Remote-related bug in tokio-core >= 0.1.13

[srijs]

Hi!

It appears that the new versions of tokio-core based on tokio (tokio_core >= 0.1.13) have broken Rusoto (rusoto/rusoto#1000).

I was able to reduce it to a simple example program that works correctly with tokio_core 0.1.12 (tokio_io 0.1.5), but breaks with tokio_core 0.1.13 or 0.1.14.

Here is the program:

extern crate tokio_core;

use std::thread;
use std::sync::{Arc, Barrier, Mutex};

use tokio_core::reactor::{Core, Remote};

fn spawn() -> Remote {
    let slot = Arc::new(Mutex::new(None));
    let slot_clone = slot.clone();

    thread::spawn(move || {
        let mut core = Core::new().unwrap();
        let mut guard = slot_clone.lock().unwrap();
        *guard = Some(core.remote());
        drop(guard);
        println!("turning core...");
        loop {
            core.turn(None);
        }
    });

    loop {
        let mut guard = slot.lock().unwrap();
        if let Some(remote) = guard.take() {
            return remote;
        }
    }
}

fn main() {
    let remote = spawn();
    let barrier = Arc::new(Barrier::new(2));
    let barrier_clone = barrier.clone();

    remote.spawn(move |_| {
        println!("waiting in remote");
        barrier_clone.wait();
        println!("waking in remote");
        Ok(())
    });

    println!("waiting in main");
    barrier.wait();
    println!("waking in main");
}

The expected behaviour of the program is to print the following and then exit:

turning core...
waiting in main
waiting in remote
waking in remote
waking in main

However, using tokio_core 0.1.13 or 0.1.14, the output is the following, and it never exits:

turning core...
waiting in main

Superficially it looks like the closure passed to remote.spawn is never executed.

[srijs]

Upon further digging, if I replace

loop {
    core.turn(None);
}

with

core.run(futures::future::empty::<(), ()>()).unwrap();

the program works as expected.

Hope that helps!

[carllerche]

Thanks for the report. The simple repro case was very helpful.

I pushed #312 with a fix. Could you check against Rusoto to see if it fixes it?

[pgerber]

Tested the fix, the issue I reported against Rusoto no longer occurs with the fix applied.