#10114: Kind mismatches with AnyK in rank-2 types -------------------------------------+------------------------------------- Reporter: cam | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.4 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: GHC rejects | Unknown/Multiple valid program | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Revisions: -------------------------------------+------------------------------------- Changes (by goldfire): * cc: sweirich@… (added) Comment: Replying to [comment:5 simonpj]:
So I think I retract my claim in comment:1. Perhaps we do want to generalise "inside". I agree with your (now retracted) claim.
If `T = forall k1 k2 (f :: k1 -> k2) (a :: k1). f a`, then I think `T`'s type would have to be `k2`, which is nonsense. As I understand it, the kind of a forall-type is the kind of the body of the forall-type.
Can anyone say why it is "generally more common to restrict the body of
a forall to have kind `*`"? I can't articulate a clear argument. This claim of mine is direct from Stephanie. Stephanie, do you know why this is?
In GHC's case, a good reason not to do this was types like `forall a. a
-> (# a, a #)`, but levity polymorpism will help here. I don't think that's quite what you want to say, because the kind of the body of that forall is `*`! Maybe `forall a. (# a, a #)`? I don't see how levity polymorphism fixes this problem. Even with levity polymorphism, the kind of the body of that last forall is still `#`. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/10114#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler