With impl trait, some variables are not recognised as `'static` when used with `AsRef`.

[P-E-Meunier]

An example is the following code:

fn f() -> impl Future<Item = (), Error = Error> {
    my_future.and_then(move |_| {
        foo("blah")
    })
}
fn foo<T: AsRef<str>>(t: T) {
     unimplemented!()
}

The current nightly rustc complains about unnamed lifetimes in impl trait. It accepts the same code with "blah" replaced by "blah".to_owned().

I believe "blah" should be recognised of lifetime 'static, and this code should be accepted by the compiler.

[eddyb]

cc @aturon @nikomatsakis I believe this is hard to accomplish without forcing all lifetimes in the concrete type, that were not inferred to be 'a from impl Trait + 'a, to be 'static. Because they would be inferred if there is an + 'a (AFAIK), so that means impl Trait without lifetimes allows no lifetimes (outside of type parameters of the function) to be anything other than 'static.

On second thought, that sounds close to the new RFC - if it's compatible with it, that's great!

[jyn514]

Triage: I'm not sure how to reproduce the original error, since neither futures::Future nor core::future::Future has Item or Error associated items. I would expect this to probably be fixed though since impl-trait is no longer experimental.

[P-E-Meunier]

Yes, definitely fixed.