asio coroutines with stdexec

[mathisloge]

I know that there is #1133 already, but it wasn't really answered.

I was wondering if someone knows a way to combine stdexec's concepts and asio's io_context.
Main reason is to allow to use stdexec but still use the existing ecosystem around asio. E.g. boost process v2 and so on.

e.g. could a sender be created from an asio::awaitable? I'm quite new to stdexec and trying to get my mental model in a place where I can use it in any form. So I'm sorry if my question is on a quite high level, but I would also be happy if someone could just give me an opinion of whether this could work and whether I still have to wait for P2855.

[maikel]

Do you ask for an asio completion token that makes asio functions return a sender? Iirc asio at least implemented some support for S&R (and deprecated it) but maintaining such a thing is difficult from the perspective of asio, because stdexec is a moving target.

I think its a good suggestion to have something like this in this repo

[maikel]

I had some time and made a CompletionToken for asio here https://godbolt.org/z/61jznd7x7

[mathisloge]

Oh wow, thank you for putting your time into this!!
It will try it this weekend!

[SamGaaWaa]

Try this https://github.com/SamGaaWaa/asio2exec

[maikel]

I've updated my example here with cancellation. But, for me, cancellation_signal/slot is still a bit weird. I think I need to synchronize something between threads since it seems to be completely unsynchronized, right?

https://godbolt.org/z/7hPc3ne1s

[maikel]

I think there is indeed a race between registering a cancelation handler at initiation and a possible stop request from the receiver 😯

[SamGaaWaa]

Yes, see this example https://github.com/NVIDIA/stdexec/blob/main/test/test_common/senders.hpp

[mathisloge]

I've tested the header of @SamGaaWaa . Thanks.
Since I was just beginning with my new project I've ported most of the async code from asio to stdexec.
Just some asio::*_pipes are left and it connects quite well.

@SamGaaWaa maybe you could add a license file to your repo.

[SamGaaWaa]

@mathisloge Ok.

[ericniebler]

I would be thrilled to merge ASIO integration into stdexec.

[pfeatherstone]

It would be good to see a roadmap for networking in senders. After all, senders was born out of a dissatisfaction with Asio's executor's model and the committee presumably thought a better asynchrony model would suit networking. Is that right? I would have thought networking would be the #1 application for senders, ahead of cpu-bound tasking and GPUs.

[MoFHeka]

Same request. There is a lot of code based on the asio ecosystem, such as asynchronous Redis client. I really don't want to reinvent the wheel myself.