async wait/sleep implementation

[russkel]

Feature request

Feature description

I have been writing some hardware "test" scripts used for checking the integration of hardware in real time (not using any test framework or assertions but simply stimulating a system with inputs and manually confirming the hardware is performing as expected). I have found writing this using async has made things much simpler and nicer:

e.g.

    async def test_1(self):
        self.get_logger().info("Setting Rudder to Center")
        await self.set_and_wait_until_rudder_angle(0.0, 5.0)

        self.get_logger().info("Setting Rudder to Full Port")
        await self.set_and_wait_until_rudder_angle(-1.0, 5.0)

To which I kick off like so:

def main(args=None):
    rclpy.init(args=args)
    executor = SingleThreadedExecutor()
    node = TestNode()
    executor.add_node(node)

    try:
        node.get_logger().info("Starting Test 1")
        task = executor.create_task(node.test_1())
        executor.spin_until_future_complete(task)

        node.get_logger().info("Testing complete!")
    except KeyboardInterrupt:
        pass

    node.destroy_node()
    executor.shutdown()

Now there is pretty much no examples of create_task being used like this so I have been experimenting and reading a bit of the rclpy code. After looking at the executor code in depth I couldn't find out how to use the executor to do an equivalent to await asyncio.sleep. I had a quick look at how asyncio implemented theirs and I came up with this hack using a Timer:

    async def async_wait_for(self, rel_time: float):
        fut = Future()
        timer = None

        def done_waiting():
            fut.set_result(None)
            timer.cancel()

        timer = self.create_timer(rel_time, done_waiting)
        await fut

This was the only missing piece and I was able to asynchronously "wait" and write the co-routines as expected.

My request is something like this is added to the Executor class, perhaps using a more elegant interface than abusing a Timer like this.

#279 might be related.

[arjo129]

I like this idea. I recently ran into a similar need. My solution was spin_until_future_complete has a timeout option which I abused with a default empty future that never gets set. It probably is not great but 🤷.

[russkel]

I like this idea. I recently ran into a similar need. My solution was spin_until_future_complete has a timeout option which I abused with a default empty future that never gets set. It probably is not great but 🤷.

If I am not mistaken that will block the thread you're in? How do you use that to await in a co-routine?

[arjo129]

Yeah it does. I wasnt using async await so... If an async/await option were available I'd probably have prefered that. Just echoing the sentiment that an API like this would be useful for people.

[russkel]

Yeah it does. I wasnt using async await so... If an async/await option were available I'd probably have prefered that. Just echoing the sentiment that an API like this would be useful for people.

Yep I understood what you meant, I was just wondering if I had missed something.

[jmblixt3]

I like this idea. I recently ran into a similar need. My solution was spin_until_future_complete has a timeout option which I abused with a default empty future that never gets set. It probably is not great but 🤷.

If I am not mistaken that will block the thread you're in? How do you use that to await in a co-routine?

I was also looking into this, and looks like since #1316 spin_until_future_complete now works in callbacks. I'm still not a fan of relying on an empty future never finishing though as the "right" way to do this