Need a better way to pass global communicator (w/o I/O tasks) to physics

[climbfuji]

Description

When the CCPP physics were introduced, it was required to pass the global communicator that includes all compute tasks, but not the I/O tasks, to the physics for the purpose of broadcasting lookup tables at physics init time etc.

This was implemented by @climbfuji by exporting commglobal from the dycore to populate a member of the GFS control DDT. As @bensonr pointed out, we should be able to get this communicator directly from ESMF (which passes it to the dycore in the first place).

Solution

Pass the global communicator that includes all compute tasks, but not the I/O tasks, from ESMF directly to atmos_model.F90 and stop exporting commglobal from the dycore.

Alternatives

Keep the current, inelegant solution.

Testing:

n/a

Dependent PRs:

n/a

[junwang-noaa]

@bensonr the forecast tasks communicator is passed to fms_init in the code here, can we get the communicator out from mpp or fms functions?

[bensonr]

@junwang-noaa - we could, but I think the communicator should be pushed down to atmos_model_init as an argument. It should then be included as an element of the GFS_init_type and sent down to the physics via the Init_parm ddt - and not as yet another argument to GFS_Initialize (commglobal) At the same time, the argument to GFS_Initialize being satisfied via mpp_npes() should also become an element in GFS_init_type. The whole reason for this particular ddt is to keep from having to make interface changes as you find another variable you'd like to pass down into another component or sub-component, in this case the physics initialization. If this is allowed to continue, you could quickly find yourself back to the olden days of 100+ arguments going between components and sub-components.

[junwang-noaa]

Code changes are committed in PR#345, close issue.