raw `exit(1)` instead of `MPI_Abort`

[valentin-seitz]

I just ran into an issue where extrae reported the following in stderr:

Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.
Buffer_InsertMultiple: No room for 1 events.

And the job continued to run until the time limit was over hanging in the other processes. So apparently my OpenMPI did not really capture the dying processes?...

I just did a quick grep and found the following responsible code line:

extrae/src/tracer/wrappers/API/buffers.c

Lines 617 to 618 in ea04ec1

	fprintf (stderr, "Buffer_InsertMultiple: No room for %d events.\n", num_events);
	exit(1);

Maybe it would be maybe a better idea for the *mpitrace.so objects to call MPI_Abort here to at least make an "MPI-compliant" effort to abort the execution and not the just process that has a buffer allocation problem?