case.build --clean lnd and --clean-depends lnd don't work for shared lnd lib

[billsacks]

When running ./case.build --clean lnd or ./case.build --clean-depends lnd, the land build is not cleaned if using a shared land build (as is standard for CESM). It looks like the problem is that COMP_LND isn't passed to the make command if you're doing a clean, and some Makefile logic depends on that in order to determine where the land build resides.

Steps to reproduce:

1. Create a case in CESM that includes CLM (e.g., an I compset; e.g., SMS_D.f10_f10_musgs.I2000Clm50BgcCropQianRsGs.cheyenne_gnu)

2. Run ./case.build if not already done

3. Run ./case.build --clean lnd and/or ./case.build --clean-depends lnd.

4. Observe that the relevant build files have NOT been removed (e.g., somewhere like bld/gnu/mpich/debug/nothreads/mct/mct/noesmf/clm/obj; I'm not sure if that's exactly the right path on cheyenne - e.g., the mpich might be something else)

5. Optionally, in the Makefile, introduce the following diffs to further observe the issue:

   @@ -991,6 +991,7 @@ clean_dependsesp:
           $(RM) -f $(EXEROOT)/esp/obj/Srcfiles
   
    clean_dependslnd:
   +       @echo $(COMP_LND) $(USE_SHARED_CLM) $(LNDOBJDIR) >> ~/makeout
           $(RM) -f $(LNDOBJDIR)/Srcfiles
   
    clean_dependscsmshare:

   then rerun ./case.build --clean-depends lnd and take a look at the contents of ~/makeout: you should see that COMP_LND is an empty string, USE_SHARED_CLM is FALSE, and LNDOBJDIR points to the location that's used in the non-shared-build case.

I started to look into how to fix this, but wasn't sure exactly how to proceed. My first thought was to add COMP_LND to the _CMD_ARGS_FOR_BUILD variable set near the top of build.py:

cime/scripts/lib/CIME/build.py

Lines 12 to 18 in 1529bac

	_CMD_ARGS_FOR_BUILD = \
	("CASEROOT", "CASETOOLS", "CIMEROOT", "COMP_INTERFACE",
	"COMPILER", "DEBUG", "EXEROOT", "INCROOT", "LIBROOT",
	"MACH", "MPILIB", "NINST_VALUE", "OS", "PIO_VERSION",
	"SHAREDLIBROOT", "SMP_PRESENT", "USE_ESMF_LIB", "USE_MOAB",
	"CAM_CONFIG_OPTS", "COMPARE_TO_NUOPC", "HOMME_TARGET",
	"OCN_SUBMODEL", "CISM_USE_TRILINOS", "USE_ALBANY", "USE_PETSC")

But then I noticed that, when doing the build, these COMP values are passed in via the environment:

cime/scripts/lib/CIME/build.py

Line 451 in 1529bac

os.environ["COMP_{}".format(comp_class)] = comp

So that made me wonder if the preferred method would be to add the necessary environment variables when calling case.build with the --clean or --clean-depends argument.

@jedwards4b or @jgfouca can you provide any guidance here?

[billsacks]

Side note: given how frequently issues arise with the shared land build: Long-term, I am tempted to suggest getting rid of this mechanism, and instead expanding the mechanism that @jgfouca has started to put in place to share the entire executable between tests. We would need to think more about how to leverage this from our test suites, to set up build subsets that can use a single build (because we can't get away with just one build: we need separate builds for debug vs. non-debug and a few other attributes).

[billsacks]

From CIME telecon: preferred way for anything new is to use _CMD_ARGS_FOR_BUILD. (It should be fine to pass COMP_LND in both ways as long as they're consistent.) I'll do that.