Re: [gpsd-dev] Broken Regressions

[Fred Wright]

On Mon, 8 Apr 2019, Eric S. Raymond wrote:

> Fred Wright <address@hidden>:
> > It looks like GPSD didn't handle the week rollover as well as most of the
> > receivers did, since there are now ten failing regression tests plus a whole
> > bunch of "gpsd:ERROR: date is negative!" messages.  Fortunately, I think
> > this only affects the tests and not the normal realtime operation.
> > A couple of spot checks look like it's hallucinating week-rollover issues on
> > some tests.
>
> Yes, this was expected - I even lrft a comment somewhere noting that the
> test suite needs to be rebuilt after an era rollover.  There is no help for
> this, it is an unpresent and inevitable consequence of the way the GPS
> calendar was designed.

Well, rebuilding the test data to consider dates in 2037 to be "correct" 
is insane (for the next couple of decades, anyway), and far worse than the 
current broken leap-second handling.  It also doesn't fix the negative 
date problem (unless the test framework is modified to handle "expected 
errors").

The can could probably be kicked down the road a ways by using a base week 
that's not aligned to an era boundary, but not having it properly 
parameterized doesn't help.  The *right* fix is to provide a way to give 
GPSD an appropriate reference date/time during tests, rather than always 
basing its behavior on "now".

BTW, the era-aligned wraparound handling is exactly the sort of incorrect 
thinking that's encouraged by the "pivot" terminology.

I plan to look into this, but it might be a couple of days before I have 
time.

Fred Wright