Fix .testing Makefile with WORKSPACE

[herrwang0]

When WORKSPACE is specified, the recursive make for target code (for regression tests) does not work properly. This PR fixes this bug.

[marshallward]

Can you describe the problem you are seeing? I just compiled and ran the test with the current Makefile and following settings:

WORKSPACE = $HOME/xxx
DO_REGRESSION_TESTS = true

and everything passed.

(Note that WORKSPACE has been phased out in favor of BUILD and WORK, which decouples the build and run directories, but I still don't understand why we need to set WORKSPACE in this situation.)

(Edit: I also tried WORKSPACE = $HOME/a/b/c/d to check for a directory tree depth issue, but it still seems to work.)

[herrwang0]

This is actually a bit tricky.
The problem occurred when I was using the command-line variables. So the variable WORKSPACE got passed to target's make call (line 285). The config.mk method does not have this problem, because, well, TARGET_CODEBASE directory does not have a config.mk, so WORKSPACE is always ..

[marshallward]

because, well, TARGET_CODEBASE directory does not have a config.mk, so WORKSPACE is always ..

OK, that explains the problem. Thanks for working that out.

This is a very nice solution, and I'd suggest the following small modification:

 # Target codebase should use its own build system
 $(BUILD)/target/MOM6: $(BUILD)/target FORCE | $(TARGET_CODEBASE)
-       $(MAKE) -C $(TARGET_CODEBASE)/.testing build/symmetric/MOM6
+       $(MAKE) -C $(TARGET_CODEBASE)/.testing BUILD=build build/symmetric/MOM6
 
 $(BUILD)/target: | $(TARGET_CODEBASE)
+       $(MAKE) -C $(TARGET_CODEBASE)/.testing BUILD=build build/symmetric/
        ln -s $(abspath $(TARGET_CODEBASE))/.testing/build/symmetric $@

The first change is to replace WORKSPACE with BUILD. This is because we have phased out the WORKSPACE macro.

The second change also helps to prevent a dead build/target symlink by also build its target directory.

[herrwang0]

@marshallward Thanks for the explanation! I modified the PR following the suggestion.

[marshallward]

Oops! My "innocent" suggestion broke the MacOS tests. I think it's something about GNU vs BSD Make and how end-slashes are handled.

Anyway, let's leave that for another time. I think just using BUILD in place of WORKSPACE should be good enough. Sorry for making this more complicated than it needed to be.

[herrwang0]

Oops! My "innocent" suggestion broke the MacOS tests. I think it's something about GNU vs BSD Make and how end-slashes are handled.

Anyway, let's leave that for another time. I think just using BUILD in place of WORKSPACE should be good enough. Sorry for making this more complicated than it needed to be.

That's good know! Removing the end-slash and using build/symmetric seems to work. But I am not sure if that'd be the general solution (my rough understanding is BSD doesn't like end-slashes). Does .POSIX: do anything?

Anyway, I will go ahead without the mkdir -p fail-safe.

[marshallward]

Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/mom6ci/MOM6/-/pipelines/25044 ✔️

[marshallward]

Thanks for the additional revision.

.POSIX: could help with the problem above, but I don't think the slash usage is consistent enough in the file to get it all working, so probably best to just hold off and do a proper POSIX compatibility check some time in the future. (Or not... 🤷)