Regression: Crash when called from Rmpi
we have three machines at different hardy patchlevels,
one with openmpi1-1.2.5-1 (around 2008-03-17)
and two current with 1.2.5-1ubuntu1. All are AMD64 architecture.
When initializing Rmpi I get a
*** caught segfault ***
which actually happens in
Rmpi.c:73 MPI_Init((void *)0,(void *)0);
Program received signal SIGSEGV, Segmentation fault.
0x00007f7e9ffb5b8b in _int_malloc () from /usr/lib/
#0 0x00007f7e9ffb5b8b in _int_malloc () from /usr/lib/
#1 0x00007f7e9ffb6e58 in malloc () from /usr/lib/
#2 0x00007f7e9ff98bfb in opal_class_
#3 0x00007f7e9fface2b in opal_malloc_init () from /usr/lib/
#4 0x00007f7e9ff99d97 in opal_init_util () from /usr/lib/
#5 0x00007f7e9ff99e76 in opal_init () from /usr/lib/
#6 0x00007f7ea0889723 in ompi_mpi_init () from /usr/lib/
#7 0x00007f7ea08ab15f in PMPI_Init () from /usr/lib/
#8 0x00007f7ea0aef866 in mpi_initialize () at Rmpi.c:73
a simple "mpirun -n 4 date" works fine on all machines.
Although all three hachines have completely different libc-versions
(between 2.7-5ubuntu2 and 2.7-9ubuntu2) it is sufficient to copy the
to the two other machines. Rebuilding 1.2.5-1ubuntu1 on the
newer machines doesn't help.
Since the only changes in ubuntu-1 are fixed dangling pointers
and a maintainer field, I suspect it has to do with the build process.