Send a blank email to [email protected] to get a copy of this message
Your commit reminded me that we should probably look into providing
access to the low-level monotonic clock.
eg. CLOCK_MONOTONIC_RAW described here:
http://www.kernel.org/doc/man-pages/online/pages/man2/clock_getres.2.html
There is a monotonic clock library here that should help with portability:
https://github.com/ThomasHabets/monotonic_clock
As to why? The leap-second bug is a good example, but in general,
anything that schedules events or times soething with fine granularity
is going to be wildly inaccurate due to ntp clock drift if the timing
source is a regular gettimeofday call. It simply doesn't work reliably
for measuring sub-second intervals. If you need to reliably trigger a
timeout 10ms from now, you have to use a monotonic clock call.
-Rasmus