"D x", for an x that is not an instance of C, is still inhabited by "undefined". Additionally, on the implementation side, the dictionary C is not included inside of a D, so you still need to pass it in to call m; a function

test :: C a => D a -> Bool gets translated in Core into a system F type like this: data Dict_C a = Dict_C { m :: D a -> Bool } test :: forall a. Dict_C a -> D a -> Bool

-- ryan On Fri, Oct 8, 2010 at 4:55 AM, Eric Walkingshaw wrote:

The following code compiles happily in GHC:

...

{-# LANGUAGE TypeFamilies #-}

class C a where   data D a   m :: D a -> Bool

test :: C a => D a -> Bool test = m

My question is why do I need the context in the function "test"?  It seems like since "D" is associated with class "C", the compiler can safely assume that any time I have a "D a", "a" must be an instance of C.  But GHC complains if the context is removed.

At first I thought that maybe the associated type was just syntactic sugar for a non-associated data type family.  But this doesn't seem to be the case since I cannot instantiate D outside of an instance of C. Google and the type family documentation provided no other leads.

Am I missing something here?  Is the context assumption invalid?  Or is it just an assumption that GHC doesn't make (yet)?

Thanks in advance for any replies.

-Eric _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe