While handling associated types in rust-analyzer, I ran into two problems that I want to ask about. These are probably not related, so I can split this up if necessary, but I wanted to discuss them a bit first 🙂

Number 1:
(moved to #235)

Number 2:

#[test]
fn projection_equality() {
    test! {
        program {
            trait Trait1 {
                type Type;
            }
            trait Trait2<T> { }
            impl<T, U> Trait2<T> for U where U: Trait1<Type = T> {}

            struct u32 {}
            struct S {}
            impl Trait1 for S {
                type Type = u32;
            }
        }

        goal {
            exists<U> {
                S: Trait2<U>
            }
        } yields {
            "Unique"
        }
    }
}

I'd expect this to pass, but it yields ambiguity, I think because the ProjectionEq is ambiguous between u32 and Trait1::Type<S>. But those are really the same type. I found #74, but it's not clear to me how this kind of thing is supposed to work with ProjectionEq. #111 is maybe related?