Yeah, hard. I moved the late in the packet processing test for
fix downgrading back into the NMEA driver where it belonged.
But that broke gpsd_error_model(), as well as a ton of regression tests.
That led to not just a few compensating changes. I could not figure
any way to change just a few things at a time. So I ended up with
a pretty large change to the way driver_nmea0183 decides the
current fix mode. Also changes to the error modeling.
Seems like a lot of the error modeling was just plain broken with NMEA,
and likely some other drivers.
The results just feels a lot better.
The end result was just pushed to git head.
Please test.