Re: [gpsd-dev] PPS in 3.11

[Gary E. Miller]

Yo Eric!

On Tue, 26 Aug 2014 23:14:23 -0400
"Eric S. Raymond" <address@hidden> wrote:

> Gary E. Miller <address@hidden>:
> > Uh, oh.  I see Miroslav's patch broke some unstated assumptions.
> > 
> > His patch used the CALCULATED time of the last NMEA timestamp
> > (last_fixtime_real).  Not the RECEIVED time of the last NMEA
> > timestamp (last_fixtime_clock).
> > 
> > That introduces a race between PPS and NMEA time and breaks the 
> > assumption that the PPS time is on the edge of the next second.
> > 
> > I can also see how some misleading variable names and unneeded
> > copying and renameing of variables is confusing things.
> > 
> > For example 'drift' implies to me a delta or rate of some kind, like
> > the drift of a boat.  Instead it is an exact benchmark point in
> > time.
> > 
> > I'm gonna have to noodle over it a bit, but it is clearly broken.
> 
> I see a short-cycle release in our future, then.

Possibly.
 
> And if it's broken, it's not very.  I've been live-checking the
> results on my GR-601-W and the observed offset has been stable to a
> small number of milliseconds through the patch.

And that is why you do not see it!  Miroslav's patch is to improve
slewing.  When you are stable you do not see the effect.

>  It's possible the
> reason Miroslav's patch doesn't look busted is that for reasonable
> hardware like the uBlox it actually isn't.

It is a race, some hardware will have it, some not.  The sort of thing
a bit of heat will affect.  Or sticking your finger on an RS-232 signal
wire.

> Rename whatever varuiables and structures you like.  That code is so
> complex that removing barriers to comprehensiion has to be top
> priority.

I'm noodling on it.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97701
        address@hidden  Tel:+1(541)382-8588