Namelist input checking (subroutine chk_var) doesn't print error message?

[MichaelWhitall]

Hi folks,

In the branch for one of my issues:
MetOffice/lfric_apps#247
vn3.1_pc2_homog_reversible
I just tried running a couple of rose-stem apps without running the branch upgrade macro on the rose-stem apps first. This should have triggered a failure as the multi-option switch i_pc2_homog_g_method added to the namelist had not actually been added to the rose-stem rose-app.conf files, so the value of this setting would not match one of the expected allowed values.

As expected, my test did fail when loading settings from the namelist. However, it did not print any error-message to explain why it failed! The old UM subroutine chk_var is still used in LFRic to check that namelist inputs match their allowed values or ranges. The traceback for the failing azspice gnu job I ran (run_lfric_atm_nwp_coma9-C12_azspice_gnu_fast-debug-32bit-crun0) did at least point to chk_var (and the contained call to ereport), but the error message did not print. The failing ex1a cce job didn't even give a relevant traceback (run_lfric_atm_nwp_coma9-C12_ex1a_cce_fast-debug-32bit-crun0).

Any ideas why the namelist checking is not printing useful error messages? Or alternatively, can psyclone autogenerate working namelist checking code directly from the meta-data, removing the need for the old chk_var calls that seem to be broken?

Cheers!
Mike

[Andrew Coughtrie (andrewcoughtrie)]

Ah this is because when you're running an MPI job messages all get output to the PET files (one per rank) in the working directory. These files aren't visible in the CYLC Gui. So when a job fails it is always worth looking at the PET files and/or the xios* files in the working directory. In your case you have this:

20260330112520.358+0000:P1:ERROR: CHK_OPTS_MOD:CHK_INT_ARR called by CHECK_RUN_CLOUD
Problem performing defensive checks (CHECK_RUN_CLOUD)
i_pc2_homog_g_method,(-32768), has failed defensive checks, [1,2,4].

in:
<home_dir>/vn3.1_pc2_homog_reversible/run1/work/1/run_lfric_atm_nwp_coma9-C12_azspice_gnu_fast-debug-32bit-crun0/PET1.lfric_atm.Log

[Andrew Coughtrie (andrewcoughtrie)]

P.S. there is a flag to reduce the output to only rank 0 if you're doing big runs and you don't want thousands of files.

[MichaelWhitall]

Awesome thanks Andrew, that's really god to know about the PET files; I'll look there next time :)

Cheers!
Mike

[iboutle]

There's probably a good question here on whether the PET0 output should be diverted to the job.out/job.err files, which is essentially what happens in the UM and is therefore the place most people look

[Andrew Coughtrie (andrewcoughtrie)]

There are a number of different improvements that can and should be made to the logging output etc. there is however no resource to do this work at the moment as simple as it might be, it falls a long way down the importance scale in current work.