[3.9] bpo-44815: Always show deprecation in asyncio.gather/sleep()

[Dreamsorcerer]

https://bugs.python.org/issue44815

[serhiy-storchaka]

A change in gather() LGTM, but useless get_running_loop() is called for sleep(0). I suggest to change the code in the same way as in gather():

if loop is not None:
    warnings.warn(...)
if delay <= 0:
    ...
if loop is None:
    ...

It will add smaller overhead.

[ambv]

Serhiy is right, and additionally we need to silence all the new warnings in test_asyncio and test_asyncgen. I'm looking into that since it's a bit tricky.

[serhiy-storchaka]

I'm looking into that since it's a bit tricky.

How was it solved in 3.10? If the corresponding test code (for loop is not None) was just removed or wrapped with assertRaises(TypeError), we should just wrap it with assertWarns(DeprecationWarning). If it was rewritten in a way that does not pass a loop, we perhaps need to backport these changes.

[ambv]

How was it solved in 3.10?

The deprecations reached their EOL and functionality was removed. But it's a problem that will return so I think it's worth approaching systematically. We will need something like GH-27634 to do it.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @ambv for commit 783e66c 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @ambv for commit 217c789 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @ambv for commit d234e0d 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[TBBle]

I think this has introduced an unavoidable deprecation warning in BaseEventLoop.create_server(..., start_serving=True), which is the default value of start_serving:

https://github.com/python/cpython/blob/v3.9.7/Lib/asyncio/base_events.py#L1514-L1518

        if start_serving:
            server._start_serving()
            # Skip one loop iteration so that all 'loop.add_reader'
            # go through.
            await tasks.sleep(0, loop=self)

It also appears to affect Server.start_serving, which does the same thing, and from a brief glance, self._loop here is not going to be None.

https://github.com/python/cpython/blob/v3.9.7/Lib/asyncio/base_events.py#L349-L353

    async def start_serving(self):
        self._start_serving()
        # Skip one loop iteration so that all 'loop.add_reader'
        # go through.
        await tasks.sleep(0, loop=self._loop)

Edit: Late update, this new failure was also reported as https://bugs.python.org/issue45097 and fixed in 3.9.8.