On 7/26/06 10:51 AM, "Borenstein, Bernard S" <bernard.s.borenst...@boeing.com> wrote:
> ld: Warning: size of symbol `mpi_fortran_argv_null_' changed from 1 in > chimera/discmo.o to 16 in > /home/bsb3227/openmpi_1.1.1b3/lib/libmpi.a(test_constants_f.o) > ld: Warning: size of symbol `mpi_fortran_argvs_null_' changed from 4 in > chimera/discmo.o to 16 in > /home/bsb3227/openmpi_1.1.1b3/lib/libmpi.a(test_constants_f.o) > ld: Warning: size of symbol `mpi_fortran_status_ignore_' changed from 20 > in chimera/discmo.o to 16 in > /home/bsb3227/openmpi_1.1.1b3/lib/libmpi.a(test_constants_f.o) > ld: Warning: size of symbol `mpi_fortran_errcodes_ignore_' changed from > 4 in chimera/discmo.o to 16 in > /home/bsb3227/openmpi_1.1.1b3/lib/libmpi.a(test_constants_f.o) Rats. It took me several days of hunting, but I finally figured this one out. It seems like this only showed up in static builds of some compilers (intel and gfortran, in my testing). This is due to something we introduced in 1.1b3 as a fix for OSX. It's a long, horrid story of Fortran, linkers, and bears. I committed a fix to the trunk at r11057, and included a lengthy log message explaining what I did to fix the problem (https://svn.open-mpi.org/trac/ompi/changeset/11057). It works on all the configurations that I could try -- could you verify that it works for you? Here's what I tested (static and dynamic builds for all): - Linux - AMD, gfortran 4.0 - AMD, intel 9.0 - AMD, pgi 6.1 - AMD, pathscale 2.3 - OSX - x86, gfortran 4.2 -- Jeff Squyres Server Virtualization Business Unit Cisco Systems
