use system library, if GIT_SUBMODULE=OFF

[dschwoerer]

No description provided.

[ZedThree]

This is failing because we've not yet checked out the submodules, so it tries to use the (non-existent) system versions of mpark/fmt.

As it is, this essentially makes using system versions the default on first checkout, unless the user has cloned recursively (which is not the default behaviour of git clone). This isn't really what we want.

What's the use case you're thinking of for this change?

[dschwoerer]

The use case would that if the user explicitly sets GIT_SUBMODULE=OFF and the submodules aren't there, that then the system ones are used.
I think it should have been:

if(GIT_SUBMODULE OR (EXISTS externalpackages/mpark.variant/CMakeLists.txt))
  option(BOUT_USE_SYSTEM_MPARK_VARIANT "Use external installation of mpark.variant" OFF)
else()
  option(BOUT_USE_SYSTEM_MPARK_VARIANT "Use external installation of mpark.variant" ON)
endif()

[ZedThree]

Ok, I see now. Still, I'm not sure of the advantage of this over setting BOUT_USE_SYSTEM_* explicitly, aside from being a little bit less typing. Setting GIT_SUBMODULE=OFF doesn't really mean "don't use the submodules", it means "don't update the submodules".

What if we made the error more explicit: "mpark_variant not found. Either set GIT_SUBMODULE=ON or use a system version (BOUT_USE_SYSTEM_MPARK_VARIANT=ON)"? Perhaps also a better name for GIT_SUBMODULE, BOUT_UPDATE_GIT_SUBMODULES?

Alternatively, having a BOUT_USE_SYSTEM_LIBRARIES option and using
CMakeDependentOption to set the specific BOUT_USE_SYSTEM_* variables. Note that this will correctly update the dependent options if BOUT_USE_SYSTEM_LIBRARIES changes value.

[dschwoerer]

Sure, but if they are nether present, nor the user don't want to install them, I thought it makes sense to check whether they might be present. I don't have a strong opinion, and am happy to explicitly pass the flags if needed. But I thought it might be nicer to do that automatically, i.e. less typing and just do the right thing.

I found the aggressive caching from cmake very annoying, and always rm -rf * before running cmake .. again, so don't know about those caching issues.

Anyway. I am happy if you close this, just thought it might be neat.