Better CMake compiler/MPI passing

[zbeekman]

I'm wondering how common it is to pass the MPI wrappers to CMake directly. The folks @Homebrew seem to be under the impression that this is somewhat abnormal. I'm going to inquire on the CMake list and see what the consensus is there.

[zbeekman]

Yes, it seems that the findMPI module might be the way to go. At least one person thinks that using the wrappers from CMake is a bad idea: https://cmake.org/pipermail/cmake/2011-June/045037.html

Also see Homebrew/legacy-homebrew#46547

[zbeekman]

Here is some documentation for using the FindMPI CMake module: https://cmake.org/cmake/help/v3.4/module/FindMPI.html

[zbeekman]

Here is the documentation for FindMPI: https://cmake.org/cmake/help/v3.4/module/FindMPI.html

This should probably be used along with CMake's configure_file capabilities to substitute in the right EXE name, flags etc. in the CAF wrapper scripts.

[zbeekman]

OK, 193546d seems to solve this quite elegantly, if I do say so myself. I would like to rework the caf and caf_run scripts to use configure_file and full paths to the compiler and MPI implementation, and also run it through shellcheck. I'll open a new issue for this, I think.