Conversation
| @@ -69,6 +69,8 @@ def get_unix_pip_env(venv, venv_source, execroot): | |||
| env["F77"] = env["F90"] = os.path.join(os.getcwd(), ARGS.fortran_compiler) | |||
| env["AR"] = os.path.join(os.getcwd(), ARGS.archiver) | |||
| env["CC"] = os.path.join(os.getcwd(), ARGS.compiler_executable) | |||
There was a problem hiding this comment.
Would it be reasonable to do the simpler thing of adding env["MPICC"] = env["CC"] here and not doing any of the other changes?
There was a problem hiding this comment.
My only concern is that there might be build scripts that assume that the presence of the environment variable MPICC that it was a valid mpi compiler wrapper and hence try to enable mpi features even if the compiler doesn't work as an MPI compiler.
The reality is though that I am setting it to the same value as CC in _add_vpip_compiler_args and its working for us so I guess its good enough? If we come across any packages this causes an issue we can revisit this with a similar approach to the fortran feature (we're already doing this to select the mpi compiler wrapper for normal cc_* targets).
There was a problem hiding this comment.
Right, my proposal and this change seem totally equivalent, which is why I asked.
|
|
3 participants
Add an argument to vpip to allow passing in an mpi compiler to be set in the environment.
This is needed to compile mpi4py.