commits so that a) they were individually buildable and pass CI (though I haven't check :-)) and b) so that the Windows tests can re-enabled with a simple revert when #8451 is addressed.