The Output should not be an association type of the FnOnce

[823984418]

Currently, we require that the lifecycle that appears in an association type appear in a generic type.
This is to fix #32330 .
But this because the Output is an associative type of the function.

---

trait F {
    type Output;
}

Except for functions, no same type can implement different lifecycle Output.
So the Output as an association type is just an error.

See #33685 (comment)

[workingjubilee]

...But what exactly would we gain by making this change?

[823984418]

...But what exactly would we gain by making this change?

Make it a legal statement:

fn a<F: for<'s> Fn() -> Vec<&'s i32>>(f: F) 

And don't need to be like this:

fn a<F: for<'s> Fn(PhantomData<&'s ()>) -> Vec<&'s i32>>(f: F)

[823984418]

See https://internals.rust-lang.org/t/reopen-fnonce-why-is-output-an-associated-type/18430?u=qsh

[fmease]

The invalid bound for<'s> Fn() -> Vec<&'s i32> can just be expressed as Fn() -> Vec<&'static i32> or am I missing something? A closure that returns a vector of references that live at least as long as any possible lifetime that the caller can choose can only return 'static references since without any function arguments that are of that lifetime, any other lifetime that the callee would select, would simply be incorrect in general.

[823984418]

The invalid bound for<'s> Fn() -> Vec<&'s i32> can just be expressed as Fn() -> Vec<&'static i32> or am I missing something?

I have to admit that I am a bad example, I mean, 's is invariant for return type.

[823984418]

any other lifetime that the callee would select, would simply be incorrect in general.

I can say clear, this is wrong, the life cycle of designated by the caller is exist and reasonable.

One example is: RR::default

#[derive(Default)]
struct RR<'a, 'b> {
    r: Option<&'a mut &'b i32>,
}

Although it doesn't seem to make much sense, what about self-referencing life cycle?

#[derive(Default)]
struct RR<'a> {
    r: Cell<Option<&'a RR<'a>>>,
}

[jyn514]

This appears to be a breaking change: https://internals.rust-lang.org/t/reopen-fnonce-why-is-output-an-associated-type/18430/8?u=jyn514

Given that, it doesn't seem strongly enough motivated to be worth changing.