> select() and kqueue() are able to sleep less than 1 ms. Using a slack of
1 ms would reduce the accuracy. I don't see why we should limit the
accuracy. Why 1 ms? Because of poll/epoll? What about Windows and its
resolution of 15.6 ms?
>
> Well, under 1 ms, don't expect accurate results :-) For example on my
fast Linux box, a select() with a timeout of 61 us takes 137 us (elapsed
time measued in Python).

That's precisely why I suggested 1ms.

And, by definition, it would only be used if the select()/poll() call
returns earlier in the first place, so it doesn't reduce the accuracy.

> I would like to avoid this *useless* loop, because such case can be
common.
> What's the point of polling again? We just polled all file descriptors.

"If it ain't broke, don't fix it": for the 10th time, all event loops work
like this, and this has *never* been an issue.
So I suggest we stop making up complicated solutions to non-existent
problems.