Omit more parens for wildcard type signature

[sergv]

This is a followup to #2764

I noticed that for cases like foo :: _ the GHC always prints • In the type signature:, but when part of a subtype it will print a stack of contexts ending with In the type signature: but without •. Also for the 'toplevel' case the type signature in the In the type signature bit will be normalized into foo :: _\n which can be checked for.

For instance this is the message when hole is not 'toplevel' - there's a stack of where the hole was located ending with In the type signature

source/library/Language/Haskell/Brittany/Internal.hs:130:20: error:
    • Found type wildcard ‘_’ standing for ‘GhcPs’
      To use the inferred type, enable PartialTypeSignatures
    • In the first argument of ‘HsDecl’, namely ‘_’
      In the type ‘HsDecl _’
      In the type signature: decl :: HsDecl _
    • Relevant bindings include
        hsmodDecls :: [LHsDecl GhcPs]
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:34)
        hsmodAnn :: EpAnn AnnsModule
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:24)
        m :: HsModule
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:13)
        splitAnnots :: HsModule -> ([LEpaComment], [LEpaComment])
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:1)
    |
130 |     decl :: HsDecl _
    |                    ^

And this is the error message for the 'toplevel' case - the line starting with • In the type signature also ends with :: _ which is easy to check and that's what this PR does.

source/library/Language/Haskell/Brittany/Internal.hs:131:13: error:
    • Found type wildcard ‘_’ standing for ‘HsDecl GhcPs’
      To use the inferred type, enable PartialTypeSignatures
    • In the type signature: decl :: _
      In an equation for ‘splitAnnots’:
          splitAnnots m@HsModule {hsmodAnn, hsmodDecls}
            = undefined
            where
                ann :: SrcSpanAnnA
                decl :: _
                L ann decl = head hsmodDecls
    • Relevant bindings include
        hsmodDecls :: [LHsDecl GhcPs]
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:34)
        hsmodAnn :: EpAnn AnnsModule
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:24)
        m :: HsModule
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:13)
        splitAnnots :: HsModule -> ([LEpaComment], [LEpaComment])
          (bound at source/library/Language/Haskell/Brittany/Internal.hs:127:1)
    |
131 |     decl :: _
    |             ^

Still not a general-purpose solution and a somewhat fragile heuristic that may break with newer GHC releases (but could be solved when GHC introduces something like structured errors). Yet it addresses a, what appears to me, pretty common case and sems to be reasonably simple.

[michaelpj]

LGTM, thanks for adding tests! Just a couple of doc comments.

[michaelpj]

Can you add a comment explaining what you explained in the PR description? Otherwise all that knowledge will get lost!

[sergv]

Added in a new commit after rebase.

[michaelpj]

Why does the prefix being empty here matter?

[sergv]

The idea is that breakOn didn't return any prefix if its pattern was not found. So the check is that the error message really contained • In the type signature: as a substring with the •.