I recently ran into a problem using kind polymorphism with GADTs. The answer there was to give a "complete user-specified kind" (CUSK) for my data type. I've read the relevant documentation, but I'm still getting essentially the same error when I try to apply that to a class.
Concretely, once I give a CUSK, the following does compile:
{-# LANGUAGE DataKinds, PolyKinds, GADTs #-}
data Foo (x :: k) where
C :: Foo x -> Foo '(x,x)
but when I move that definition to a class:
{-# LANGUAGE DataKinds, PolyKinds #-}
class Foo (f :: k -> *) where
foo :: (f :: k1 -> *) (x :: k1) -> (f :: (k1,k1) -> *) ('(x,x) :: (k1,k1))
I get the error:
• Expected kind ‘(k1, k1) -> *’, but ‘f’ has kind ‘k -> *’
• In the type signature:
foo :: (f :: k1 -> *) (x :: k1) -> (f :: (k1, k1) -> *) ('(x, x) :: (k1, k1))
In the class declaration for ‘Foo’
I expect there's something small I need to do to convince GHC that f
is kind-polymorphic in the second example.
This is a job for GHC 8 and TypeInType
, which allows much more amusing forms of dependency. The following minor edit of your code typechecks.
{-# LANGUAGE PolyKinds, RankNTypes, KindSignatures,
DataKinds, TypeInType #-}
module KP where
import Data.Kind
class Foo (f :: forall k. k -> *) where
foo :: (f :: k1 -> *) (x :: k1) -> (f :: (k1,k1) -> *) ('(x,x) :: (k1,k1))
Crucially, it is no longer a syntax error to use forall
in the (ha ha!) type of a type class parameter.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With