QueuingExecutor is not panic-safe

[Patryk27]

Hi, seems you're already aware of it:

crux/crux_core/src/capability/executor.rs

Line 119 in 238b8cc

// FIXME: are we potentially sending ourselves `Unavailable` and reading it

... but I thought it'd be nice to drop an example as well, since it's quite easy to trigger this behavior:

#[test]
fn test_panic() {
    let (executor, spawner) = executor_and_spawner();

    struct MyTask;

    impl Future for MyTask {
        type Output = ();

        fn poll(self: std::pin::Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
            cx.waker().wake_by_ref();
            panic!();
        }
    }

    spawner.spawn(MyTask);

    _ = std::panic::catch_unwind(|| {
        executor.run_all();
    });

    executor.run_all(); // runs forever
}

(note that the deadlocking is not caused by calling MyTask::poll() in a loop - that poll happens just once, .run_all() gets blocked via self.ready_sender.send(task_id) that gets picked up on the next iteration of the loop)

[charypar]

Any idea whether the command executor suffers from the same problem? They have a similar structure, but slightly different internals.

I'll try this myself, but if you're in a position to test it easily, it would help 🙏🏻

[Patryk27]

Command executor seems correct - tried to break it and failed 😄