[MDEV-4004] Empty variable names allowed and break replication Created: 2013-01-06  Updated: 2013-05-08  Resolved: 2013-05-08

Status: Closed
Project: MariaDB Server
Component/s: None
Affects Version/s: 5.5.29
Fix Version/s: 5.5.31

Type: Bug Priority: Major
Reporter: Federico Razzoli Assignee: Sergei Golubchik
Resolution: Fixed Votes: 0
Labels: upstream

Issue Links:
Blocks
is blocked by MDEV-4333 5.5.31 merge Closed

 Description   

The original description can be found further in this field

The problem is reproducible with MySQL 5.1-5.6 and filed as http://bugs.mysql.com/bug.php?id=68135.

A variable can be created with an empty name. Apparently, it's confusing (see the original request below) and potentially error-prone; but it also causes problems with binary logging and replication:

SET @`` = 1;
 
CREATE TABLE t1 AS SELECT @`` AS f;
 
SHOW BINLOG EVENTS;
 
ERROR HY000: Error when executing command SHOW BINLOG EVENTS: Wrong offset or I/O error

Consequently, replication fails with ER_SLAVE_RELAY_LOG_READ_FAILURE.

Currently it's only reproducible on maria/5.5, but not on maria/5.1, 5.2, 5.3 or 10.0-base. However, I suppose it's just a matter of time, since it came to maria/5.5 with the last mysql-5.1.67 merge, and it exists in all current trees of MySQL.

Test case:

--source include/master-slave.inc
 
--source include/have_binlog_format_statement.inc
 
 
 
SET @`` = 1;
 
CREATE TABLE t1 AS SELECT @`` AS f;
 
 
 
--sync_slave_with_master

==================================

Original description

I'm not sure if this is a bug or a feature request. Empty names seem to be allowed for variables. No warnings.
Empty names are not allowed for schemas, tables, columns, views, etc.

 
 
MariaDB [test]> SET @`` = 1;
 
Query OK, 0 rows affected (0.00 sec)
 
 
 
MariaDB [test]> SELECT @``;
 
+------+
 
| @``  |
 
+------+
 
|    1 |
 
+------+
 
1 row in set (0.07 sec)
 
 
 
MariaDB [test]>
 
 

The same happens with local vars in stored routines:

 
 
DROP FUNCTION IF EXISTS `test`;
 
CREATE FUNCTION `test`()
 
	RETURNS BOOLEAN
 
BEGIN
 
	DECLARE `` TINYINT UNSIGNED;
 
	SET `` = 1;
 
	RETURN ``;
 
END;
 
 



 Comments   
Comment by Elena Stepanova [ 2013-01-22 ]

Hi Federico,

I suppose as a feature request it wouldn't stand much of a chance, not because it's unreasonable, but because it would almost inevitably cause compatibility issues – replication from an old slave to a new master, stored procedures which use an empty name, etc., and of course for MariaDb also compatibility with the upstream.

However, it turns out there is a bug with empty names, most likely a regression, which causes replication problems. I've updated the description and refiled it to MySQL bug base, and mentioned your request there, too. I doubt, however, that the bug will make Oracle forbid empty names, on the same reason of breaking compatibility with older versions; and if they don't, we can't either.

Thanks for the report, it revealed a serious issue.

Comment by Federico Razzoli [ 2013-01-22 ]

Hi Elena
I agree that it is serious, for this reason: I can hardly beilive that someone wants to use an empty name. I did it, but my SQL was generated by a buggy PHP script. In such cases, a warning would make debugging easier.
However, I admit that in some weird cases buggy code can provide the expected output, so I understand that breaking compatibility is a bad thing.

Comment by Elena Stepanova [ 2013-01-31 ]

A comment from the corresponding MySQL bug report:

"
Fixed in 5.1+. Documented in the 5.1.69, 5.5.31, 5.6.11, and 5.7.1 changelogs as follows:

A zero-length name for a user variable was incorrectly considered to be a sign of data corruption and thus was not written to the binary log.
"

So I suppose they haven't made an empty name illegal.

Comment by Sergei Golubchik [ 2013-05-08 ]

fixed by the mysql-5.5.31 merge

Generated at Thu Feb 08 06:52:58 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.