Re: [gpsd-dev] gpsdate program, similar to 'ntpdate'

[Terje Mathisen]

Ed W wrote:
> On 21/06/2013 08:17, Harald Welte wrote:
> > Hi C.J.,
> >
> > On Thu, Jun 20, 2013 at 03:56:00PM -0700, C.J. Adams-Collier wrote:
> > > I hear ntpdate has been deprecated long time and obsoleted in new
> > > releases.
> > >
> > > Try ntpd -q
> > I have tried this long ago, and would be happy if things would have been
> > that simple.
> >
> > ntp_refclock:refclock_process_f() calls libntp:clocktime() and the
> > latter aboslutely cannot deal with the fact if e.g. your embedded
> > system without RTC boots up assuming it is January 1st 1970 and a GPS
> > reference time in the year 2013 has been received.
> >
> > I looked into fixing the ntpd/libntp code some months ago, but gave up
> > as it seems to be written on the assumption that there simply are not 43
> > years of difference between two timestamps.

ntp-dev was updated a few years ago to allow the maximum possible gap 
(i.e. 2^31 seconds or 68 years) during startup.

> > Having said that, I'm clearly not a ntpd expert and happy to try
> > whatever method people suggest here.  But after lots of debugging and
> > ntp source code reading I arrived at the conclusion that it is not
> > possible using the existing code, which prompted the implementation of
> > gpsdate.
>
> I don't have this problem with Chrony when my board boots up with 2000
> as the date... However, I haven't tested with wider than 13 year gap.

As has been noted here, the current ntp-dev code also has logic to 
assume the time is within 68 years of the time when the binary was compiled.

This is actually a _very_ safe assumption imho. :-)

Terje
> Also I use gentoo on these boards and the service "swclock" which
> implements a weak monotonic clock source by setting the initial clock to
> match some file on the filesystem. Touch this file as often as you wish
> to get the initial clock as for forward in time as you wish.  The main
> goal is simply that the clock doesn't move (to far) backwards in time
> rather than it's accurate
>
> Ed W


--
- <address@hidden>
"almost all programming can be viewed as an exercise in caching"