[GHC] #11519: Inferring non-tau kinds

[GHC]

#11519: Inferring non-tau kinds
-------------------------------------+-------------------------------------
        Reporter:  goldfire          |                Owner:
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by goldfire):

 Replying to [comment:2 simonpj]:
 > Yes I think that using `ExpType`s is exactly the right thing here.
 Interestingly we used to have a variant of `ExpType`, called `ExpKind`:
 > {{{
 > data ExpKind = EK TcKind (TcKind -> SDoc)
 > }}}
 > but that has gone now.  I'm not sure how you managed to keep the error
 messages good without that `SDoc` argument, but I'm all for the
 simplification.

 That was actually quite a bit of work. All the kind errors used to be
 reported eagerly, during constraint generation. Thus the `SDoc` argument
 came into play. Now kind errors are naturally reported from !TcErrors, so
 the `ExpKind` stuff doesn't work. Instead, look at `mkExpectedActualMsg`,
 which re-creates the old messages. You might notice that the function is a
 terrible mess.... I would love to find a theory or specification of error
 messages against which we can write an implementation. Because !TcErrors
 is inscrutable. I made it considerably worse, but it wasn't in that good a
 state when I arrived. In any case, that's where the error messages have
 gone.

 >
 > So now perhaps `ExpType` will do just fine.
 >
 > I also agree with Adam about explicit quantification.  (Both are
 independently useful.)  And in fact I think that the declaration he give
 is accepted already -- an entirely un-documented feature.   Adding some
 docs for it would be good.

 I actually seem to recall implementing something along these lines once
 upon a time. But the following full module doesn't work for me:

 {{{
 {-# LANGUAGE RankNTypes, TypeInType, GADTs, ConstraintKinds #-}

 module Bug where

 import GHC.Exts
 import Data.Kind

 data TypeRep (a :: k) -- abstract

 data TypeRepX :: (forall k. k -> Constraint) -> Type where
   TypeRepX :: forall k (c :: forall k'. k' -> Constraint) (a :: k).
               c a => TypeRep a -> TypeRepX c

 instance Show (TypeRep a)  -- elsewhere

 instance forall (c :: forall k. k -> Constraint). Show (TypeRepX c) where
   show (TypeRepX tr) = show t
 }}}

 Is there something I'm missing? I haven't looked at the GHC code for
 instance declarations.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11519#comment:3>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler