GHC suppresses too much kind information

GHC tends to hide important kind polymorphism in error messages and in GHCi. (I came across this when looking at #9017 (closed).) For example, try this:

bash$ cat Test.hs
{-# LANGUAGE PolyKinds #-}
module Test where
foo :: m a
foo = foo

bash$ ghci Test.hs
ghci> :t foo
foo :: m a
ghci> :i foo
foo :: m a

bash$ ghci Test.hs -XPolyKinds
ghci> :t foo
foo :: m a
ghci> :i foo
goo :: m a

bash$ ghci Test.hs -fprint-explicit-foralls -fprint-explicit-kinds
ghci> :t foo
foo :: foo :: forall (m :: * -> *) d. m d
ghci> :i foo
foo :: forall (k :: BOX) (m :: k -> *) (d :: k). m d

bash$ ghci Test.hs -XPolyKinds -fprint-explicit-foralls -fprint-explicit-kinds
ghci> :t foo
foo :: forall (k :: BOX) (m :: k -> *) (d :: k). m d
ghci> :i foo
foo :: forall (k :: BOX) (m :: k -> *) (d :: k). m d

Pretty confusing eh?

• Without -fprint-explicit-foralls -fprint-explicit-kinds you don't see the kinds on the types at all.
• Without -XPolyKinds in GHCi, when you say :t foo you are asking for the type of the expression foo (it could be an arbitrary expression). So foo is instantiated and then re-generalised; but without -XPolyKinds we don't get a poly-kinded type. Hence the difference in what is printed by :type and :info.

Here's a proposal: even without -fprint-explicit-foralls, we should print any foralls that bind a type variable whose kind includes a kind variable.

Trac metadata

Trac field	Value
Version	7.8.2
Type	FeatureRequest
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture