[Haskell-cafe] What's the story of this weird type?

[David Feuer]

I was reading an answer by Luis Casillas on Stack Overflow

http://stackoverflow.com/a/23418746/1477667

and he used a somewhat peculiar type, with apparently not-so-useful constraints:

newtype f :-> g =
    Natural { eta :: forall x. (Functor f, Functor g) => f x -> g x }

There are a couple things about it that I don't understand.

1. I can construct a :-> from a function that doesn't satisfy the
constraints, and I can pattern match to get access to it, but I can't
use it. Why am I allowed to construct it at all?

data Mb a = N | J a -- Not a Functor instance

-- This is accepted (why?)
pot = Natural (\case
  [] -> N
  (a:_) -> J a)

-- And I can even do this:
mb :: Functor Mb => Mb Int
mb = case pot of
       Natural f -> f [3]

But then of course I can't satisfy the constraint to use it.

2. I would have thought that such a weird/broken thing as this would
only be allowed with -XDatatypeContexts, but in fact it is allowed
without it. Why?

David