[gpsd-dev] New slowdown option for regression tests

[Eric S. Raymond]

Some of you have seen flaky (non-reproducible) failures compromising
the GPSD regression tests; this seems especially prone to happen on
heavily loaded machines.

I don't know what causes the failures (which look like segments of the
regression loads were not fed to the daemon) but suspect that race
conditions in kernel pty layers are implicated.  The GPSD test
framework works those layers exceptionally hard, and I wouldn't be
surprised if it wanders into regions of the code's behavior space
that are otherwise seldom or never reached.

I have added a 'slow' option to the SCons build to be used with the
"check" target.  If you specify slow=yes, this will get passed to the 
test framework, which will insert additional delays.  (The sizes of
the additional delays are constants in gps/fake.py.)

The effect on my desktop box is to slow down testing by about 4x, but
I think there's a good chance the delays may banish the failures.

Bernd: You should, in particular, try this option out on the Debian
porterboxes.  It will probably banish all failures except the 
AIS Type 6 mismatches. If it only makes them less frequent, try
upping the SOWDOWN delays.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>

What if you were an idiot, and what if you were a member of Congress?
But I repeat myself.        -- Mark Twain