The value no longer satisfies the constraint after passing through `identity`, which seems counterintuitive

[A4-Tacks]

use std::convert::identity;

fn foo<F: for<'a> FnOnce(&'a ()) -> &'a ()>(f: F) {
    f(&());
}

fn main() {
    //foo(identity(|x| x));
    foo(|x| x);
}

I'm not sure if this is a generic or something issue

[LunarLambda]

My guess is that passing the closure as-is causes it to coerce to a fn-pointer which fulfills the HRTB, but passing it through identity causes it to be passed as whatever anonymous closure struct type it is, which the compiler cannot prove fulfills the HRTB since it's not part of the signature (in other words, identity() could feasibly return a different value of the same type, so the compiler is conservative and the inferred type doesn't fulfill the HRTB)