> It's not easy because I collect informations from various buildbots and
different virtual machines. But I planned to try to summarize the overall
work done on time in asyncio, select and selectors.

Thanks for the summary.

> The first problem is that clocks and selectors have a different
resolution. On Linux, poll() has a resolution of 1 ms, whereas
clock_gettime(CLOCK_MONOTONIC) has a resolution better than 1 us (0.3 us
according to my last test).

time() having a better resolution than select() isn't an issue, so we're
good.

> On Windows, GetTickCount64() has a resolution of 15.6 ms, whereas
GetQueuedCompletionStatus() has a resolution of 1 ms when HPET is enabled.
>
> Note: It looks like GetQueuedCompletionStatus() has the same resolution
than GetTickCount64() when HPET is disabled.

OK, so basically this means that with HPET, select() does block, but the
clock resolution is too low, and reports an inaccurate elapsed time (and
sometimes returns 0).
So actually, no busy loop occurs even in this case.
Wouldn't it be possible to use another clock for asyncio on Windows? I find
surprising that we can't get better than 15.6ms with HPET...