impl vs param_env candidates in normalization

[aliemjay]

While investigating rust-lang/rust#112583, it seems that it reveals a bug in the new solver rather than being fixed by it. The new solver accepts this:

trait Trait {
    type Ty;
}

impl<T> Trait for T {
    type Ty = ();
}

fn test<T: Trait>() {
    let _: <T as Trait>::Ty = ();
}

It uses the impl candidate for normalization instead of the ParamEnv candidate.

This means that we're using a different candidate when solving T: Trait than when normalizing <T as Trait>::Ty. I don't think it is unsound but it can lead to some surprising behavior like this requiring 'a == 'static despite using the impl candidate for normalization:

trait Trait<'a> {
    type Ty;
}

impl<T> Trait<'_> for T {
    type Ty = ();
}

fn test<'a, T: Trait<'static>>() {
    let _: <T as Trait<'a>>::Ty = ();
    //~^ ERROR lifetime may not live long enough
}

[compiler-errors]

This is kinda related to #1 since the new solver totally doesn't even consider "rigid" associated types as valid normalization candidates

[compiler-errors]

Also see #12

[aliemjay]

Oh, this is clearly a duplicate of #12.