ambiguity due to "maybe-rigid" rigid assoc type

[lcnr]

• cc https://github.com/rust-lang/trait-system-refactor-initiative/issues/89

trait Trait<'a> {
    type Assoc;
}

trait Outer<'a>{}

impl<'a, T: ?Sized> Outer<'a> for T
where
    T: Trait<'a>,
    T::Assoc: Trait<'a>,
{}

checking T::Assoc: Trait<'a>, we try to normalize the self type

• this tries to prove T: Trait<'a> either when checking for wf for coinduction(?), or currently in discard_impls_shadowed_by_env
  • two ParamEnv candidates, T: Trait<'a> and T::Assoc: Trait<'a>
    • T: Trait<'a> holds, but adds a region constraint due to uniquification
    • T::Assoc: Trait<'a> may hold if T::Assoc normalizes to T
    • this causes an inductive cycle

[compiler-errors]

fixing this will probably fix #109

[lcnr]

this does not affect a test which uses an impl as indirection, as we prefer param env candidates over impls

trait IntoIterator<'a> {
    type IntoIter;
}

trait Iterator<'a> {}
impl<'a, T: Iterator<'a>> IntoIterator<'a> for T {
    type IntoIter = T;
}

trait Outer<'a>{}

impl<'a, T: ?Sized> Outer<'a> for T
where
    T: IntoIterator<'a>,
    T::IntoIter: Iterator<'a>,
{}

[lcnr]

closing due to https://github.com/rust-lang/trait-system-refactor-initiative/issues/114#issuecomment-2123832503