Hm, I wonder if this idiom can actually be saved, as follows: - When a loop is created its _stop flag is False. - Calling stop() sets the _stop flag to True. - Calling run_forever() does *not* reset the _stop flag at the beginning. - However when run_forever() exits it resets the _stop flag to False.
Let me work on a new patch for that. (bugs.python.org seems to have some database slowness issues, or maybe it's my hotspot, but I'll get this done ASAP so you can test.) On Thu, Nov 19, 2015 at 7:23 AM, Andrew Svetlov <andrew.svet...@gmail.com> wrote: > Since > loop.stop() > loop.run_forever() > is broken in the patch aiohttp test suite as well as almost all my > other libraries and private projects hangs. > Replacing loop.stop() with loop.call_soon(loop.stop) is easy and > obvious fix but the change is required everywhere for all projects. > > Obviously it is backward compatibility break and people will complain > that asyncio starts hanging in new release but worked perfectly before > 3.5.1 > > Personally I can live with that if there is no way to keep current > loop.stop() approach. > Maybe we can make `loop.stop()` as shortcut for actual > `loop.call_soon(loop._do_stop())`? -- --Guido van Rossum (python.org/~guido)
