Memory leak when dropping waker after runtime drop

[Paraworker]

Consider the following case:

static WAKER: AtomicWaker = AtomicWaker::new();

struct LeakWatcher;

impl Future for LeakWatcher {
    type Output = ();

    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<()> {
        WAKER.register(cx.waker());
        Poll::Pending
    }
}

impl Drop for LeakWatcher {
    fn drop(&mut self) {
        println!("LeakWatcher is being dropped!!!");
    }
}

fn test_memory_leak() {
    let rt = Runtime::new().unwrap();

    rt.block_on(async {
        runtime::spawn(LeakWatcher).detach();
    });

    // Dropping the runtime clears all runnables in the queue, which breaks the reference cycle.
    drop(rt);

    // Dropping the waker after the runtime is dropped may push the runnable back into the queue,
    // recreating the reference cycle.
    let _ = WAKER.take();
}

Using Weak in the schedule closure might be necessary.

[Berrysoft]

Interesting. Would you propose a PR?

[Paraworker]

Interesting. Would you propose a PR?

I’m on it.