Surprising soundness trouble around `PollFn`

7

The alternative way of creating a struct, and think about whether some field is structurally pinned, feels too heavy for an ad-hoc future only used in one place. I imagine poll_fn is used because of that reason.

With RFC 2781, one can easily write an ad-hoc future without any unsafe, like:

let future = fut();
let f: impl FnPin(&mut Context) -> Poll<()> = move |ctx| {
    pin_mut!(future);
    loop {
        yield future.poll(ctx);
    }
};
let f: impl Future = poll_fn_pin(f);