Bug #46419 Memory corruption when creating tables with different character sets in RQG
Submitted: 28 Jul 2009 1:51 Modified: 13 Oct 2009 8:19
Reporter: Hema Sridharan Email Updates:
Status: Not a Bug Impact on me:
None 
Category:MySQL Server Severity:S3 (Non-critical)
Version:mysql-6.0, azalea OS:Linux
Assigned to: Philip Stoev CPU Architecture:Any
Tags: regression

[28 Jul 2009 1:51] Hema Sridharan 
Description:
Create grammar file that creates database db1 and tables in db1.
The grammar file will also set different character sets and collations randomly while creating tables.

Execution of the grammar will lead to memory corruption. This is observed only in latest trees.

How to repeat:
Run the script with RQG load as follows,

perl runall.pl --basedir=/export/home/tmp/WL4732f/mysql-6.0-backup/
--grammar=conf/memory_corrupt.yy

# 23:50:06 Error: 'Duplicate column name ÿc8ÿ' indicates memory corruption.

The grammar file is attached below.
[28 Jul 2009 1:52] Hema Sridharan 
Grammar file for test execution

Attachment: memory_corrupt.yy (application/octet-stream, text), 1.51 KiB.
[28 Jul 2009 5:46] Sveta Smirnova 
Thank you for the report.

Verified as described. Version 5.1 is not affected.
[13 Oct 2009 8:19] Philip Stoev 
The ErrorMessageCorruption Validator in the Random Query Generator attempts to detect server memory corruption by checking for non-alphanumeric characters in the error messages that are returned by the server.

This Validator is enabled by default and has uncovered numerous issues, however unfortunately it is easily mislead by non-ascii character sets and data. So, it is best that all tests that work on character sets disable this Validator by providing an explicit list of validators using the --validators option.