[Haskell-cafe] Type families versus functional dependencies question

[Manuel M T Chakravarty]

Alexey Rodriguez:
> We are having trouble with the following program that uses type  
> families:
>
> > class Blah f a where
> >   blah :: a -> T f f a
>
> > class A f where
> >   type T f :: (* -> *) -> * -> *
>
> the following function does not type:
>
> > wrapper :: forall a f . Blah f a => a -> T f f a
> > wrapper x = blah x
>
> GHC gives the error:
>
>     Couldn't match expected type `T f1 f1 a'
>            against inferred type `T f f a'
>     In the expression: blah x
>     In the definition of `wrapper': wrapper x = blah x
>
> Maybe it is a problem with ambiguous types, namely "f" appears only  
> in applications of "T". But that is not the case, there is a "naked"  
> f appearing as the argument of "T f". But perhaps the type checker  
> does not want to unify those two f's precisely because they are the  
> arguments of "T f".

The problem is that blah's type is ambiguous, as f does only occur as  
an argument to the type family.  If you'd define

   class Blah f a where
     blah :: a -> f -> T f f a

(and change the rest of the program accordingly) then all will be  
fine.  See this thread for a more in-depth discussion of the problem:

   http://www.haskell.org/pipermail/haskell-cafe/2008-April/041385.html

Manuel