Previous formula was not compiling anymore.
Scotch needs more recent POSIX standard implementations of pthread than OSX supports.
Disabling pthread in Scotch fixes the compilation
Fix and Update scotch
Is the pthread issue specific to the new 6.x version of Scotch and is there an upstream discussion or issue that we can track to keep tabs on this?
This post in stackoverflow answers the question about pthread and missing barrier functionality
The previous version of scotch (version 5.1.2) did not make use of pthread_barrier, and worked with the PTHREAD flags ON, so YES, this is specific to the new 6.x version.
The threads don't offer substantial advantage, and mpi is used for distributed graph partitioning anyway. No functionality is lost.
However, compilation of version 5.1.2 was not working any longer due to an unrelated problem:
Something is now wrong with the MPI environment.
Some checks showed that ENV['MPICC'] equals 'nil'. In other words the mpi_path variable in the function 'modify_build_environment' is nil (which mpicc --> nil), while this returns the correct path in the 'satisfied?' function. Perhaps the introduction of superenv has something to do with it? I now trust that the correct mpicc is found without specifying the path.
The MPI issue is related to superenv and is being tracked in #16409. Looks like we'll pull a workaround soon.
Is MPI working now that #16409 has landed?
scotch: bring mpi env back
Yes, the mpi-part works again now. Thanks! Still needs the pthread disabled though for now.
scotch: Update to 6.0
Also disables pthreads as OS X doesn't implement the required optional parts of
the POSIX spec:
Signed-off-by: Charlie Sharpsteen <email@example.com>
Pulled. I squashed the commits since the second really just undoes deletions introduced in the first.