HLS failure with GHC 9.14 and ExplicitLevelImports

[guibou]

I turned LANGUAGE ExplicitLevelImports ON on my codebase (1200+ haskell modules) and got it building with GHC 9.14.

However HLS fails on some modules (repro is not obvious) with:

Diagnostics:
Compilation Issue
expectJust

Call stack:
    CallStack (from HasCallStack):
      pprPanic, called at compiler/GHC/Data/Maybe.hs:77:19 in ghc-9.14.1-05a0:GHC.Data.Maybe
      expectJustError, called at compiler/GHC/Data/Maybe.hs:74:24 in ghc-9.14.1-05a0:GHC.Data.Maybe
      expectJust, called at compiler/GHC/Unit/Module/Graph.hs:579:8 in ghc-9.14.1-05a0:GHC.Unit.Module.Graph
Please report this bug to the compiler authors.

I bet that this is linked to the custom graph logic that HLS does for template haskell code. Maybe this custom logic could be simplified in the context of ExplicitLevelImports (e.g. GHC claims to be better at tracking dependencies for template haskell).

Your environment

Which OS do you use? Nixos
Which version of GHC do you use and how did you install it? GHC 9.14 from nixpkgs
How is your project built (alternative: link to the project)? Custom nix builder

Which LSP client (editor/plugin) do you use? neovim
Which version of HLS do you use and how did you install it? 2.13.0.0+some custom patchs
Have you configured HLS in any way (especially: a hie.yaml file)? using cmd line flag

Steps to reproduce

Sorry, the following is left empty until I find a proper way to reproduce. It does seem to only appear on module which depends themselfs on module which does have template haskell.

Expected behaviour

HLS should typecheck correctly.

Actual behaviour

HLS fails

Debug information

[fendor]

Thank you for the bug report!

I observed similar issues on the ghc-debugger codebase

cc @wz1000

[fendor]

closed by #4846

[guibou]

Hi @fendor.

I've retried today HLS with ExplicitLevelImports (I was blocked at work by ormolu tweag/ormolu#1192) and hit the same problem again.

Nothing to be alarmed yet, I think HLS in fine with GHC 9.14.1, but you may be aware that I'm using a custom build of GHC, especially including this wonderful fix for :load and :add, https://gitlab.haskell.org/ghc/ghc/-/merge_requests/15980

Nothing to be alarmed yet, I had not yet took the time to confirm that the origin in the regression is in GHC, or in the way I rebased the above merge request on top of my GHC, or if the problem comes from my custom HLS build, ...

I'll try to confirm that in the following days, but in the meantime, I wanted to notify you, in case that's something obvious.

Here is the error I'm observing in HLS:

Compilation Issue
expectJust

Call stack:
    CallStack (from HasCallStack):
      pprPanic, called at compiler/GHC/Data/Maybe.hs:77:19 in ghc-9.14.1-0837:GHC.Data.Maybe
      expectJustError, called at compiler/GHC/Data/Maybe.hs:74:24 in ghc-9.14.1-0837:GHC.Data.Maybe
      expectJust, called at compiler/GHC/Unit/Module/Graph.hs:605:8 in ghc-9.14.1-0837:GHC.Unit.Module.Graph
Please report this bug to the compiler authors.

And indeed the GHC.Unit.Module.Graph module was heavily changed in the GHC MR reference above AND was the biggest source of conflict when I rebased on top of GHC 9.14.1, so 99.9% chance that the problem comes from a mistake in my rebase.