shadowed trait in a generic argument causes incorrect name resolution.

[jhgg]

Consider this code:

trait A {
    fn woop(&self);
}
trait B {
    const X: usize;
    fn woop(&self);
}

impl<A> B for A
where
    A: B,
{
    const X: usize = 1;

    fn woop(&self) {
        println!("{}", A::X);
        //             ^ this is considered as `trait A` not the generic param `A` by r-a.
    }
}

fn main() {}

See:

This also impacts rename, which gets a tad confused:

Rustc however does resolve this properly and the program provided compiles fine.

[lnicola]

The might be two issues here:

• we don't support const generics much outside of parsing
• this still happens if you replace the const with an associated type, which may or may not be Associated Type Refs of qualified paths dont resolve #8282

[jhgg]

• we don't support const generics much outside of parsing

This occurs if you were to do:

     fn woop(&self) {
        println!("{}", A::woot());
        //             ^ this is considered as `trait A` not the generic param `A` by r-a.
    }

As well, so probably not related to const generic 🤔 I believe the issue here is neither (but I'm no expert.) It looks to prefer to resolve A in that context to the trait A rather than the generic parameter A (from impl<A> B for A).

[flodiebold]

A::X gets correctly resolved, so this seems to be a problem purely in the IDE layer name resolution.