D.3.4 Changes in release 4.0.12 (15 Mar 2003: Production)

Functionality added or changed:

• mysqld no longer reads options from world-writeable config files.
• Integer values between 9223372036854775807 and 9999999999999999999 are now regarded as unsigned longlongs, not as floats. This makes these values work similar to values between 10000000000000000000 and 18446744073709551615.
• SHOW PROCESSLIST will now include the client TCP port after the hostname to make it easier to know from which client the request originated.

Bugs fixed:

• Fixed mysqld crash on extremely small values of sort_buffer variable.
• INSERT INTO u SELECT ... FROM t was written too late to the binary log if t was very frequently updated during the execution of this query. This could cause a problem with mysqlbinlog or replication. The master must be upgraded, not the slave. (Bug #136)
• Fixed checking of random part of WHERE clause. (Bug #142)
• Fixed a bug with multiple-table updates with InnoDB tables. This bug occurred as, in many cases, InnoDB tables can not be updated ``on the fly,'' but offsets to the records have to be stored in a temporary table.
• Added missing file mysql_secure_installation to the server RPM subpackage. (Bug #141)
• Fixed MySQL (and myisamchk) crash on artificially corrupted .MYI files.
• Don't allow BACKUP TABLE to overwrite existing files.
• Fixed a bug with multi-table UPDATE statements when user had all privileges on the database where tables are located and there were any entries in tables_priv table, that is, grant_option was true.
• Fixed a bug that allowed a user with table or column grants on some table, TRUNCATE any table in the same database.
• Fixed deadlock when doing LOCK TABLE followed by DROP TABLE in the same thread. In this case one could still kill the thread with KILL.
• LOAD DATA LOCAL INFILE was not properly written to the binary log (hence not properly replicated). (Bug #82)
• RAND() entries were not read correctly by mysqlbinlog from the binary log which caused problems when restoring a table that was inserted with RAND(). INSERT INTO t1 VALUES(RAND()). In replication this worked ok.
• SET SQL_LOG_BIN=0 was ignored for INSERT DELAYED queries. (Bug #104)
• SHOW SLAVE STATUS reported too old positions (columns Relay_Master_Log_File and Exec_master_log_pos) for the last executed statement from the master, if this statement was the COMMIT of a transaction. The master must be upgraded for that, not the slave. (Bug #52)
• LOAD DATA INFILE was not replicated by the slave if replicate_*_table was set on the slave. (Bug #86)
• After RESET SLAVE, the coordinates displayed by SHOW SLAVE STATUS looked un-reset (though they were, but only internally). (Bug #70)
• Fixed query cache invalidation on LOAD DATA.
• Fixed memory leak on ANALYZE procedure with error.
• Fixed a bug in handling CHAR(0) columns that could cause wrong results from the query.
• Fixed rare bug with wrong initialisation of AUTO_INCREMENT column, as a secondary column in a multi-column key (see section 3.5.9 Using AUTO_INCREMENT), when data was inserted with INSERT ... SELECT or LOAD DATA into an empty table.
• On Windows, STOP SLAVE didn't stop the slave until the slave got one new command from the master (this bug has been fixed for MySQL 4.0.11 by releasing updated 4.0.11a Windows packages, which include this individual fix on top of the 4.0.11 sources). (Bug #69)
• Fixed a crash when no database was selected and LOAD DATA command was issued with full table name specified, including database prefix.
• Fixed a crash when shutting down replication on some platforms (for example, Mac OS X).
• Fixed a portability bug with pthread_attr_getstacksize on HP-UX 10.20 (Patch was also included in 4.0.11a sources).
• Fixed the bigint test to not fail on some platforms (for example, HP-UX and Tru64) due to different return values of the atof() function.
• Fixed the rpl_rotate_logs test to not fail on certain platforms (e.g. Mac OS X) due to a too long file name (changed slave-master-info.opt to .slave-mi).

User Comments

Add your own comment.