The issue was that homebrew's openmpi 1.10.1 does not link to any gfortran compiler. I rebuilt openmpi primarily to include the gcc fortran compiler. Though I think the cleaner way would be to modify the mpif90-wrapper-data.txt to ensure that the compiler points to a working gfortran.

On Tuesday, December 1, 2015, Lawrence Mitchell <lawrence.mitchell@imperial.ac.uk> wrote:

Hi Justin,

On 30/11/15 23:59, Justin Chang wrote:
> Onno, that's strange. Are you sure you followed my instructions
> correctly? Because simply reinstalling Homebrew is not going to fix
> the fortran issue. You need to manually reconfigure openmpi to use the
> latest gcc compilers. First ensure you have the latest gcc via
> Homebrew (I think it should be gcc-5). Then you "install" openmpi
> initially (via brew install openmpi). Finally you need to rebuild
> openmpi via
>
> brew reinstall openmpi --build-from-source
>
> Because by default, the openmpi links to the apple compilers, and for
> some reason mpif90 is not hooked up to any gfortran. Hence you need
> those environment variables I stated.
>
> They do something similar
> here http://stackoverflow.com/questions/26812780/linking-homebrew-compiled-openmpi-or-mpich2-to-homebrews-gcc
>
> This should fix that fortran problem within the PETSc build process.
>
> Spatialindex needs to be rebuilt the same way.

It shouldn't be necessary to rebuild openmpi yourself unless you want
to use the GNU compilers everywhere. Clang (+ gfortran) works fine as
far as I can tell. If you do that, you then won't need to rebuild
spatialindex either.

Cheers,

Lawrence