Relaxed bounds in associated type bounds don't get rejected

[fmease]

The following code gets erroneously accepted:

fn f<T>() where T: Trait<Ty: ?Sized> {}
//                           ^^^^^^ not good

trait Trait { type Ty/*: ?Sized*/; }

Relaxed¹ bounds are only meant to be put on type parameters and associated types "declared in the immediate vicinity" which isn't the case here. This is supported by the fact that Trait<Ty:> (sic!) doesn't elaborate to Trait<Ty: Sized>.

Compare this to the snippet below which gets rightfully rejected:

fn f<T>() where T: Trait, T::Ty: ?Sized {}
//                               ^^^^^^
//~^^ ERROR `?Trait` bounds are only permitted at the point where a type parameter is declared
trait Trait { type Ty/*: ?Sized*/; }

---

Obviously a fix would be a breaking change. However I doubt anyone is writing such code (unless macro generated). In any case, we should run crater on the future PR.

Footnotes

1. Also known as unbounds (an ancient name for them), maybe-bounds (colloquially) or "question mark bounds" (very colloquially). ↩

[compiler-errors]

😭

[theemathas]

However I doubt anyone is writing such code (unless macro generated).

I found 2 (two) files containing code like this:
https://github.com/Gui-Yom/turbo-metrics/blob/1c6a196210cc85f00a9f3735d2c09adb80d00b08/crates/cudarse/cudarse-driver/src/mem.rs#L174
https://github.com/ireina7/summon/blob/506d5733b5326798c015c57026e87b072a4c3493/src/summon.rs#L33