GeezerVison

Please remove the [asyncio] prefix from the title.

Ok

I'm not sure that this issue solves MultiLoopChildWatcher design issue. The MultiLoopChildWatcher doc says that it can be used even if no event loop runs in the main thread, whereas add_signal_handler() must be called from the main thread:

https://docs.python.org/dev/library/asyncio-eventloop.html#asyncio.loop.add_signal_handler

I'm not sure that this issue solves MultiLoopChildWatcher design issue. The MultiLoopChildWatcher doc says that it can be used even if no event loop runs in the main thread, whereas add_signal_handler() must be called from the main thread:

https://docs.python.org/dev/library/asyncio-eventloop.html#asyncio.loop.add_signal_handler

This PR changes MultiLoopChildWatcher in a way that it still doesn't require a running loop in the main thread. So the basic idea remains the same. The only change is that MultiLoopChildWatcher now can't be used from any other thread. I've tested it in the REPL and also the test ran for about 30 minutes, so I guess this patch works?

Breaking backwards compatibility is not an option for 3.9. I'd say the same for 3.10, since we're past feature freeze; beta phase is all about fixing regressions (and potentially ripping out newly introduced features).

Breaking backwards compatibility is not an option for 3.9. I'd say the same for 3.10, since we're past feature freeze; beta phase is all about fixing regressions (and potentially ripping out newly introduced features).

For 3.9, I agree. But since this is a bug wouldn't it be better to fix it in 3.10 beta?

IMO, it's better to deprecate the class in 3.10 and remove it in 3.12, and for 3.9 do nothing. But, like Victor, I'm not an asyncio user, nor do I have any knowledge of that module, so my opinion does not really matter here.

This PR is stale because it has been open for 30 days with no activity.

MultiLoopChildWatcher is deprecated now and the issue is closed so closing. Thanks.