Optionally add links to load_compass_env.sh in test cases

[xylar]

This is specified either in the config file or at the command line.

[xylar]

@matthewhoffman, this would add a capability we've already been using a little bit on ocean/develop to have a script that knows about machines we commonly use COMPASS on and which can be sourced to load the appropriate conda environment for COMPASS. An example of such a script is here, though I will need to move it into the ocean subdirectory in a separate PR.

If this functionality sounds useful to you, you will likely want to add a copy of the script to landice and update the version number for the compass conda env as needed. If it doesn't sound useful, that's fine, and the only review I would request from you is to make sure this change doesn't cause you any trouble (either by inspection or by running a landice test case).

[xylar]

@mark-petersen, I am having trouble with other testing I'm doing because of conflicts between develop and ocean/develop. I think our quick-and-dirty approach of modifying setup_testcase.py in ocean/develop has turned out to be a bad move and I will strongly advocate against doing that in the future.

I realized in the process of putting in the changes we have made (which, after reversion of some of my MPI work, is just linking load_compass_env.sh) that there will be issues if different cores try to update the same load_compass_env.sh. Instead, we need one per core and it will be clear which core owns each file and which branch it should be modified in. In #618, I have moved load_compass_env.sh into the ocean directory on ocean/develop.

[xylar]

Testing

I tested by setting up the ocean/global_coean/QU240wISC/init test case with and without the --link_load_compass flag (after merging this branch into #618). Without the flag, there is no load_compass_env.sh link in the test case, as expected. With the flag, the link is there and sourcing it loads the expected compass_0.1.6 environment on Grizzly.

[matthewhoffman]

@xylar , this sounds useful. I will try it out.

[xylar]

@matthewhoffman, keep in mind that the more recent compass_* environments include MPI, and that causes problems if you load things in the wrong order. The order of operations would be to first source load_compass_env.sh and then load the modules for the system compilers and MPI (so they take priority). I know this is a pain but I haven't figured out a better way.

[mark-petersen]

@xylar this is fine with me if you've tested it. I've seen those merge conflicts as well, so it would be nice to merge it in.

@matthewhoffman any comments, or should we proceed?

[matthewhoffman]

@xylar , thanks for pointing this out to me. I don't see any problems with adding this. I'll wait and see how you use it in the ocean before thinking about adapting it for our testing.