Parentheses between some single class constraints conflict with hlint rules

[arybczak]

As pointed in #264 (comment), the change introduced with #975 made it so that now specific code formatted with ormolu causes hlint to complain about redundant parentheses.

This is quite inconvenient and the workaround described in the above comment, i.e. manually adding hlint annotations is not very satisfying.

Would you consider (partially) reverting this change? I think hlint only complains about parentheses around constraints with no type parameters (like HasCallStack), so at least these could be made to be formatted with no parentheses.

[amesgen]

Thanks for raising a separate issue for this. There also is a potential alternative of tweaking hlint not to warn about these redundant parentheses at the constraint level (or adding an option for that behavior), as suggested in #264 (comment). WDYT?

[mrkkrp]

I think we should probably keep parentheses in all cases, since the original motivation for adding parentheses was that like this it is easier to add new constraints. This still holds no matter what kind of single constraint we have.

[arybczak]

WDYT?

I'm fine with any solution that doesn't make ormolu and hlint conflict.

[amesgen]

Re #1003 (comment): I opened ndmitchell/hlint#1503.

[tbidne]

Incidentally, I wonder if there is any interest in adding hlint to the CI here? It could help mitigate such conflicts in the future (or make it less likely people are caught off guard, at least).

I experimented adding it here and it was very easy. Sadly this would not have caught the current issue, as it appears the ormolu code base does not have any such single class constraints. Still, seems beneficial to me (+ the benefit of hlint itself).

[mrkkrp]

IMO about 20-30% of HLint suggestions degrade readability of code (e.g. its obsession with writing everything as concisely as possible eliminating case expressions), so if we go that path, it would need quite a bit of tweaking.

[ocharles]

While I agree, with a big codebase it doesn't take long to build a good ruleset. I think it's a net win

[amesgen]

ndmitchell/hlint#1503 has been resolved; hlint does no longer warn about parentheses around class constraints in all cases. We could close this issue now, or wait until there is an hlint release with the relevant fix.

[mrkkrp]

I think we can close. Thanks for opening that issue!