Use futures_io::{AsyncRead, AsyncWrite} for L2Cap Streams

[abezukor]

As discussed in my previous PR (#18) and (#33), it is a better interface to implement AsyncRead and AsyncWrite from futures-io, rather than use custom functions. This PR switches to that interface.

Open Questions

Should we remove the unstable feature? Are there any more changes expected for L2Cap?

Todos

• After this gets merged, make another PR adding linux support.

[alexmoon]

That looks pretty good to me. @lulf can you review how the proposed L2CAP API would work for your use-case?

[lulf]

I'll give this a spin and a review, thanks @abezukor.

[lulf]

Could we still expose the non-blocking try_write or is it available via some traits?

[andrewzhurov]

I've been using it as writer.write_all(&buf).await, from future-lite's pub trait AsyncWriteExt: AsyncWrite.
I think it does repeated poll_write(&whatevers_left_buf) until it returns Ok((0)) (written bytes) or Err.
Worked no prob for writing 100kb bytes buf

[lulf]

Yeah, but I need to be able to write to the output buffer in a non-async context, without blocking.

[alexmoon]

You could use writer.write(...).now_or_never(). Or (equivalently) you could call writer.poll_write(Context::from_waker(Waker::noop()). They're not the most ergonomic options, but we could provide a try_write impl based on them if needed.

[lulf]

Ah, thanks. now_or_never() was what I was looking for. It's not that easy to discover, but works for me.

[lulf]

Ok, so I confirm this works well on both android and ios for us

[alexmoon]

Thanks! @abezukor I think this should be ok to merge one the CI issues are resolved. I'd like to keep L2CAP unstable until we have the Linux implementation, just to be sure there's nothing we need to change.

[abezukor]

Thanks! @abezukor I think this should be ok to merge one the CI issues are resolved. I'd like to keep L2CAP unstable until we have the Linux implementation, just to be sure there's nothing we need to change.

Thanks. I am on vacation without a laptop this week, but I should be able to fix the CI next week. I can also follow up with a Linux implementation.

[andrewzhurov]

+1, kudos for the PR! I needed l2cap on android and linux, so been playing with it, adding linux l2cap via AsyncRead AsyncWrite traits as here (been of great help as reference). Could submit a PR once this gets merged, if that's of use?

(also added l2cap listener for android and linux, advertising, and extended scan fn with sercive-uuid filter)
Able to start listeners, advertise their PSM (on Android it's dynamic, assigned on listener creation), scan these ads to learn addresses + PSM of another device, open l2cap from Android to linux, send and read some bytes.
(for some reason opening l2cap from linux to Android time outs, investigating atm)
Anyhow, bunch of stuff, could start with that l2cap for linux PR.