Re: [gpsd-dev] ntp reference clocks

[juergen perlinger]

Hello Gary, hello Sanjeev!

On 03/24/2015 08:14 PM, Gary E. Miller wrote:
> Yo juergen!
>
> On Tue, 24 Mar 2015 19:24:24 +0100
> juergen perlinger <address@hidden> wrote:
>
> > > Due to the old style fake IPv4 naming of ref clocks, you need a link
> > > from you GPS to /dev/gpsX.  Then the refclock last octet is the X.
> > >
> > > Also no way to get a naked /dev/ppsX from gpsd, which is the
> > > feature I'm working on now.
> > If that seems useful, I could use 'readlink()' to get the underlying
> > 'true' device name for the communication between NTPD and GPSD. But
> > that would not remove the need for a symlink, though it might make
> > things easier for GPSD. Gary, what's your opinion here?
> A link is a link.  The current way works, given the contraints.  Ideally
> the gpsd-json driver should query for available gpsd-json time sources
> and autoconfig.
>
> I appreciate that will take a while to happen, but autoconfig is always
> the goal, at least for trivial cases.  But after the current code
> gets some real world usage.
Aye. The autoconfig / autoselect will be an additional benefit, but we 
should be satisfied with the data processing before we go in that direction.
> > What NTPD is really missing is a standard way for IO configuration,
> > including the device names. Not to mention baud rates and alike...
> > But adjusting all drivers to such a new schema would be quite some
> > work, I think.
> Yup.  Long overdue work.  Someday.
>
> > > > I downloaded 4.3.10 and am building it on a Raspberry Pi, connected
> > > > to the GR601W.  Will take a few more hours to complete :-)
> > > Cool.  You'll note I have been adding RasPi comments to the INSTALL
> > > file.
> > >
> > > That prolly needs to be forked, maybe: INSTALL.raspi.
> > Attention -- I have not yet integrated the updated driver into
> > ntp-dev or -stable! I wanted to be sure that there are no obvious
> > bugs in the code...
> Understood.  No gpsd doc mentions it yet, but the word is out and
> devs want to play with it.
>
> > Gary, did you find anything that needs repair soon? If not, I would
> > create a bug/featrure report and have my changes merged.
> gpsd needs to send you the precision since there is no way for you
> to guess it.  On my short list to add to the json.
That would be cool... Plucking an estimation from thin air is always a 
shaky thing at best, and getting a solid number from GPSD makes much 
more sense. Plus, I can remove some code.

Just keep me informed on the progress!  I would like to support that 
feature before I have my changes integrated. I hope we can let it escape 
into the wild afterwards, so more people can play/work with it to 
provide more feedback.

@Sanjeev: If you want to do some alpha testing, drop me a note, or have 
a look at

http://people.ntp.org/perlinger/for_test/refclock_gpsd/ntp-4.2.8p1-gpsdjson.tar.gz

This is not exactly a 4.2.8p1 tree; it's based on it but contains my 
latest changes I gave to Gary. It should build out of the box, but doing 
so in a separate tree is recommended.

Cheers,
    Pearly.