Accept a leading single quote for data constructor promotion

[andrewthad]

Resolves #12

[RyanGlScott]

Thanks for doing this, @andrewthad.

Question: have you tested these changes with a promoted data constructor that itself ends with a single quote? I ask since there's a special lexical trick you have to employ to promote those:

{-# LANGUAGE DataKinds #-}
{-# LANGUAGE GADTs     #-}

data Foo = Foo' | Bar

data Indexed ix where
  Indexed :: Indexed ' Foo'

main :: IO ()
main = print $ #size int

Now that I take a look, it appears that all hsc2hs tests are within GHC itself, not within this repo, which is a bit of shame. Would you mind adding some tests for this feature after this patch lands?

Some comments inline.

[andrewthad]

I cannot figure out how to reply to a review, so I'll do it here. Good point about:

data Indexed ix where
  Indexed :: Indexed ' Foo'

I had never encountered that before. I've made another tweak, it should now handle this case correctly. I've tested it out locally. I'll get the sample code I've added in the note into GHC's test suite whenever GHC bumps to using a version of hsc2hs that has this change.

[RyanGlScott]

Thanks, this is looking clearer to me. Some more questions/comments inline.

[RyanGlScott]

Might there be other whitespace characters besides (e.g., Unicode whitespace) that we'd have to consider? Similarly, is isAlphaNum sufficient to parse everything, or would Unicode characters add an additional wrinkle to this?

Disclaimer: I never use Unicode when writing Haskell, so I have no idea if this is a well formed thought. But I'm sure that if we get this wrong, someone who //does// use Unicode will complain about this at some point :)

[andrewthad]

I hope that isAlphaNum is sufficient for all of the characters that could show up in an identifier. It’s what this library already uses in other places when it’s parsing identifiers. (So, if it didn’t work, I suspect that someone else would have already complained about it since this library would be broken for non-ascii content). But, I think that it is correct.

The question about alternative space characters is a good one. I’m not familiar with what GHC accepts here. I suspect that GHC probably accepts any number of white space characters of any kind. So, I think I’m currently doing the wrong thing right here.

[andrewthad]

I've addressed this to the extent that my knowledge permits. I've assumed that in source haskell, anything that isSpace returns True for is considered whitespace.

[RyanGlScott]

Great! This patch looks to be in great shape. The only other thing that would need to be changed is the changelog.md file. If you feel up to it, feel free to update this yourself (under the version number next); otherwise, we'll do so ourselves before the next release.

Can anyone else (@hvr?) think of any potential issues with this patch? If not, I'll merge this some time soon(-ish).

[andrewthad]

I've added a note in the changelog. Thanks for all the prompt feedback!

[RyanGlScott]

I went ahead and merged this. @andrewthad, would you be willing to open up a MR for GHC that:

1. Bumps the hsc2hs submodule to use this commit
2. Add some test cases for this

If we do that, I think we can close Trac #11004 for good. Thanks!