Suggestion for small performance improvement

[gammelalf]

Hi,
nice project!

I just did a small prototype before knowing of this crate, which ended up implementing the same thing as you did.
Although mine is just a prototype, I spotted a simple change you could do to improve your speed a tiny bit:

How

Replace your create function in waker.rs with:

pub fn create() -> impl std::ops::Deref<Target = Waker> {
    // Safety: The waker points to a vtable with functions that do nothing. Doing
    // nothing is memory-safe.
    std::mem::ManuallyDrop::new(unsafe { Waker::from_raw(RAW_WAKER) })
}

Nothing else would have to change because you only ever use the waker as reference and ManuallyDrop implements Deref.

Why

Your waker's drop method is a noop because it has no data to clean up.
However the Waker instance doesn't know this and will always call drop resulting in a virtual function call whose body does nothing.
By wrapping the Waker in ManuallyDrop you avoid this single unnecessary function call which is done in each advance call.