ice, segfault, ...
Hi,
kullervo is still experiencing spurious segfaults and ICEs, and I still
don't see what causes them. I noticed this again when it was building
opencc, the package built fine but then failed during some tests. I built it
again (manually), and it failed other tests:
The following tests FAILED:
- 3 - zht2zhs_convert (Failed)
- 4 - zht2zhs_compare (Failed)
+ 7 - mix2zhs_convert (Failed)
+ 8 - mix2zhs_compare (Failed)
On the next manual build, everything went fine. I am convinced this causes
many of the failed builds (at least on kullervo). The longer a package needs
to build, the more probable it is to fail, and no, it is not an increase in
cosmic rays, I measure that, everything is calm at the moment.
Can we be sure that packages that build fine are actually ok, or may there
be hidden problems in them? After all, I do not test every package that
kullervo is building. Do we have any way to check this? As Ingo alreay
noted, sometimes the errors come in bunches, maybe during times of high load
or heavy swap use. Maybe there is some test program (memtest?) with which we
can reproduce these errors to finally track them down? Or is this just
kullervo getting old?
Christian
Reply to: