Ah... I see that I have a bug in my proposal, perhaps corrected below.
2. In the second and fourth instances, the type variable x appears twice in the parameters of the type function built from the fundep (a b -> c). This causes an error. If I try adding (x ~ x') to the context and making one of them x' instead of x, I get different errors.
I see the problem with the x parameter now:
I'm starting to this this is an example of using functional dependencies that has no translation to type families. Can anyone free me of such thoughts?
Code with fundeps is:
data Var a = Var Int (Maybe String) data RHS a b = RHS a b
class Action t a b c | a -> t, a b -> c, a c -> b instance Action t () y y instance Action t (Var x) (x -> y) y instance Action t [t] y y instance (Action t a b c) => Action t (RHS (Var x) a) (x -> b) c instance (Action t a b c) => Action t (RHS [t] a) b c
I have never used associated types or type families. I assume you cannot rely on GHC lazy goodwill for the a->t constraint. You will have to be explicit. Not using associated types but type families I think the syntax should be:
[snip]
type instance A_C_B (RHS (Var x) a) c = x -> A_C_B a c
The above is the corrected line. I had forgotten the (x->) on the RHS.
[snip]