Re: [gpsd-dev] buzzkill is gone, please re-test

[Eric S. Raymond]

Gary E. Miller <address@hidden>:
> On Thu, 29 Jan 2015 15:21:39 -0500
> > The effect should be that device reports come in in N-character chunks
> > except for the last N characters of a packet burst, which could be
> > delayed by 0.1sec.  The number of select(2) returns per cycle should
> > drop by N (13 for SiRF binary, 16 for NMEA 0183).
> 
> This will affect all the ntp fudge values based on serial time.  The ntp
> time is based on the time of the first message in the cycle.  So this
> should be noted as a behavior change in the upgrade notes.

Let's see if this change fixes the pending problem (the occasional data loss
Hal is reporting) first.  If not, I may back it out - no point in adding latency
to no good purpose.  Bulletproofing against buggy select implementations
would be nice, but probably isn't worth screwing up time service to do it.

Besides, I'm not sure you're right about this, and if you are the Time Service
HOWTO needs a fix.  I thought the fudge was supposed to be the offset between
time of the received PPS pulse and actual top of second.  Changing the latency
of the other, in-band report should not affect this value. But perhaps I'm
just confused; expkain it to me so I can clarify the HOWTO.
 
> Everyone should be using PPS anyway...

Agreed.
 
> > Please test.
> 
> Testing...

Please check to see if the fudge value is actually disturbed, too.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>