How to implement Promises with Tokio-Core?

[asomers]

I'm coming from the world of Seastar. Seastar has chainable futures, like Tokio. But it also has Promises. A Promise is an object that can be used to make Future ready. Typically, you use them like this:

Future do_something() {
    std::pair<Future, Promise> pair = make_future();
    send_to_some_other_time_or_place(pair.second);
    return(pair.first);
}

And then at some later time, the receiver of the Promise calls promise.set_value(), and suddenly its Future is ready.

I'm trying to do this same thing with Tokio-core. It seems that mio supports it by using mio::SetRegistration. I think I can make it work by issuing the low-level I/O at the time my future is created. However, that violates the principal that a future should do nothing until polled. Plus, it's pretty convoluted. I would rather not issue the low-level I/O until my future's poll method. But that's a problem, because the low-level I/O is itself using a Tokio library, and it returns a future. There's no way to return a future from a poll method, so my low-level library won't do anything. It's like this:

fn poll(&mut self) -> futures::Poll<(), io::Error> {
    if ! self.issued {
        let future = LowLevelLibrary::do_something().and_then{||
             self.io.set_readiness(true); //No way to send future to the reactor!
        }
    }
    self.issued = true;
    if self.io.poll_read() == futures::Async::NotReady {
        return Ok(futures::Async::NotReady);
    }
    Ok(futures::Async::Ready(())
}

Has anybody done this before? Any good ideas? If it can be done in a straightforward and reusable way, I might try to write a middleware crate for it.

[carllerche]

This sounds exactly like sync::oneshot in futures-rs. Is that correct?

[asomers]

Actually @carllerche I don't think that solves my entire problem in this case, because I want my promise to be triggered by a future that doesn't yet exist at the time when I need to return a future. But it looks like tokio_core::reactor::Handle::spawn may be able to help me, because I can call that from within my poll method. I'll try using those two methods together.

[alexcrichton]

Ok sounds like oneshot and perhaps some other methods should solve this, so I'm going to close this.