Overview:

• There a many minor changes in the GHC AST in GHC 9.0:
  • The new module hierarchy is mostly a mechanical translation.
  • The extension constructors now have NoExtCon as a strict field, and the GHC 8.8+ pattern match coverage checker is clever enough to detect that we don't have to match it (no more noExtCon calls).
  • Some unreachable constructs (e.g. ConPatOut) removed (less notImplemented calls)
  • The RealSrcLoc constructor of SrcLoc now has an extra field Maybe BufPos which indicates the raw location of a syntactic construct (so it ignores e.g. LINE pragmas). We simply ignore this field.
  • Due to parsing changes in whitespace sensitivity, some code that is valid in older GHC versions will no longer be able to be parsed.
    As a side effect, we can enable TypeApplications be default, (Guess that TypeApplications is used without being enabled #452) as I am not aware of any code snippets where parsing with or without TypeApplications would make any difference.
  • Some language extensions got added or changed:
    • LambdaCase can now be used in the proc notation of Arrows src
    • QualifiedDo is enabled by default
    • LinearTypes and LexicalNegation are disabled by default.
• Previously, ormolize was applied to the haskell packages of the current GHC version. Some packages have different versions at different GHC versions (hlint, lens), so we now use the haskell packages of 8.10.4 for ormelization or all GHC versions.

Two questions to be resolved:

1. The leksah test revealed a haddock edge case:

   -- |
   --
   ---
   module Test where

   gets formatted (both with and without this PR) to

   ---
   
   -- |
   module Test where

   but with this PR (due to parser changes), the AST is no longer considered to be equal (so formatting fails without -u).
2. This snippet

   test ::
     -- |
     a -> 
     b

   was a fixed point before this PR, but now it is formatted to

   test ::
     a ->
     b

   and this works even without -u, as empty haddock strings no longer appear in the AST.

I am not sure whether these are considered to be bugs, and if so, whether they should block this PR. EDIT see this comment for an update.

Update to the questions above: We decided that 2. is OK, and created #726 to track 1.

Question: I'm running ghc-8.10.4 due to issues with 9.0.1 in the plugins I'm writing. Should I stay on 0.1.4.1, or is it OK to switch to 0.2?

@Tritlo ormolu-0.2.0.0 should compile just fine on GHC 8.10.x; it is also tested on CI.

Ok thanks! I just saw the lib-parser scroll by for 9.0.1, and was worried it might be inconsistent.