Ormolu doesn't work in certain cases when comments contain $

[tomjaguarpaw]

Ormolu doesn't work in certain cases when comments contain $. To reproduce, run ormolu on a file containing

foo = ()
 -- $

(the space before -- is essential to demonstrate the bug). Ormolu says

The GHC parser (in Haddock mode) failed:
  ormolu.hs:2:2-5
  parse error on input `-- $'

I expected it to be able to parse this file successfully. It is a valid Haskell file. This is version

ormolu 0.0.5.0 UNKNOWN UNKNOWN
using ghc-lib-parser 8.10.1.20200412

[aspiwack]

The -- $ comment is parsed as a Haddock reference, I assume, which Haddock deems misplaced. There is really nothing Ormolu can do, since it needs to care about the placement of Haddock comments.

But you have workarounds: such as adding an extra space between -- and $, it should suffice.

[tomjaguarpaw]

Ah yes, that's true. That workaround is OK for now.

But why is Ormolu parsing in Haddock mode? Does it also format Haddock comments?

[mrkkrp]

Does it also format Haddock comments?

It may change how they are placed. For example -- ^ may be reformatted as -- | this is why we need to parse in Haddock mode.

[tomjaguarpaw]

Interesting, thanks. And thanks for writing Ormolu!

[aspiwack]

More importantly, Ormolu needs to understand Haddock comments, otherwise, they could easily become detached from the definition they refer to, hence change the generated html documentation (probably to something nonsensical). A problem which, I believe, early versions of Ormolu had.

The best way to understand Haddock comments is to share Haddock's parser. When you get there, it also gives you the freedom to manipulate them a smidge, which @mrkkrp is speaking of above.

[tomjaguarpaw]

I suspect I ought to consider my code incorrect if it doesn't satisfy Haddock. I don't actually have any Haddock docs in it, but it's probably a good habit to get into. On the other hand, I didn't realise that Haddock was actually part of GHC!