I use openmpi, gfortran, netcdf (and for some old code hdf5) for testing models before using the code on a larger mainframe. I later use fortran, c, cpp, netcdf and hdf5 for analysis, including 3-D graphics output. However when I look at all the options available with opensuse there seem to be a complex mix of options. I have done reasonably well in choosing a mixture of openmpi2, hdf5 and netcdf packages - ones with names like “hdf5_a_b_c-gnu-openmpi2-…” which seem to share the same regions of /usr space for the lib64 and include directories. However hdf5-fortran, netcdf-fortran and netcdf-c++ seem to be unrelated with their own directory structures which then need to be added to the makefiles and LD_LIBRARY_LOAD. I also find that their values of a_b_c can be different.
So although the system usually compiles and runs OK, in cases where the running program fails with lots of mixed netcdf/hdf5 backtrace lines, I am unsure whether this is an incompatibility in the library versions or some other bug of mine.
So has anyone produced a web page which explains all the openmpi/hdf5/netcdf/fortran/c++ options and how they are related - or if not is there any documentation anywhere (or a telephone number I can ring) which will help me to make sense of the present system? I could if necessary compile everything from source but it would seem to be a great waste of time.
Thanks …
David Webb.