GenericWaitForEventFuture doesn't handle task migration. about futures-intrusive

[Matthias247]

You are right. The same probably also applies to other Futures. I now recall a discussion on discord (or somewhere else?) where the question was whether Wakers should actually be updated. I think some mentioned "no", but strictly speaking it's the right thing to do. I will push an update.

I am not sure whether you can really move the Future between tasks. It at least has to be pinned, and thereby it is already very restricted what you can do with it. Executors will likely always be passing the same Waker, but it's better not to rely on it.

will_wake() sounds good to avoid the refcount churn - but I just discovered that it's broken due to the executor optimizations: Those sometimes pass a different Waker reference which only on .clone() gets transformed into the real Waker. E.g. futures-rs does it here. That saves the refcount increase and decrease on the executor - but now causes will_wake() to always return false.

[sfackler]

I am not sure whether you can really move the Future between tasks.

It is possible, but I'm not sure you'd ever really see it in "normal" code. You could for example make a future that contains a Pin<Box<impl Future>>, polls it once and then sends it on a channel to a task running on another executor.

[Matthias247]

This should be fixed now. I checked all Pending returns and added updates for the Waker.