Hello Hans, You can move the setFoo into a multiparameter type class, to which you can add that constraint: class (HasFoo a, HasFoo b, b ~ NoFoo a (Foo b)) => SetFoo a b where setFoo :: Foo b -> a -> b instance (HasFoo b, b ~ NoFoo (FooT f a) (Foo b)) => SetFoo (FooT f a) b where setFoo f (FooT _ x) = FooT f x In my opinion, that's not as nice as the fundep solution which has less names: class Has s t a b | s -> a, t -> b, s b -> t, t a -> s On the topic of defining setFoo in terms of setFoo', you might be interested in -XDefaultSignatures, which can work with the SetFoo class defined above: class HasFoo a where default setFoo' :: SetFoo a a => Foo a -> a -> a setFoo' = setFoo setFoo' :: Foo a -> a -> a Finally, there seems to be some overlap with generic programming. Changing type parameters isn't supported by the ones I know (syb), but you can still fake it: http://www.haskell.org/haskellwiki/SYB#fmap. Regards, Adam On Mon, Dec 16, 2013 at 7:37 PM, Hans Höglund wrote:

Hello,

I am working with a set of type classes of the following form (http://lpaste.net/97110). The idea is that every such class provide an associated type Foo, and a lens to the Foo in every instance. I.e. this class is used to provide view/set/modify for all types that contain a Foo somewhere deep in its structure.

For "simple lenses", i.e. functions that does not modify the associated Foo, this is straightforward. However to support polymorphic updates it seems necessary to add another associated type NoFoo, which must be used to constraint the return type of set.

What bothers me is the redundancy of the two set functions. I would intuitively expect set' to be implemented in terms of set (as it seems to be a restriction of that function), but this is not possible, as the compiler can not deduce that (NoFoo a (Foo a) ~ a). Is there a way to add this constraint to the type class?

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