GitLab

Documentation improvements: the linear regressions should have more standard naming and documentation

The compiler does not matter, the libc does!

execinfo is a Gnu C library extension, this is not available for example on musl
libc

Now relying on a FindProj.cmake macro in order to find libproj (there was no way
to use the files provided by the libproj project, therefore we have our own file
for now), renamed all instances of PROJ4 into PROJ (as this is the new name of
the project from version 7 onward) including in the ini files (so the ini files
MUST be migrated). This should make a smoother transition to the newer version
of libproj at a latter time (since many distributions still ship proj version 4
or 5, we might have to keep using the deprecated api for a long while).

The Posix signal SIGTERM is now trapped by meteoio_timeseries in order to
trigger the printing of a stack trace. This means that something like "kill
{meteoio_timeseries process}" stops the process and prints a stack trace (making
it easier to understand where a seemingly stuck process was). This will be used
by operational systems in order to kill processes that might have been running
for too long and still getting some useful debug information (we've had multiple
long running processes in the last few days, I suspect because of a unreachable
mount, this should help in validating or not this hypothesis on the running
system if this happens again).

Fixed a bug when processing meteoFilters with timerestrictions: in some cases, ovec was not initialized...

In order to ease time series comparison (between the reference and the test), a
new version of the input dataset has been created so the rounding is done the
same way for both data paths (input->test and input->reference data) otherwise
there were some small differences...

Now the SMET specifications have been edited  in order to rename OSWR into RSWR.
The SMET plugin has also been edited as well as the example input data and the
input data for the tests.

But this support has been tweaked: now the CSV_FIELDS must be defined with one
filed named PARAM that will be replaced by the PARAM name extracted from
elsewhere (either file name or metadata in the header). Therefore there is much
more flexibility (it is even possible to read n fields + PARAM).

A while ago the CSV plugin learned to read Cartesian coordinates,
and then how to scan scan individual files for coordinates. Now it can
read coordinates from the csv files, and coordinate system parameters
from [INPUT]. I hope this has no undesired implications... also not sure
the final "else" clause is even needed, but if not it at least gives
good debugging info if something gets messed up.
(issue #794)

When writing the computed offsets to a file, the file is first deleted (in the
constructor) and then appended to (so multiple calls to the filter append their
data).

in ProcShift, the sampling rate is now rounded when computing the offsets so the generated timestamps are more user-friendly

Moreover, small code cleanup in GRIBIO and larger change in ProcessingStack (the
SHIFT filter can actually change the number of points, so this had to be handled
including for DATA_QA).

Moreover, for increased generality the offset_spec::offset has been renamed into
offset_spec::value.

As well as code cleanup