Add support for LambdaCase

[daoo]

Implemented support for the LambdaCase extension by adding a new expression constructor (LCase).

I have tested this with hlint, and it worked (!).

I'm eagerly waiting for comments/input on how to improve this patch.

[niklasbroberg]

Two comments, one easier to fix than the other.

First, there's a discrepancy in the parser vs the exact-printer. You treat the '' 'case' as two different lexemes (like GHC does), but the exact-printer treats them as a single "printeme". That means that the last of your commits should be un-done, and the exact-printer updated accordingly.

Test code:

foo = \    case { _ -> True }

Second, and much harder to fix, GHC allows LambdaCase to play with NonDecreasingIndentation. Implementing this requires that the lexer distinguishes between the case in a normal case and the case in a lambda case, which in turn requires that the lexer can inspect the previous token in the stream. That in turn requires rewriting the internals of the parser monad.

I would be happy to merge the patch even without support for this though, but we shouldn't forget that the problem isn't fixed.

[daoo]

I tried to come up with a relevant test case for your second point but was unsuccessful. I didn't get anything to work in GHC, maybe I've not understood the extension properly. Maybe it should be noted as a bug report as well, I could make one.

[ndmitchell]

Note that users in HLint are asking for this feature, so it would be useful.

[bergmark]

+1, this is important for stylish-haskell, verified that building it against this works.

[UnkindPartition]

Merging based on @niklasbroberg initial feedback and popular demand.

@niklasbroberg regarding your second point — please open a separate issue with a test case.

@daoo — thanks for your work and patience :)

[robstewart57]

Is there a timescale for uploading a fresh release to hackage that'll incorporate this merged commit? An increasing number of users (e.g. hlint users) may stumble into this problem as use of LambdaCase become more frequent. I don't think the 1.14.0.1 version on hackage reflects this commit.

[UnkindPartition]

Initially I was going to wait for #54 / #90 to get reviewed/merged in, but I don't have much free time now (and neither does @niklasbroberg), so maybe I should release what we have by now.

Hopefully will do it this weekend, then.

[mgsloan]

@feuerbach That's fine with me - feel free to ping me to rebase the changes in #90 atop whatever gets released.

[nh2]

Version 1.15.0.1 on Hackage contains the LCase constructor, but the README still claims to not support it.

Was it forgotten to update that?

[UnkindPartition]

Updated the readme.