Fn trait associated type (return type) considered unconstrained

[afranchuk]

I tried this code:

struct S<F>(F);

impl <'a, F, G> S<F>
where
    F: Fn(&'a u8) -> G
{
    fn call(&mut self) -> G {
        (self.0)(&1)
    }
}

I expected to see this happen: G should be considered a constrained type.

Instead, this happened:

error[E0207]: the type parameter `G` is not constrained by the impl trait, self type, or predicates

Note that if you use '_ as the lifetime instead of 'a, it works. It also works if there is no lifetime on the input parameter (e.g. if it's not a reference). However I want to constrain G to the same lifetime (eventually) so that's not an option. Is this an issue due to some lifetime assumptions about the return type from Fn traits, specifically that having G: 'a makes it unconstrained as it can't reason about that?

This is identical for FnOnce and FnMut as well.

Meta

rustc --version --verbose:

rustc 1.53.0 (53cb7b09b 2021-06-17)
binary: rustc
commit-hash: 53cb7b09b00cbea8754ffb78e7e3cb521cb8af4b
commit-date: 2021-06-17
host: x86_64-unknown-linux-gnu
release: 1.53.0
LLVM version: 12.0.1

[b-naber]

@rustbot claim

[b-naber]

This seems to not be a bug, but an intentional error, see #25041

[b-naber]

@rustbot release-assignment

[b-naber]

Maybe to summarize why this is an error. The rules for constrained parameters are in this rfc:

If T appears in the impl trait reference,
  then: T is constrained

If T appears in the impl self type,
  then: T is constrained

If <T0 as Trait<T1...Tn>>::U == V appears in the impl predicates,
  and T0...Tn are constrained
  and T0 as Trait<T1...Tn> is not the impl trait reference
  then: V is constrained

Since the lifetime parameter 'a is unconstrained when we determine whether G is constrained the third rule does not apply.

[afranchuk]

Okay, that's what I suspected. I guess I really want a more powerful for<'a> that can apply to the return type (which I've discussed before on Zulip). It'd be very helpful in the future if the error message indicated exactly what makes a particular type unconstrained.

[Dylan-DPC]

Closing this, as it's intentional as per #86934 (comment)