Scientificpython needs to be compiled with mpi wrapper if sci-libs/netcdf[mpi], otherwise it fails to build.
Created attachment 218116 [details] build log
Created attachment 218118 [details] emerge --info
Created attachment 218171 [details, diff] adding patch to ebuild
Created attachment 218173 [details, diff] adding mpicc flags for netcdf module
Created attachment 218177 [details, diff] adding mpicc flags for netcdf module (better way)
Now fixed in cvs, thanks.
Um, this fix actually broke my world update. I had to revert the netcdf patch in my local overlay to make it build successfully. Note that I do not have openmpi installed, rather I have a perfectly working mpich2 install. This seems to lend some weight to my argument against patching other ebuilds to try to work around a broken openmpi library. This current approach is bad juju...
netcdf lost its mpi USE. I removed the USEdep. If you want a different solution, please correct it.
Sorry about the messy fix previously. Anyway, reclosing.