- User Since
- Oct 16 2018, 12:53 PM (34 w, 6 d)
Dec 17 2018
Regrettable changes can be observed in tests T12447, UncurryNamed, WildcardInstantiations, T11670,...to name but a few. I also saw this while I was running the testsuite:
Dec 10 2018
Dec 6 2018
Dec 5 2018
Dec 4 2018
ooops, somehow I can't add inline comment...But here is the failed outputs for:
Oct 18 2018
I see two possible solutions:
- Currently, ' is parsed as an empty list, which means it's type-checked as a list. This causes trouble for visible kind application, because lists expect to get their type from their elements. We could instead parse it as a ntgtycon by adding SIMPLEQUOTE '[' ']' as a production there and changing the list production from atype to require at least one type in the brackets. If ' is parsed into an HsTyVar, then type-checking will automatically work.
- Special-case the empty list in the type-checker.
I will special-case  in the type-checker.
I'm not sure which commit your haddock submodule corresponds to. You've put 5035a25c2020a04f124b05221b209c57403f1cab, but that doesn't appear to live anywhere in your fork of Haddock. (The latest commit on the wip/kind-app branch is c6b2f4a3bd96a28c7a2ec34108e4705f9b4a8d59, for comparison.)
Sorry @RyanGlScott, I didn't know how to respond to direct comments until now. 5035a25c2020a04f124b05221b209c57403f1cab is the head commit on my local wip/kind-app after I rebased my utils/haddock to origin/ghc-head, so technically my local utils/haddock has diverged from my github fork. I don't see why it would matter, but seems like it does. I'm not very good with git and would appreciate any help with this!
- Even though it's possible to use the value-level empty list constructor  with visible type applications (e.g.,  @Bool), it's impossible to do the same with the promoted empty list constructor:
Thanks for the catch! I'll go with adding a special case for  in the typechecker, although I agree this is not the most elegant solution.
Oct 17 2018
My rebase last time was missing some commits