LSP completion tests seem to be flaky

[cocreature]

The LSP completion tests seem to fail occasionally with the following rather scary looking error:

    completion on stale file:                      FAIL (3.46s)
      external/ghcide_ghc_lib/test/src/Development/IDE/Test.hs:35:
      Could not find (DsError,(3,0),"Parse error") in List [Diagnostic {_range = Range {_start = Position {_line = 0, _character = 0}, _end = Position {_line = 100000, _character = 0}}, _severity = Just DsError, _code = Nothing, _source = Just "compile", _message = "Compilation Issue\nlookupGlobal\n\nCan't find interface-file declaration for variable mkInteger\n  Probable cause: bug in .hi-boot file, or inconsistent .hi file\n  Use -ddump-if-trace to get an idea of which file caused the error\nCall stack:\n    CallStack (from HasCallStack):\n      callStackDoc, called at compiler/utils/Outputable.hs:1164:37 in ghc-lib-parser-8.8.1.20200115:Outputable\n      pprPanic, called at compiler/typecheck/TcEnv.hs:133:32 in ghc-lib-8.8.1.20200115:TcEnv\nPlease report this bug to the compiler authors.\n", _relatedInformation = Nothing},Diagnostic {_range = Range {_start = Position {_line = 3, _character = 0}, _end = Position {_line = 3, _character = 14}}, _severity = Just DsWarning, _code = Nothing, _source = Just "typecheck", _message = "/tmp/extra-dir-15084968049022/Foo.daml:4:1: warning:\n    The import of \8216DA.List\8217 is redundant\n      except perhaps to import instances from \8216DA.List\8217\n    To import instances alone, use: import DA.List()", _relatedInformation = Nothing}]

I have absolutely no clue what is going wrong here. So far I have only seen this on CI and never locally.

cc @aherrmann-da (just want to make you aware of it)

[cocreature]

Judging from the error, it seems to be coming from compileModule (since that seems to be the only thing that uses "compile" as the stage). Given that completion only relies on the typechecked module but not the core, I don’t really understand why this seems to be the only test with this problem.

[cocreature]

It looks like the code trying to lookup mkInteger probably comes from CorePrep (presumably due to RebindableSyntax) https://gitlab.haskell.org/ghc/ghc/blob/77f1d11a9ed42eda149a45786294253b4853fce7/compiler/coreSyn/CorePrep.hs#L1533 so that seems consistent with the problem being caused by core generation but I still have no idea why.