Bad diagnostics from RPITIT

[Noratrieb]

Given the following code: playground

#![feature(return_position_impl_trait_in_trait)]
#![allow(incomplete_features)]

trait OnlySized<T> {}

impl<T> OnlySized<T> for () {}

trait LetsGo {
    fn uwu() -> impl OnlySized<[u8]>;
}

struct Owo;

impl LetsGo for Owo {
    fn uwu() -> () {}
}


fn main() {}

The current output is:

error[E0277]: the size for values of type `[u8]` cannot be known at compilation time
  --> src/main.rs:15:17
   |
15 |     fn uwu() -> () {}
   |                 ^^ doesn't have a size known at compile-time
   |
   = help: the trait `Sized` is not implemented for `[u8]`
   = help: the trait `OnlySized<T>` is implemented for `()`
note: required for `()` to implement `OnlySized<[u8]>`
  --> src/main.rs:6:9
   |
6  | impl<T> OnlySized<T> for () {}
   |         ^^^^^^^^^^^^     ^^
note: required by a bound in `LetsGo::uwu::{opaque#0}`
  --> src/main.rs:9:22
   |
9  |     fn uwu() -> impl OnlySized<[u8]>;
   |                      ^^^^^^^^^^^^^^^ required by this bound in `LetsGo::uwu::{opaque#0}`

The problem here is that (): OnlySized<T> where T: Sized, but [u8]: !Sized, therefore (): !OnlySized<[u8]>

The doesn't have a size known at compile-time pointing at () is quite confusing, since (): Sized.

Also the {opaque#0} name from the desugaring can be improved as well.

[Noratrieb]

@rustbot label F-return_position_impl_trait_in_trait

[compiler-errors]

The span is fixed by #101676:

error[E0277]: the size for values of type `[u8]` cannot be known at compilation time
 --> /home/gh-compiler-errors/test.rs:9:22
  |
9 |     fn uwu() -> impl OnlySized<[u8]>;
  |                      ^^^^^^^^^^^^^^^ doesn't have a size known at compile-time
  |
  = help: the trait `Sized` is not implemented for `[u8]`
note: required by a bound in `OnlySized`
 --> /home/gh-compiler-errors/test.rs:4:17
  |
4 | trait OnlySized<T> {}
  |                 ^ required by this bound in `OnlySized`
help: consider relaxing the implicit `Sized` restriction
  |
4 | trait OnlySized<T: ?Sized> {}
  |                  ++++++++

But, for example, this code:

#![feature(return_position_impl_trait_in_trait)]

trait Foo {
    fn bar() -> impl std::fmt::Display;
}

impl Foo for () {
    fn bar() -> () {}
}

Still mentions {opaque#0}, see:

error[E0277]: `()` doesn't implement `std::fmt::Display`
 --> /home/gh-compiler-errors/test.rs:8:17
  |
8 |     fn bar() -> () {}
  |                 ^^ `()` cannot be formatted with the default formatter
  |
  = help: the trait `std::fmt::Display` is not implemented for `()`
  = note: in format strings you may be able to use `{:?}` (or {:#?} for pretty-print) instead
note: required by a bound in `Foo::bar::{opaque#0}`
 --> /home/gh-compiler-errors/test.rs:4:22
  |
4 |     fn bar() -> impl std::fmt::Display;
  |                      ^^^^^^^^^^^^^^^^^ required by this bound in `Foo::bar::{opaque#0}

I'm not sure if I know a better way of explaining "required by a bound in an RPITIT in Foo::bar" -- do you have thoughts on what you would've wanted to see here @Nilstrieb, since you commented on that?

[Noratrieb]

Maybe we should just say required by this bound in Foo::bar. If we find a neat and obvious way to tell it that it's the RPITIT specifically it would be great, but I think just pointing at the function would be enough.

[compiler-errors]

Gonna unassign myself this issue because I don't think it's necessarily an issue, or at least I don't know exactly how to solve it in a way that makes me happy yet. I think "required by this bound in Foo::bar is also ambiguous compared to mentioning ::{opaque#0}", though someone else can take a stab at it and we can see what the diagnostic fallout looks like.

[compiler-errors]

This now says "required by a bound in OnlySized", so I'm considering this fixed.