callCC type signature?

[jqyu]

Pardon my ignorance if I'm misunderstanding what a Cont transformer is, but is there a reason why the callCC type signature is

callCC :: forall a b. ((a -> m b) -> m a) -> m a

Rather than the more general:

callCC :: forall a. (forall b. (a -> m b) -> m a) -> m a

I recently tried to write some code that looked like:

value = callCC \return -> do
  foo <- foo' `fallBackTo` return defaultValue
  bar <- bar' `fallBackTo` return defaultValue
  fn foo bar

Which failed to type check because foo and bar had different types, so the signature of return wasn't happy.

[garyb]

I think it's probably fine to move the forall up like that, since the use of b here is only in the return type anyway (because we don't care about the result value).

It's probably the way it is now because it's the same in Haskell, and a lot of stuff was borrowed or straight lifted from there at first. I assume it's this way in Haskell to avoid the need of -XRank2Types/-XRankNTypes.

[jqyu]

Hmm, tried actually implementing it and it's much easier said than done. Although the instance declarations for the other monad transformers like ExceptT, MaybeT, StateT, and WriterT should work as-is, they fail to typecheck.

The liftCallCC functions are pretty hard to translate using rank 2 types. Going to have to put some more thought into this before submitting a PR

Whoops, looks like I had the signature wrong! I should've suggested:

callCC :: forall a. ((forall b. a -> m b) -> m a) -> m a

That worked without a hitch.
PR here: #74