set CC env variable

[bknight1]

Fixes issues with finding mpi

[gthyagi]

@bknight1 please change main to development here too. Thanks.
@julesghub please merge this PR as well. Thanks.

[julesghub]

I'm unclear if this approach is the best.
As I read this we are only overwriting CC so cython can compile the extensions correctly, is that correct?
This doesn't happen on my lappy. Any specific reason why it happens on yours?

[bknight1]

@julesghub This was to fix a changein mpi4py>=4 where we can't get the CC env variable from mpi4py any more. UW won't compile if the CC env isn't set as it can't find the mpi install.

[bknight1]

Could add a flag to only set it if it's not found?

[julesghub]

interesting, I'm running mpi4py==4.0.1 and have no issues.
I wonder if this a mac thing - I can test more tomorrow.

[bknight1]

Is your CC variable already set in terminal?

[julesghub]

Nope.

[bknight1]

Okay, fair enough. Thyagi and I discussed in detail on #242

[julesghub]

mpi4py/mpi4py#532
Relevant to the discussion. Why mpi4py no longer needs to be constraint to a single implementation of MPI.

[julesghub]

ok, I gave it a go and I think this is the issue.
I installed mpi to system directories. But on mac people often install mpi with homebrew, potentially not in a system dir.
Hence cython can't find the mpi implementation in sys dirs and dies compiling uw3.

@bknight1 and @gthyagi are using homebrew?

[bknight1]

Yeah, I used homebrew to install ompi

[julesghub]

Yeah, I used homebrew to install ompi

What's your path?
Something not in system standard lib or inc dirs

[bknight1]

It's the homebrew dir:/opt/homebrew/bin/mpicc

which is included in my $PATH env variable, so should be picked up?

[julesghub]

yeah the mpicc will be picked up but not the mpi.h, which was the main issue I believe.
So I think this PR is good to go - will accept it after lunch.

[gthyagi]

I am using manual install /Users/tgol0006/manual_install_pkg/openmpi-4.1.6/bin/mpicc. Even then I encounter that error in Mac.

[julesghub]

@gthyagi you need to use mpicc, not mpirun.

[bknight1]

Looks good!

[julesghub]

I add a rework to effectively get what petsc was compiled with, not directly get mpicc. Show work all the same.
This approach is slightly safer for those funny hpc issues where we have had different mpi implementations for UW and PETSC (and hdf5). Will test using conda soon. Could @gthyagi and @bknight1 please test and report back.
Thanks!

[gthyagi]

I am now able to install the code without any MPI issues using this fix. Please merge it.