Fix the Bifoldable instance of Coyoneda

[Topsii]

Coyoneda's Foldable instance currently requires Bifunctor in its context. The plan is for Foldable to become a superclass of Bifoldable, see haskell/core-libraries-committee#93. If this happens, then Coyoneda's Bifoldable instance will require Bifunctor, because it is a requirement of its superclass Foldable.

[Topsii]

Currently the Foldable and Traversable instances of Coyoneda are implemented using bimap. The use of bimap requires the context Bifunctor p in Coyoneda p's Bifoldable (and Foldable) instance. It is possible to implement Foldable and Traversable without bimap as follows:

instance (forall x. Foldable (p x)) => Foldable (Coyoneda p a) where
  foldMap = \f (Coyoneda _ yb pxy) -> foldMap (f . yb) pxy

instance (forall x. Traversable (p x)) => Traversable (Coyoneda p a) where
  traverse = \f (Coyoneda xa yb pxy) -> Coyoneda xa id <$> traverse (f . yb) pxy

To me this appears to match their Bifoldable/Bitraversable instances more closely. Note that the Bifoldable instance does not require Bifunctor p in its context in this case.

instance Bifoldable p => Bifoldable (Coyoneda p) where
  bifoldMap = \f g (Coyoneda h i p) -> bifoldMap (f . h) (g . i) p

Unfortunately the use of quantified constraints makes the instance contexts more restrictive than they were before.

[RyanGlScott]

Sorry for not noticing this earlier!

@ekmett, do you have an opinion on whether to:

1. Update the Bifoldable instance to instance (Bifoldable p, Bifunctor p) => Bifoldable (Coyoneda p), as is done in this patch, or
2. Changing the Foldable instance to instance (forall x. Foldable (p x)) => Foldable (Coyoneda p a), which would avoid needing to change the Bifoldable instance's context

?

[ekmett]

If we can get Bifunctor, etc. changed in base to have quantified superclass constraints, then I'd be all about getting these instances changed to use quantified constraints, as it'd be almost a pure win.