Higher-ranked trait bounds on associated types are not elaborated

[shepmaster]

This code works:

trait Boring {}

trait Usage
where
    Self::Thing: Boring,
{
    type Thing;
}

fn value_example<T: Usage>(t: T::Thing) {
    use_the_trait(t);
}

fn use_the_trait<T: Boring>(_: T) {}

fn main() {}

While this code fails:

trait Boring {}

trait Usage
where
    for<'a> &'a Self::Thing: Boring,
{
    type Thing;
}

fn ref_example<T: Usage>(t: T::Thing) {
    use_the_trait(&t);
}

fn use_the_trait<T: Boring>(_: T) {}

fn main() {}

error[E0277]: the trait bound `for<'a> &'a <T as Usage>::Thing: Boring` is not satisfied
  --> src/main.rs:10:1
   |
10 | / fn ref_example<T: Usage>(t: T::Thing) {
11 | |     use_the_trait(&t);
12 | | }
   | |_^ the trait `for<'a> Boring` is not implemented for `&'a <T as Usage>::Thing`
   |
note: required by `Usage`
  --> src/main.rs:3:1
   |
3  | / trait Usage
4  | | where
5  | |     for<'a> &'a Self::Thing: Boring,
6  | | {
7  | |     type Thing;
8  | | }
   | |_^

This is probably related to #20671, but seems slightly different from the examples posited there. Specifically, the "standard" case (the first example) does work.

Also highly relevant: #44656 and #32722.

Tested with Rust 1.25.0 and 1.27.0-nightly (2018-04-29 79252ff)

[adityashah1212]

Not sure how come, but this seems to be working on Rust Playground. I am not sure if there was something that changed recently

[shepmaster]

The code posted in the original comment still fails in Rust 1.73 and 1.75.0-nightly (2023-10-15 42b1224), so it seems your situation is distinct.

[fmease]

Might get fixed by #120752, haven't double-checked (yet).

[compiler-errors]

This is not fixed by #120752. This is just because we don't imply/elaborate arbitrary where clauses. To the trait solver, for<'a> &'a _: Trait is just a random type and doesn't really correspond to a supertrait or item bound like Self: Trait or Self::Item: Trait would.

[fmease]

Thanks for the correction, that makes perfect sense! I should've triaged more carefully