fix lint to catch async fn with maybe sized return type

[Aditya-PS-05]

Fixes #149438

The compiler incorrectly accepted nested RPITs with ?Sized bounds when used in associated type positions that have implicit Sized bounds. See this playgound

The fix is to only skip validation when the nested RPIT does NOT have an explicit ?Sized bound.

[rustbot]

r? @fee1-dead

rustbot has assigned @fee1-dead.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Aditya-PS-05]

@lcnr , please review it.

[lcnr]

This doesn't quite feel right, it results in the following behavior, does it not?

trait Bound {}
trait Trait {
    type Assoc: ?Sized + Bound;
}
impl Bound for () {}
impl Trait for () {
    type Assoc = ();
}

fn foo() -> impl Trait<Assoc = impl ?Sized + PartialEq> {} 
// ^ now lints that `Bound` is not implemented?

I feel like what we'd want to do is sth like this:

• check whether item bounds involve Sized
• check whether we explicitly opt out of Sized
• lint in that case

View changes since this review

[Aditya-PS-05]

This doesn't quite feel right, it results in the following behavior, does it not?

trait Bound {}
trait Trait {
    type Assoc: ?Sized + Bound;
}
impl Bound for () {}
impl Trait for () {
    type Assoc = ();
}

fn foo() -> impl Trait<Assoc = impl ?Sized + PartialEq> {} 
// ^ now lints that `Bound` is not implemented?

I feel like what we'd want to do is sth like this:

* check whether item bounds involve `Sized`

* check whether we explicitly opt out of `Sized`

* lint in that case

View changes since this review

Yeah, my fix is too broad.
I will update it.