Re: [gpsd-dev] Planning a 3.11 snap release

[Greg Troxel]

"Eric S. Raymond" <address@hidden> writes:

>> I think that if the first two things are addressed, gpsd pps support
>> will likely work on NetBSD.  (I have working pps support via
>> sys/timepps.h in ntpd with a GR601-W, using my soon-to-be-upstream USB
>> pps changes.  Real serial pps appears in NetBSD sources in 1998, and was
>> then spiffed up tracking the drafts of RFC2783 over several years.)
>
> I think it would be a nice win to have PPS support working on any non-Linux.
>
> I've pushed an SConstruct change that should fix this if your analysis
> is correct. It clears pps only if pps is set and neither timepps.h nor
> TIOCMIWAIT are available, and it does include sys/time.h where you
> suggest. The resulting build passes regressions here.

With your changes, regression seems ok on NetBSD, and live-test without
pps still works.  However, pps does not work because the code complains
about line discipline changes, which is perhaps a clue that the pps code
is really linux kpps rather than RFC2783.  (RFC2783 gives a set of
procedure calls, and none of them are about line disciplines.  However,
they are in BSD inlined to ioctls.)  I will have a look and I may be
able to just #ifdef linux around the Linux-specific code.  Or maybe
there's something else wrong.  But it's far easier to debug now - thanks
very much for those changes.

>> And finally, I built 3.10 (from git tag) on NetBSD 5 and 6 (i386), and
>> there are only some apparently-timing-based sporadic regresssion
>> failures.  I also live-tested with a GR601-W on NetBSD 5 (pps does not
>> work, but the rest seems ok).
>
> Thanks, that's nice to know. Maybe you'll even have PPS working soon!

And also I built both 3.10 and release-3.10-27-g84ea966 on OS X 10.7,
and regression passed.

pgpAh0eI8EYbX.pgp
Description: PGP signature