Greg,
I have located the source of the bus error and core dump on SGI
multiprocessor machines. Randall mentioned it awhile back (see
http://opendx.npaci.edu/mail/opendx-users/2002.08/msg00098.html ), and I
have been seeing it while testing 4.3.
What happens is exCleanup() in dxmain.c is executed twice, once for the
parent (_dxd_exMyPID = -1) and once for the master child (_dxd_exMyPID =
0). If the child gets there first, then the bus error occurs. If the parent
gets there first then a clean exit occurs.
The condition around line 1557 in dxmain.c can be modified to eliminate
this error by choosing either the parent or master child do the cleanup:
ok = ((exParent && _dxd_exMyPID==-1) || (nprocs == 1 && _dxd_exMyPID == 0
&& ! processor_status_on));
or
ok = ((exParent && !_dxd_exMyPID) || (nprocs == 1 && _dxd_exMyPID == 0 && !
processor_status_on));
Since you have worked on the multiprocessor implementation, wanted to check
with you on the correct approach. Another method that works is to sleep(1)
if _dxd_exMyPID == 0.
Jeff
