Recent change in MMM-physics breaks MPAS-Model

[eap]

NCAR/MMM-physics#16 was submitted several hours ago breaking jedi-bundle which depends on MPAS-Model 8.2.1. I'm not sure if the change breaks all existing tags of MPAS-Model or just 8.2.1, although the nature of the issue makes me suspect that all builders of MPAS-model are going to encounter the issue.

Build error

 Building Fortran object mpas/src/core_atmosphere/CMakeFiles/core_atmosphere.dir/physics/physics_wrf/module_bl_gwdo.F.o
/workdir/bundle/mpas/src/core_atmosphere/physics/physics_wrf/module_bl_gwdo.F:215:49:

  215 |                     ,errmsg=errmsg,errflg=errflg)
      |                                                 1
Error: Missing actual argument for argument 'omax' at (1)
make[2]: *** [mpas/src/core_atmosphere/CMakeFiles/core_atmosphere.dir/build.make:2769: mpas/src/core_atmosphere/CMakeFiles/core_atmosphere.dir/physics/physics_wrf/module_bl_gwdo.F.o] Error 1

Suspected root cause

I found this line of code in core_atmosphere/CMakeLists.txt:102-108 which references NCAR/MMM-Physics without a git tag or hash. So any new changes that break API compatibility will end up breaking core_atmosphere.

if(NOT EXISTS ${ATMOSPHERE_CORE_PHYSICS_MMM_DIR})
    set(PHYSICS_MMM_REPO_URL "https://github.com/NCAR/MMM-physics")
    execute_process(COMMAND git clone ${PHYSICS_MMM_REPO_URL} ${ATMOSPHERE_CORE_PHYSICS_MMM_DIR}
                    RESULT_VARIABLE GIT_CLONE_RESULT
                    OUTPUT_VARIABLE GIT_CLONE_OUTPUT
                    ERROR_VARIABLE GIT_CLONE_ERROR)
    if(NOT GIT_CLONE_RESULT EQUAL 0)

[mgduda]

This issue should only affect builds that use CMake, while builds using GNU Make should be unaffected since our Makefile obtains a specific tag of the MMM-physics repository using manage_externals.

@eap Are you attempting to build stand-alone MPAS-Atmosphere, or are you building something like MPAS-JEDI?

[guoqing-noaa]

@eap FYI, if you don't have to use the latest physics_mmm, I have a PR at ufs-community#231 fixing the MPAS-Model (and MPAS-JEDI) CMake building issue.

[eap]

@guoqing-noaa - that looks like a useful change. I'm a little hamstrung on testing it. Our mpas-jedi integration uses MPAS-model v8.2.1 and integrating with the develop branch will require some work. I tried patching a similar change back to 8.2.3 but it appears that some api-breaking changes were included in 8.2.3. I was hoping that we could back-port this to 8.2 but it appears that we have integration work no matter what approach we take.

For now I've disabled mpas-jedi and MPAS-model from our CI system and bundle.