@swift-ci create

The example you posted now type checks on master, as long as the typealiases are present. Fixing associated type inference to infer these properly seems like a separate issue. I think we have trouble looking through the superclass constraint.

I think the feature you're asking for is pretty tricky. For example, consider this:

class G<T, U> {}
class Sub : G<Int, String> {}

protocol P {
  associatedtype A : G<X, Y>
  associatedtype X
  associatedtype Y

  func foo() -> A
}

struct S : P {
  func foo() -> Sub {}
}

We'd have to infer that A := Sub, and additionally X := Int, Y := String. That's a pretty complicated inference rule. I feel like it might only introduce more problems with circular decl checking and the like.