Experimental Physics and
Industrial Control System

"Zimoch Dirk \(PSI\) via Core-talk" <[email protected]> · Thu, 25 Jul 2019 08:23:08 +0000

On Wed, 2019-07-24 at 13:08 -0700, Michael Davidsaver via Core-talk wrote:
> On 7/24/19 1:02 PM, Johnson, Andrew N. via Core-talk wrote:
> > I don't think anyone has suggested using clock_getres() for the quantum value before; that might be a better solution than just returning zero. Anyone else want to weigh in on that idea — should we just apply Dirk's patch?
> 
> cf. the posix version of epicsMonotonicResolution()
> 
> I've never looked at what is returned for CLOCK_REALTIME (which is always present).
> 
> I think the ultimate fallback has to be returning zero.  1e-3 is too arbitrary.

1e-3 is not completely arbitrary. periodicTask uses epicsEventWaitWithTimeout to
sleep which calls convertDoubleToWakeTime. The posix implementation of
convertDoubleToWakeTime uses gettimeofday if CLOCK_REALTIME is not present.
gettimeofday uses struct timeval which has only microsecond resolution.

Thus on systems where CLOCK_REALTIME does not exist, 1e-3 makes some sense.

However if there are no such posix systems, then the compiler branched can be
removed.

The posix epicsThreadSleep implementation on the other hand uses nanosleep and
is thus not limited to milliseconds.

Dirk

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System