Problem using function traits with associated types

[ghost]

Here's a small example that one might expect to work but doesn't:

trait T0 { type O; }

struct S<A>(A);
impl<A> T0 for S<A> { type O = A; }

trait T1: T0 {
    // this looks okay but as we see below, `f` is unusable
    fn m0<F: Fn(<Self as T0>::O) -> bool>(self, f: F) -> bool;
}

// // complains about the bounds on F here not being required by the trait
// impl<A> T1 for S<A> {
//     fn m0<F: Fn(A) -> bool>(self, f: F) -> bool { f(self.0) }
// }

// // complains about mismatched types: <S<A> as T0>::O vs. A
// impl<A> T1 for S<A>
// {
//     fn m0<F: Fn(<Self as T0>::O) -> bool>(self, f: F) -> bool { f(self.0) }
// }

[ghost]

Interestingly, this example does work if you move the bound on F to a where clause:

trait T1: T0 {
    fn m0<F>(self, f: F) -> bool where F: Fn(<Self as T0>::O) -> bool;
}

impl<A> T1 for S<A> {
    fn m0<F>(self, f: F) -> bool where F: Fn(A) -> bool { f(self.0) }
}

[jroesch]

@darinmorrison I would expect this to be another sad consequence of handling the bounds placed on type parameters differently then the ones placed in where clauses. @nikomatsakis is helping me close down #20020 which should ideally fix this bug.

[nikomatsakis]

This seems to work now.

[ghost]

Awesome, thanks!