I've retested with DBI 1.617 and DBI 1.618. Our program does not segfault with either of them so it has to be something specific to the 1.620 changes.
Matt On Thu, May 10, 2012 at 10:00 PM, Matthew Musgrove <mr.musk...@gmail.com>wrote: > We have this threaded program that runs as a daemonized service and it's > been happy for quite a while using DBI 1.616 and DBD::Oracle 1.30. Last > week, I updated DBI to 1.620 and DBD::Oracle 1.44 and when I restarted the > service, it segfaulted as soon as it created a new thread. I was pretty > sure that it wasn't in DBD::Oracle because I had tested a prerelease of the > new code and didn't have any problems with it. I reverted it back to DBI > 1.616 and DBD::Oracle 1.30 and when I restarted the service, it was happy > again. Next I updated DBI to 1.620 and left DBD::Oracle alone. Again, it > segfaulted as soon as it created a new thread. It isn't my code so there > could be a bug or three lurking in there waiting to attack but as I said, > it's been working just fine. The only change was to DBI. I reverted DBI > back to 1.616 and decided to wait (you know, to see if anyone would report > anything). > > Today, I had a little free time and decided to try it again. It took some > coaxing but I finally got it create a core dump (after setting > DAEMON_COREFILE_LIMIT=" > unlimited"). I loaded the core dump in gdb and did a backtrace. It looks > like it segfaults in dbi_ima_dup. Unfortunately, this perl didn't have > debugging enabled. I compiled a new copy of perl (same version and I think > the same flags except with debugging), installed the same modules for it > and created another core dump. I logged 'backtrace', 'info registers' and > 'thread apply all backtrace' to separate files. This was my first time > using gdb so that was interesting. I don't really know what sort of > information you need so... yeah. > > I'm attaching output from perl -V (from both copies of perl) and 'info > registers'. I'll send a reply with the backtrace log next. > > What other information do I need to provide? Does anyone have any tips or > pointers on how I can debug this further? > > Matt >
