Inferred type too general, ignoring functional dependencies

Consider

module FloatFDs2 where

class C a b | a -> b where
  meth :: a -> b -> ()

data Ex where
  MkEx :: a -> Ex

f x = (\y -> case x of MkEx _ -> meth x y, \z -> case x of MkEx _ -> meth x z)

GHC infers f :: forall p1 p2. (C Ex p1, C Ex p2) => Ex -> (p1 -> (), p2 -> ()). That's silly: p1 and p2 should have been unified.

Solution is on the way in a branch I'm working on about a constraint solver refactor.