Details
Major
Description
Originally spotted with mysql-5.6.24 (http://bugs.mysql.com/bug.php?id=76654), but mariadb 10.0.17 seems to have the same issue:
Description:
When building mysql-5.6.24 with gcc 5.0 on Fedora 22 or later some tests fail on big integers operations, for example bigint test case.
Full build log is available here:
https://kojipkgs.fedoraproject.org//work/tasks/3420/9443420/build.log
How to repeat:
Run bigint test or a small reproducer:
MySQL [(none)]> select -(-9223372036854775808); +-------------------------+ | -(-9223372036854775808) | +-------------------------+ | 854775808 | +-------------------------+
Which is obviously wrong.
Gliffy Diagrams
Attachments
Issue Links
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
mysql's patch without whitespace changes.
I don't like this fix, it's a hack that converts the loop to an equivalent loop that magically works in gcc 5. It does not fix the root of the bug (undefined behavior) but simply changes the following code to work with how gcc 5.0 happens to behave in this undefined case. If gcc 6 will start doing something different here, it will break again.
I will directly fix the undefined behavior instead.
See the mysql bug report, which contains a patch (not tested with mariadb yet though).