[MDEV-6202] MariaDB 10.0.10 doesn't work with KDE's Akonadi Created: 2014-05-02  Updated: 2014-06-22  Due: 2014-06-10  Resolved: 2014-05-10

Status: Closed
Project: MariaDB Server
Component/s: None
Affects Version/s: 10.0.10
Fix Version/s: None

Type: Bug Priority: Major
Reporter: Bartłomiej Piotrowski (Inactive) Assignee: Unassigned
Resolution: Fixed Votes: 0
Labels: None
Environment:

Arch Linux, gcc 4.9, Akonadi 1.12.1

Attachments: Text File mysql-manual-create.err.txt     Text File mysql.err.txt    
Issue Links:
Relates
relates to MDEV-6340 Mariadb 10.0.12 fatal "Lost connectio... Closed

 Description   

Originally reported here:
https://bugs.archlinux.org/task/40191
https://bugs.archlinux.org/task/40186

After I've put MariaDB 10.0 in our [testing] repository, KDE users are reporting issues with Akonadi; mysqld segfaults whether the database is old (5.5.x) or freshly created. See attachments for more detailed logs.

I'm not really sure what's the culprit,, I'm not excluding packaging issue…

The problem has been also reported on KDE bugtracker: https://bugs.kde.org/show_bug.cgi?id=334205

 Comments   
Comment by Bartłomiej Piotrowski (Inactive) [ 2014-05-02 ]

Seems that gcc 4.9 is the problem… When Maria is built using 4.8, everything works as expected.

Comment by Elena Stepanova [ 2014-05-03 ]

I tried to build with gcc 4.9 (on Sid), didn't get a crash, so there should be something else in addition to the compiler version.
I also tried an upgrade from 5.5, no crash either.
But it still looks suspicious that, according to your log, mysql_upgrade was not run after upgrade. Did you try to run it manually and see if the problem goes away?

Comment by Bartłomiej Piotrowski (Inactive) [ 2014-05-05 ]

May it be glibc or binutils? I'll try to rebuild everything with debug flags and see if anything more interesting in backtrace shows up. Changing the priority as it's not blocker, thanks Elena.

Comment by Sergei Golubchik [ 2014-05-10 ]

Discussion in the KDE bug tracker has a suggestion that this might be related to a gcc-4.9 bug. This bug causes libQtScript to crash (when compiled with gcc 4.9) and there's a further link to the gcc bug report and a patch.

Anyway, the crash happens only when Akonadi uses Qt, compiled with gcc-4.8, and libmariadbd, compiled with gcc-4.9. May be there's something unsafe in this configuration.

Comment by Bartłomiej Piotrowski (Inactive) [ 2014-05-10 ]

We tried rebuilding Akonadi and Qt with the new gcc, but it had no effect.

However our toolchain maintainer recently uploaded gcc built from new snapshot dated 20140507 with an additional patch[1]. Since then everything builds and works as expected.

[1] https://projects.archlinux.org/svntogit/packages.git/tree/trunk/gcc-4.9-tree-ssa-threadedge.patch?h=packages/gcc&id=518aead30e1ac6a0430893de63bdbb960317ba76

Comment by Sergei Golubchik [ 2014-05-10 ]

Okay, thanks.

For the reference, this is the corresponding (to this patch) bug report in the gcc bug tracker: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60902

I'm going to close this issue then.

Comment by Bartłomiej Piotrowski (Inactive) [ 2014-05-25 ]

Apparently the new snapshot hasn't fixed everything. I have added -fno-delete-null-pointer-checks to CFLAGS and CXXFLAGS and now MariaDB behaves as it should.

Generated at Thu Feb 08 07:10:07 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.