I see. The type inference algorithm requires type variables to betype constructors, not functions. Equality constraints aresimplified assuming that is the case. I knew that typefunctions had to be fully applied, but I didn't know that they alsocouldn't be "taken apart" by unification. This explains why thefollowing doesn't typecheck:type family T a :: *my_id :: f a -> f a; my_id = idx :: T a -> T a; x = my_idIMHO, this was not clear from the documentation or from the errormessage (It certainly _looks_ like "T a" should match "f a"...).Thanks for the explanation.-heatsink _________________________________________________________________ Hotmail: Trusted email with Microsoft’s powerful SPAM protection. http://clk.atdmt.com/GBL/go/201469226/direct/01/