Bring sleep back to WgpuAwaitable. Slowly back off on the amount of time sleeping.

[fyellin]

No description provided.

[fyellin]

I've probably overengineered this.

1. Should I be using async_sleep (defined in that file) or should I change async_sleep to use anyio? Now that we're actually sleeping, it's more important to use Event when sleeping. (Some other operations poll() might cause this callback to happen). Does everything that sniffio supports have an Event?

2. This sleep backoff just felt kind of right to me, but I don't really have a good feel. Happy to make changes.

[almarklein]

Yeah, we should only rely on sniffio, because the async stuff might actually be running on rendercanvas.utils.asyncadapter, which anyio does not know about. Both sleep and Event can be obtained in the way we now get the sleep func for async_sleep (it works for asyncio/trio/rendercanvas).

[fyellin]

@almarklein

The real problem is

                with anyio.move_on_after(next(sleep_generator)):
                    await self.event.wait()

This is the only way to wait on an event with a timeout intrio. (And anyio makes it work on asyncio). I can't find any documentation on rendercanvas.utils.asyncadapter so I don't know if it has something compatible. I can get rid of the events if necessary, but they do make the code more responsive.

[almarklein]

Ah, I see what you mean. We could technically add support for this to rendercanvas, but the simpler solution is to not use Event here. Since we scale up the wait time, the possibly time we waited too long should only be a fraction of the total time we had to wait, so I don't see harm in that.

Could even make that explicit:

sleep_time = time_since_started / 100

[almarklein]

Thanks @fyellin !