bevy implied bounds hack no longer works

[lcnr]

affected test

• tests/ui/implied-bounds/bevy_world_query.rs

somewhat unsure about why this happens, but

// Needed due to <https://github.com/rust-lang/rust/pull/137253>.
#![crate_name = "bevy_ecs"]

pub trait WorldQuery {}
impl WorldQuery for &u8 {}

pub struct Query<Q: WorldQuery>(Q);

pub trait SystemParam {
    type State;
}
impl<Q: WorldQuery + 'static> SystemParam for Query<Q> {
    type State = ();
    // `Q: 'static` is required because we need the TypeId of Q ...
}

pub struct ParamSet<T: SystemParam>(T) where T::State: Sized;

fn handler<'a>(x: ParamSet<Query<&'a u8>>) {
    let _: ParamSet<_> = x;
}

fn ref_handler<'a>(_: &ParamSet<Query<&'a u8>>) {}

now errors with "the type &'a u8 does not fulfill the required lifetime"

[compiler-errors]

In the implied bounds compat query, we use the "regular" wf with normalization:
https://github.com/rust-lang/rust/blob/ed49386d3aa3a445a9889707fd405df01723eced/compiler/rustc_trait_selection/src/traits/query/type_op/implied_outlives_bounds.rs#L180

That normalizes obligations using normalize_with_depth_to:
https://github.com/rust-lang/rust/blob/ed49386d3aa3a445a9889707fd405df01723eced/compiler/rustc_trait_selection/src/traits/wf.rs#L360

...Which instead of registering the type outlives obligation into the infcx's region storage (which when normalizing a <Query<&'a u8> as SystemParam>::State: Sized wf predicate, gives us &'a u8: 'static from the impl), it just puts it into the vec of wf clauses.

In the new solver, even if we were to use deeply_normalize on the wf predicate here:
https://github.com/rust-lang/rust/blob/ed49386d3aa3a445a9889707fd405df01723eced/compiler/rustc_trait_selection/src/traits/query/type_op/implied_outlives_bounds.rs#L209

...we would still need to manually take the type outlives obligation that has been registered as part of normalizing the type out of the infcx here:
https://github.com/rust-lang/rust/blob/ed49386d3aa3a445a9889707fd405df01723eced/compiler/rustc_trait_selection/src/traits/query/type_op/implied_outlives_bounds.rs#L253

[compiler-errors]

I think it's doable, and the fallout is likely very small b/c it's just for the compat hack (which I recently even reduced the scope of rust-lang/rust#137253).

[compiler-errors]

It's arguable whether or not we should fix this in the new solver, because we'd rather just remove the hack. If we do need to fix bevy by the time that the solver is ready for stabilization in 2050, then we can do a hack just like the PR I put.

[lcnr]

fixed by rust-lang/rust#137633