unit tests failing to build

[Adrian Lock (Adrian-Lock)]

My PR MetOffice/lfric_apps#504 is adding new namelist parameters to the existing intial_temperature namellst. I've currently broken several build unit_test jobs that are complaining my new namelist variables are missing, eg build_gungho_unit_tests. Looking in the code there's what appears to be a subroutine call with the namelist items of the one I've added to given as subroutine arguments. From which I'm guessing I need to add my new namelist variables? Is that correct? If yes, do I have to do this manually or have I failed to apply the upgrade macro correctly?

[Andrew Coughtrie (andrewcoughtrie)]

Hi Adrian,

This is a consequence of the need to be able to set module scope config information for subroutines that are being unit-tested. We have what are called config feigners which set these module scope variables so that the function/subroutine can use them when being unit-tested. In your case as you've added metadata you'll need to update any intitial_temperature_config feigners (here for example rhs_sample_eos_kernel_mod_test.pf), you will essentially be creating hard coded config for the unit-tests which will then be used by the kernel (even if you haven't modified the kernel the feigners have no concept of default values so they have to be set up).

We hope to get rid of feigners but the work to do so is dependent on removing module use of config, this is a big task as it is very wide spread.

Hope that makes sense.

Andy

[Adrian Lock (Adrian-Lock)]

That's fine, thanks for explaining. Can I just check that I should be changing what's there now, eg:

call feign_initial_temperature_config( bvf_square=bvf_square, &
theta_surf=theta_surf, &
pert_centre=pert_centre, &
pert_width_scaling=pert_width_scaling, &
perturb=perturb_none, &
profile_size=profile_size, &
profile_data=profile_data, &
profile_heights=profile_heights )
with
call feign_initial_temperature_config( bvf_square=bvf_square, &
theta_surf=theta_surf, &
pert_centre=pert_centre, &
pert_width_scaling=pert_width_scaling, &
perturb=perturb_none, &
profile_size=profile_size, &
profile_data=profile_data, &
profile_heights=profile_heights &
theta_pert_start=0.0_r_def, &
theta_pert_end=0.0_r_def, &
theta_pert_size=0.0_r_def)
I'm guessing that the existing setting of perturb=perturb_none means it's using perturb='none' and so my new variables won't be used

[Adrian Lock (Adrian-Lock)]

And also to check that I really will need to manually add these lines to all 19 test.pf files that reference this namelist?!!!

[Andrew Coughtrie (andrewcoughtrie)]

Unfortunately yes you will need to update them all but yes you can do it like that or set parameters, the values you put don't really matter but ideally should make sense. We normally try to discourage the use of 0 in tests (there are many ways 0 can appear that aren't what was intended) but I'm guessing these aren't being used at all so it shouldn't matter.