expose try_recv and try_send on channels

[yoshuawuyts]

Exposes Receiver::try_recv and Sender::try_send on channels. Closes #579. I'm proposing we put the types in the crate's top-level because we're still undecided. I'd like to move to unblock people first, and then consider how we might want to move types around. Thanks!

Screenshot

[k-nasa]

LGTM!!

[ghost]

How do you feel about this mild inconsistency where we sometimes use Option and sometimes Result?

fn try_recv(&self) -> Result<T, TryRecvError>;
fn recv(&self) -> Option<T>;
fn next(&self) -> Option<T>;

Or, to put it differently, should recv return a Result too?

[yoshuawuyts]

@stjepang Interesting! None implies "no more items", where Result implies "something went wrong here". Arguably the try_ variants could return Option too, but then it wouldn't be possible to distinguish between error kinds.

On the other hand mpsc::Receiver::recv does return a Result. The way items are taken out without being fallible is by using iter, which returns None when done.

This is interesting; hadn't thought of this. I was going to say that the current state is good, but now looking at what std provides I'd almost say that following their lead might make more sense? crossbeam_channel::Receiver::recv works the same way too!

[ghost]

Yeah... in that case, are you okay with recv() and next() (coming from Stream) differing in Result vs Option? They do the same thing, except recv is slightly more efficient that next.

fn recv(&self) -> Result<T, RecvError>;
fn next(&self) -> Option<T>;

[yoshuawuyts]

@stjepang yeah, that sounds good. Seems to be the same in crossbeam and std as well (:

[yoshuawuyts]

Note from triage: we should remove the Disconnected case from the TrySendError enum.

[leo60228]

Is this a good alternative?

use futures::prelude::FutureExt;

fn try_send<T>(chan: Sender<T>, x: T) -> bool {
    chan.send(x).now_or_never().is_some()
}

[yoshuawuyts]

@stjepang actually I'm unsure how to remove that case. Maybe let's leave it in for now and consider changing it as part of stabilization?

edit: defined a new type RecvError that makes recv return a Result rather than an Option.

[yoshuawuyts]

ping @stjepang

[dignifiedquire]

yes please, I was recently searching for eactly this

[yoshuawuyts]

@dignifiedquire fixed the doctests, fingers crossed it passes now.