John Belmonte

> This still has a task group straddling a `yield` which invalidates it in situations where cancel scope stack corruption can happen (e.g. pytest fixtures). And in other situations it's...

> > This wrapper absolutely has real world use cases, and the application at my workplace relies on it. > > I would be interested in seeing this use case...

It's not hard to imagine cases where you want to use a nursery or cancel scope within an async generator, and people are hitting these cases in real life (we...

I see. It's a little unfortunate as far as garbage generation. And I notice that Trio doesn't use slots for Event or ParkingLot, so that's two unslotted class instances for...

how about: * add WaitQueue to anyio API * on Trio it maps to ParkingLot * on asyncio, instantiate a new Event on each unpark This will at least make...

> And I notice that Trio doesn't use slots for Event or ParkingLot, so that's two unslotted class instances for every set(). `trio.Event` received some attention and we were able...

> and here's a PR to make ParkingLot slotted: [python-trio/trio#1971](https://github.com/python-trio/trio/pull/1971) This is good change and I overlooked it in my PR. But to clarify: trio.Event() is no longer implemented with...

this seems to work for me running autokbisw manually as root (seen on 10.13.5) (in fact running as root appears to be the only way autokbisw will work at all)

The test code is assuming that tasks are scheduled in `start_soon()` order, which isn't guaranteed. Perhaps the test could be clarified by using testing.Sequencer. You haven't mentioned whether your memory...

> It doesn't matter which order the multiple send_value() invocations begin. I see it now, sorry for the noise. > In my code the order that values get sent into...