Monday, December 26, 2011

Showing Thrists Revisited

More than three years ago I lamented that it's impossible to to define a show function on thrists, generally, even when the element type is member of the Show class.

In the meantime I succeeded to declare certain thrist parametrizations and convince GHC to accept show instances for them. The Appli thrist is such an example.

With the upcoming GHC 7.4 things may improve considerably, as it will bring constraint kinds. Remember, thrists are (currently) parameterized like this (* → * → *) → * → * → *, i.e. on types, and with the new kind variable feature we can probably generalize to user-defined kinds. Then the parameters may match Ī©mega's: (a → a → *) → a → a → *. So we still result in a type, but we can choose our indices from a vastly bigger domain. Enter constraint kinds! These may be part of our user-defined kinds, so we can form stuff like (Show, *) and supply it for the parameter a. With some luck deriving Show can be more successful and not attached to particular parametrizations.

I regrettably still haven't gotten around building a GHC v7.4.1 candidate, so I cannot verify the above, but my gut feeling is that this'll work out...

1 comment:

Brent said...

I don't think this will work. Show is not a kind, it is a constraint (a type-like thing) whose kind is Constraint. So (Show,*) does not make sense. There is still no kind that means "types with a Show instance".