Michael Banck <mbanck@gmx.net> writes:
> How much timer resolution do we require from the system? GNU Mach seems
> to (at least try to) guarantee that the timer won't go backwards, but it
> does not guarantee (currently) that two consecutive clock_gettime()
> calls will return something different in all cases.
I think it is reasonable to require the clock to not go backwards
during a test run, but it's not at all reasonable to require the
clock to advance by more than zero between two successive readings.
We used to encounter the no-advance case all the time, back when
machines had clock resolutions measured in milliseconds. It's
relatively rare now though, so it's possible that some test case
has crept in that expects that. But I'd call it a bug in the
test case if so.
It'd be interesting to see the output of a pg_test_timing run
from your Hurd machine.
regards, tom lane
