Move ccase_SList into All class

[ryantrinkle]

This allows us to provide more type information to the continuations.

Specifically: cpara_SList doesn't allow the cons continuation to know that there's a connection between (y ': ys) and xs. With this change, ccase_SList does provide that connection, i.e. that (y ': ys) ~ xs.

I'm not sure whether this is a breaking change or not: it should provide strictly more info to the continuations, and I don't think it's sensible for anyone to be defining their own instances of All.

[kosmikus]

Thanks for the PR, @ryantrinkle . I'm currently still a bit reluctant to just merge, because I'd like to understand the underlying issue. To me, cpara_SList is one of the most general functions there are and if ccase_SList after your change can no longer be expressed in terms of cpara_SList, then I'd think that cpara_SList should be changed. Also, is it possible for you to give an example of a definition that requires this change?

[kosmikus]

Ok, I don't think ccase_SList has to be moved into the class, as I think it can be expressed in terms of cpara_SList as follows:

newtype CCase xs r ys = MkCCase { unCCase :: (xs ~ ys) => r ys }

ccase_SList ::
     All c xs
  => proxy c
  -> ((xs ~ '[]) => r '[])
  -> (forall y ys . (xs ~ (y : ys), c y, All c ys) => r (y ': ys))
  -> r xs
ccase_SList p nil cons =
  unCCase (cpara_SList p (MkCCase nil) (const (MkCCase cons)))
{-# INLINE ccase_SList #-}

I would furthermore think that it's then even sufficient to change the type of ccase_SList to

ccase_SList ::
     All c xs
  => proxy c
  -> ((xs ~ '[]) => r '[])
  -> (forall y ys . (xs ~ (y : ys)) => r (y ': ys))
  -> r xs