Wrong argument order in cdeps buildlib

[billsacks]

I think the argument order on this line is wrong:

CDEPS/cime_config/buildlib

Line 130 in c999578

caseroot, libroot, bldroot = parse_input(args)

As noted in ESMCI/cime#4034, this is using argument parsing that is appropriate for components, which (confusingly!) differs from the order of arguments in calls to libraries' buildlibs. CDEPS is built using the library code, not the component code.

See (for example) buildlib.csm_share for what I think is a correct argument parsing.

This is only relevant if CDEPS's buildlib is called as a subprocess instead of as a function. Typically it is called as a function, so this isn't critical, but it would be good to fix it to avoid subtle issues in the future.

@jedwards4b @mvertens @uturuncoglu

[mvertens]

@jedwards4b - can you please look at this. Has this been resolved?

[jedwards4b]

@billsacks please consider https://github.com/ESMCI/cime/blob/master/scripts/lib/CIME/buildlib.py#L41
in light of that this line looks correct to me:
https://github.com/ESCOMP/CDEPS/blob/master/cime_config/buildlib#L152

[jedwards4b]

Note that the buildlib.csm_share is using it's own copy of parse_command_line and not the one in CIME/buildlib.py

[billsacks]

@jedwards4b - The (confusing) situation is that the argument order differs when CIME calls component buildlib scripts vs. library buildlib scripts. Typically the library buildlib scripts are called as python functions rather than subprocesses, so it doesn't really matter, but if they were called as a subprocess, it would matter. It looks like https://github.com/ESMCI/cime/blob/master/scripts/lib/CIME/buildlib.py is intended for component buildlibs, not for library buildlibs. So library buildlibs need their own parse_command_line, and cannot use the version in CIME's shared buildlib.py.

There are some additional details in ESMCI/cime#4034

[billsacks]

I thought I'd try to wrap this one up this afternoon, since I thought I knew what was needed. It turns out it's a little trickier than I first realized because the same buildlib script is used for both (1) building the CDEPS shared library, and (2) building the individual data models. The problem is that the argument order differs in those two cases, so fixing one breaks the other.

It looks like there is essentially no code shared between these two situations, though, so my plan is to introduce a new buildlib_comps that is used for building the individual data models, then trim each one down to only contain what's actually needed.