8935e0a5e0
There is more comprehensive work regarding MPI_SIZEOF coming, but the Fortran working group in the MPI Forum is debating this internally, and I'm still doing more testing to get a final solution. So for the moment, just remove real*16 and complex*32 support so that it compiles porperly with older compilers (that do not support real*16 and complex*32). This commit was SVN r32048.