Beta feedback (and my response) from Hans Gleisner (DMI).
We clearly have a problem with the ropp_io user-test scripts & files. I found
the same issue when I was checking the packages at home under Cygwin last night.
It doesn't prevent the build or running of the tools, but it hardly gives
the user a warm glow that all is well! I will deal with the corrections
pre-release, if not before.
Hans,
I have just installed ROPP -- tools, io, fm, and 1dvar -- on a Linux machine
> (using gcc/ifort) and under Windows/Cygwin (using gcc/g95).
>
Thanks for the good news!
To pass the ROPP IO test on the multi-file ("mropp_test.nc") I need
> to add the attribute 'phase_method' to the ROPP file header. This
> attribute is already there in the single file ("ropp_test.nc") but not in
> the multi-file. If this attribute is not in the file header, an error is
> generated
> by 'ropp_io_read_ncdf_get.f90'.
>
I found this problem myself only yesterday. Our proper solution will
be to re-generate the test files provided with the IO package
(and I've modified the test script to _first_ do the single-->
multifile test (so generating a multifile) and use that as the input
to the multi-singlefile test afterwards. So no need to supply
a multifile in the package (for this test, anyway)
If you have the time and interest, if you cd to the data directory
and do the following:
$ ../build/ropp2ropp ropp_test.nc ropp_test.nc -m -o mropp_test.nc
this will re-generate the multifile from the base test file
(used twice), which I think is ok. The test script should then run.
It's possible the last test fails to find a file called
oc_(something).nc because the script expects a particular
input file with a certain content. You may need to edit
the script to the name actually created. I've modified the script
to make this a variable.
> How is the ROPP file format actually defined? Which global attributes
must be in the header and which are optional?
>
They are currently all mandatory in the v1.0 format definition
(in that the reading code expects them all, and doesn't check
first.) This is something we need to reconsider, and any new
parameters we might add, like 'thin_method', would have to be optional
from the start.
> Concerning the testing of ROPP FM: the file "FASCOD_Scenarios.nc",
> which the test program tries to read, is not there. And the files that are
> there also lacks the 'phase_method' attribute.
>
Yes, same problem as above, which needs re-generation of the supplied test files (and including the missing one!)
We've obviously not kept the user-testing files in sync with our internal test system. We'll correct this for the full release.
Many thanks for this feedback, which is very timely for the
upcoming review by EUMETSAT. In some ways, it's good to have
problems noted by beta testers which are relatively easy to solve
prior to proper release - it shows the system works!
best regards,
Dave
Test files in ropp_fm updated to v1.0 and FASCOD_Scenarios.nc added to distribution list prior to release of ROPP-1 v1.0.
All test files updated to v1.1 format and committed at [1308].