[Haskell-cafe] Failed inference: maybe bug?

[Edward Z. Yang]

On the prompting of napping, I humbly submit the following code to haskell-cafe:

    ezyang at javelin:~/Dev/haskell/generic-typeclass$ cat > Bar.hs
    {-# LANGUAGE TypeFamilies #-}
    {-# LANGUAGE MultiParamTypeClasses #-}
    {-# LANGUAGE FlexibleContexts #-}

    data Foo b = Foo
    class FooTwo m b where

    type family Ctor a :: (* -> *)
    type instance Ctor (f a) = f

    type family CtorVal a :: *
    type instance CtorVal (f a) = a

    instance FooTwo Foo Int where

    class FooTwo (Ctor mb) (CtorVal mb) => FooOne mb where

    foo :: FooOne (m Int) => m Int
    foo = fooGeneric

    fooGeneric :: FooTwo m b => m b
    fooGeneric = undefined
    ezyang at javelin:~/Dev/haskell/generic-typeclass$ ghci Bar.hs 
    GHCi, version 6.12.1: http://www.haskell.org/ghc/  :? for help
    Loading package ghc-prim ... linking ... done.
    Loading package integer-gmp ... linking ... done.
    Loading package base ... linking ... done.
    [1 of 1] Compiling Main             ( Bar.hs, interpreted )

    Bar.hs:19:6:
        Could not deduce (FooTwo m Int) from the context (FooOne (m Int))
          arising from a use of `fooGeneric' at Bar.hs:19:6-15
        Possible fix:
          add (FooTwo m Int) to the context of the type signature for `foo'
          or add an instance declaration for (FooTwo m Int)
        In the expression: fooGeneric
        In the definition of `foo': foo = fooGeneric
    Failed, modules loaded: none.
    Prelude> 
    Leaving GHCi.

It seems that GHC is unable to unify the two instances.  So, bug or expected
behavior?

Cheers,
Edward