AST of input and AST of formatted code differ on certain empty comments

[nikita-volkov]

Formatting the following code:

data A = A
  {
    -- |
    -- 
    a :: Int
  }

Results in the following error:

@@ -14,8 +14,6 @@
  data A = A
-   {
-     -- |
-     -- 
+   { -- |
      a :: Int
    }

  AST of input and AST of formatted code differ.
    at ***
  Please, consider reporting the bug.
  To format anyway, use --unsafe.

Environment

• macOS 14.1.2
• ormolu 0.7.7.0 using ghc-lib-parser 9.10.1.20240511

[amesgen]

Thanks, I think this is related to the special handling of empty Haddock comments, also see #1068 and #1065 and the upstream ticket https://gitlab.haskell.org/ghc/ghc/-/issues/23935

[nikita-volkov]

Reproduced on:

ormolu 0.8.0.0 using ghc-lib-parser 9.12.1.20250105

[amesgen]

FTR the problem here is that in

data A = A
  {
    -- |
    -- 
    a :: Int
  }

there is a space after --. That causes the GHC parser to not insert DocD; the comment only shows up as non-Haddock AST comment (see #901 (comment) for context); but in the output, it does, so the AST differs.

I think fixing this (in a non-adhoc manner) will either require https://gitlab.haskell.org/ghc/ghc/-/issues/23935 or a change to how we format Haddock comments, cf #901 (comment).