Brandon S Allbery KF8NH wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 8/15/10 09:00 , Andrew Coppin wrote:
class (Vector (Point x)) => HasSpace x where type Point x :: *
(...)
And now things get *really* interesting. Consider this:
data Foo x = Foo !x !(Point x)
Surprisingly, GHC accepts this. This despite the rather obvious fact that "Point x" can exist if and only if "x" has a HasSpace instance. And yet, the type definition appears to say that "x" is simply an unconstrained type variable. Intriguing...
Maybe I'm missing something in all the type machinery I elided, but it looks to me like you have that backwards: HasSpace x requires Point x but not vice versa. Your actual usage may require the reverse association, but the definition of Foo won't be modified by that usage --- only applications of that definition.
Well, since Point is part of the definition of HasSpace, and therefore Point x is defined only if an instance HasSpace x exists. I'm not sure how that's "Point x doesn't require HasSpace x".