Support async recvmsg and sendmsg

[LukasWoodtli]

async recvmsg and sendmsg

Following the discussion here: https://discuss.python.org/t/expanding-asyncio-support-for-socket-apis/19277

Implemented async variants of:
loop.sock_sendmsg - async analogue of socket.sendmsg
loop.sock_recvmsg - async analogue of socket.sock_recvmsg

This is a draft PR open for discussion.

Some open tasks for this draft:

• Improve documentation
• More testing
• Platform independence (developed only on Linux)

[ghost]

All commit authors signed the Contributor License Agreement.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[gvanrossum]

I've only skimmed this, but it looks pretty straightforward (it's just a clone of recvfrom/sendto). I'm still skeptical how much use this would get (the Discourse thread has some folks stating it'd be useful for them, but not a lot). I'm also skeptical that you couldn't easily provide this as a pair of user-level coroutines (yeah, I tried in the Discourse thread and failed, but if this could be a small 3rd party library, it'd take some maintenance off our hands -- asyncio is understaffed as it is).

(@Eclips4 Do you have interest in this yourself or did you just think that this is ready to be promoted from Draft?)

[gvanrossum]

Shouldn't this use recvmsg? (TBH I don't know anything about recvmsg so I may be missing something. But it looks to me like a simple copy-and-paste bug.)

[LukasWoodtli]

My bad.
The whole file had a lot of copy-and-paste code already before.
I would love to reduce the code duplications in that module. But probably I will not manage to find time for it.

[h-vetinari]

Targeting the branch of an already released version is probably a nonstarter for a new feature? Should almost certainly be targeting main.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[1st1]

libuv doesn't support these, so it will be a bit of pain to add support for these to uvloop.

All in all, these methods are not the most commonly used, but if one needs to use them, they still can using the add_reader / add_writer low level APIs. For that reason, I don't think these are worthy of adding and growing the API surface.

[gvanrossum]

libuv doesn't support these, so it will be a bit of pain to add support for these to uvloop.

Hm. It would be the addition of a small amount of pretty standard boilerplate to uvloop, right? Just clone the existing copy for recv and send? I have a slight preference for having asyncio lead and make uvloop follow. Maybe if you provide a small hint, the author of this PR might be able to add this to the next version of uvloop. In the meantime, you can just document as one of the temporary restrictions with uvloop that you can't use recvmsg and sendmsg yet (with an invite to readers to help fix this).

@LukasWoodtli What do you think?

[ghost]

All commit authors signed the Contributor License Agreement.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[LukasWoodtli]

I'm a bit undecided about how to proceed with this PR.
On one hand, the standard library seems to be incomplete as it provides some sync API that is not available as an async variant. But on the other hand I see that it doesn't make a lot of sense to maintain some library functions that are only in use by one project.
If there will be more interest in the future, the PR can be reopened again.

[llchan]

I am interested in seeing this merged. It's a pretty general need that's not bespoke to one downstream project. For example, it's needed to be able to use e.g. IP_PKTINFO in an asyncio context.