Closed jcphill closed 10 years ago
Original date: 2014-07-06 16:18:53
I'm running into this issue again while trying to debug a segfault. Any ideas?
Original date: 2014-08-03 06:18:51
Ping. Progress on this? Everything else to be fixed for 6.6 has been merged.
Original date: 2014-08-03 12:55:24
Can't get much done over hotel wireless from Puerto Rico.
Original date: 2014-08-04 15:16:03
I just tried to reproduce this, and the exact build command shown runs to completion on Stampede. I also tried adding module load python
and reran the build, to see if there was a version skew error (2.6 vs 2.7). That build also ran successfully.
Jim: Does this still occur for you?
Original date: 2014-08-04 22:34:21
I just tested and I still get the error, both with and without the fftw module loaded:
Currently Loaded Modules: 1) TACC-paths 3) cluster-paths 5) mvapich2/1.9a2 7) TACC 2) Linux 4) intel/13.0.2.146 6) cluster 8) fftw3/3.3.2
Currently Loaded Modules: 1) TACC-paths 3) cluster-paths 5) mvapich2/1.9a2 7) TACC 2) Linux 4) intel/13.0.2.146 6) cluster
login2$ git status
On branch charm nothing to commit (working directory clean) login2$ git show commit 7ef40e7d4eca868513ea7d7fb65a5fcfc0d8b625 Author: Ronak Buch <rabuch2`illinois.edu> Date: Mon Aug 4 12:02:00 2014 -0500
Improve grammar in comments
Change-Id: I1c3e01e366253fc7647b542a2bd302870476ce3d
...
Original date: 2014-08-04 22:47:54
OK, I've reproduced it now, and I think I have a handle on its cause. I'll try to have a patch posted this evening.
Original date: 2014-08-04 23:08:52
Change 342 posted in Gerrit.
The reason I couldn't reproduce this earlier is that parallel compilation, which I assumed was inconsequential, was actually masking the problem. It generally meant that the module in question was built before the linker went looking for it. Really, that module shouldn't be linked into binaries meant to run on the build or run host at all.
Original issue: https://charm.cs.illinois.edu/redmine/issues/433