Replying slap-foreheadedly to own post... On 15 May 2008, at 11:56, Conor McBride wrote:
Folks
I'm also wondering whether it makes sense to have a "bottomless Top" type, with constructor _ and lazy pattern _ (with (undefined :: Top) equal to _). Then the constant-time shape operator makes the same sort of sense as the constant-time
inflate :: Functor f => f Zero -> f a
We've got it already. data One would do, with smart constructor only :: One only = undefined and no other operations, so no way to be strict in its values. It's the lazy version of Zero. Assuming (safely?) that representation of data in each instance of a type constructor is uniform, that means shape :: Functor f => f a -> f One shape = unsafeCoerce should be ok, right? The question about the GC implications of that move still stands, though. Cheers Conor