Re: [gpsd-dev] Direct GPS support in ntpd makes me nervous

[Harlan Stenn]

Eric,

"Eric S. Raymond" writes:
> Andy Walls <address@hidden>:
> > Check the list of reference clock drivers at the bottom of this page:
> > 
> > http://www.eecis.udel.edu/~mills/ntp/html/refclock.html
> > 
> > Type 20 is Generic NMEA, but there are others (Jupiter, Oncore,
> > Palisade, etc.).
> 
> Heads up, Harlan. You need to hear this.
> 
> Learning that there is direct GPS support in ntpd makes me nervous.
> I recommend you take a good hard look at *removing* it.
> 
> Getting GPS sentence parsing right is much harder than it looks, and
> has lots of unobvious failure modes. The problem isn't algorithmic
> difficulty, it's background issues like NMEA being very weakly
> specified. A driver based on one or a few NMEA devices is likely to
> embody regularity assumptions that will be false on many others.
> 
> Things like varying sentence orders in fix reports and the
> presence/absence of ZDA are likely to bite you in the ass.  And I have
> to tell you that aggregating a full timestamp out of the mess has
> been one of the *most* trouble-prone parts. (That and sky views have
> been the two worst pain points.)
> 
> GPSD specializes in this job and has the best test suite in existence.
> Even so it took us *years* of effort to get this right.  I trust our
> code, but I don't trust anybody else's - because you can't beat this
> problem just by being smart and careful.  It takes lots of ugly
> experience and having your nose rubbed in the take from fifty
> different kinds of idiosyncratic and semi-broken GPSes.
> 
> We've had that experience. Nobody else has. I therefore strongly
> recommend that, for the sake of ntpd's future defect rate and
> robustness under field conditions, you exit this piece of the
> hardware-management problem and leave it to GPSD.

ntpd has had direct driver support for a Very Long Time, pre-dating
gpsd.  This includes support for a number of non-GPS devices.

I'm game to identify all of the refclocks that ntpd supports, and if
gpsd is supporting those refclocks I'll talk to DLM about this and bring
it up on the appropriate NTP lists for discussion.

This seems like a reasonable plan to me, and I'd be game to look at the
chrony pipes to see about including them in ntpd, too.

H