On Wed, Feb 25, 2015 at 10:17 AM, Paul Hargrove <phhargr...@lbl.gov> wrote:
> I did that and just shipped a tarball to get Hargroved.
>>
>
> Tests have been dispatched... I will report complete results later today.
> The first of the BSD results should be in soon, and I'll plan to report
> go/nogo.
>
"NOGO"
I don't have full results (and no backtraces, yet), however things actually
look *worse* than the previous tarball.
While the previous tarball gave a pretty message, the current one SEGVs on
{Free,Net}BSD.
On FreeBSD-{8,9,10}/amd64 I don't get past "make check":
Segmentation fault (core dumped)
FAIL: dlopen_test
Oddly, my FreeBSD-10/i386 tester passed my tests (I don't test
FreeBSD-{8,9} on i386).
On NetBSD-6/amd64 I also fail "make check", but w/o the "Segmentation
fault" message (difference might not be OMPI-related):
FAIL: dlopen_test
On NetBSD-6/i386 I don't see the problem during "make check" but get a
terse failure from mpirun:
$ mpirun -mca btl sm,self -np 2 examples/ring_c'
[netbsd6-i386:05263] *** Process received signal ***
[netbsd6-i386:05263] Signal: Segmentation fault (11)
[netbsd6-i386:05263] Signal code: Address not mapped (1)
[netbsd6-i386:05263] Failing at address: 0xbb800000
Unable to print stack trace!
[netbsd6-i386:05263] *** End of error message ***
On OpenBSD-5 everything appears to "just work" now on both amd64 and i386.
When I have a chance (might not be until Thu) I will collect backtraces
from the SEGVs on FreeBSD and NetBSD, as well as sanity checking the
related configure output.
Jeff,
In the meantime you may want to visit
https://www.freebsd.org/cgi/man.cgi?query=dlopen&sektion=3&manpath=FreeBSD+10.1-RELEASE
https://www.freebsd.org/cgi/man.cgi?query=dlopen&sektion=3&manpath=NetBSD+6.1.5
and see if the OMPI usage of dlopen() is consistent with the docs for
FreeBSD and NetBSD.
-Paul
--
Paul H. Hargrove phhargr...@lbl.gov
Computer Languages & Systems Software (CLaSS) Group
Computer Science Department Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory Fax: +1-510-486-6900
