mpich-dragonegg33 @3.1.3 +fortran/mpich-devel-dragonegg33 @3.2a2 +fortran linking fails

[dstrubbe (David Strubbe)]

Linking fails for these ports for me on OSX 10.8.5. openmpi-dragonegg33 builds ok though. This is the error:

:info:build libtool: link: /opt/local/bin/dragonegg-3.3-gcc -pipe -m64 -O2 -Wl,-headerpad_max_install_names -Wl,-flat_namespace -o src/env/mpichversion src/env/mpichversion.o  lib/.libs/libmpi.a -L/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_science_mpich/mpich-dragonegg33/work/mpich-3.1.3/src/mpl -L/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_tarballs_ports_science_mpich/mpich-dragonegg33/work/mpich-3.1.3/src/openpa/src -lpthread
:info:build Undefined symbols for architecture x86_64:
:info:build   "_MPI_Abort", referenced from:
:info:build       _main in mpivars.o
:info:build   "_MPI_Comm_rank", referenced from:
:info:build       _main in mpivars.o
:info:build   "_MPI_Finalize", referenced from:
:info:build       _main in mpivars.o
:info:build   "_MPI_Init_thread", referenced from:
:info:build       _main in mpivars.o

[seanfarley (Sean Farley)]

Should we just remove the dragonegg variants from the compilers portgroup and mpi subports?

[eborisch (Eric A. Borisch)]

Replying to sean@…:

Should we just remove the dragonegg variants from the compilers portgroup and mpi subports?

I'm all for it.

[dstrubbe (David Strubbe)]

Support for dragonegg compilers has been removed from the MPICH port.

[seanfarley (Sean Farley)]

Replying to dstrubbe@…:

Support for dragonegg compilers has been removed from the MPICH port.

Should we remove those variants from the compilers/mpi port group, too?

[dstrubbe (David Strubbe)]

There are some ports that use dragonegg specifically, e.g. OpenBLAS, apparently due to AVX support. So, I think unless we find that is no longer useful, we should keep dragonegg in the compilers portgroup.

[seanfarley (Sean Farley)]

OpenBLAS (nor ATLAS) use the compilers portgroup. Plus, the mpi portgroup dependes on the compiler portgroup. I think it'd make more sense to remove them from the group and just add those variants to the ports that really need them (if any at all)

[dstrubbe (David Strubbe)]

Ok fair enough. In practice, the use of dragonegg-gfortran in OpenBLAS is not compatible with the compilers portgroup anyway since it is done along with clang rather than dragonegg-gcc.