[gpsd-dev] Half the port problems are solved; seeking help with RTCM2 driver

[Eric S. Raymond]

Hard confirmation will await the scheduled buildd run of GPSD
regression tests on all porterboxes, but I believe from results on I
see on an s390x machine that about half of the port problems I listed
on 17 April are now solved.

It looks like all the regression failures on the italk-binary,
trimble, and uBlox tests were due to a point bug in a very obscure
place; the initializer for our internal geoid separation table was,
inexplicably, declared char even though it contained signed integer
values.  I suspect that the architectures on which these tests failed
were those on which char is unsigned.

It appears that all of the remaining port failures are due to some
strange type punning in the RTCM2 driver.  That module (with the ISGPS
code it uses) is the blackest magic in the GPSD codebase; I
successfully modified it from contributed code, but I don't fully
understand it.

I need help from some very experienced C programmer with lots of
experience in weird bit-bashing and troubleshooting endianness
problems, to check my assumptions and help me fix this. If you're such
a person and up for a challenge, read isgps.c and driver_rtcm2.c and
find me on the #gpsd channel of freenode.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>

"Gun control" is a job-safety program for criminals.