@jckarter, do we have something tracking this already?

I can't find it, but this is a known design issue. The code is behaving correctly as currently designed—`U.Type` in the generic context represents a concrete metatype, so when you substitute `U = P` you get the concrete metatype for P, `P.Protocol`. This ends up asking "is S the type P", which is false, rather than "does S conform to P". A slightly different formulation works:

protocol P {}
struct S: P {}

func does<T, U>(_ x: T, extend _: U.Type) -> Bool {
  
  print(U.self) // P.Type
  
  return x is U
}

print(P.Type.self) // P.Type

S.self is P.Type // true
does(S.self, extend: P.Type.self) // false

@jckarter

A slightly different formulation works ...

Whatever you meant does(S.self, extend: P.Type.self) still returns false in your sketch (in the original environment). Though it now prints "P.Type" in the console twice, it doesn't really help with the practical application.

But it's good to know you are aware of this limitation.

Sorry, I updated the comment. The new version should print `true`. You have to provide `Protocol.Type` as a generic argument in whole in order to get the behavior you want.

@jckarter This time it really does what was expected. Thank a lot for this workaround!

Joe, what do you want to use this bug for? Diagnostics? Changing the interpretation?

It feels to me like a language design issue we should revisit at some point.