Replication: MySQL Cluster: Added a new global system variable slave_transaction_retries: If the replication slave SQL thread fails to execute a transaction because of an InnoDB deadlock or exceeded InnoDB's innodb_lock_wait_timeout or NDBCLUSTER's TransactionDeadlockDetectionTimeout or TransactionInactiveTimeout, it automatically retries slave_transaction_retries times before stopping with an error. The default is 0, and you must explicitly set the value greater than 0 to enable the “retry” behavior. (Bug#8325)

MySQL Cluster: More informative error messages are provided when a query is issued against an NDB table that has been modified by another mysqld server. (Bug#6762)

Replication: For slave replication servers started with --replicate-* options, statements that should not be replicated according those options no longer are written to the slave's general query log. (Bug#8297)

The use of SESSION or GLOBAL is now disallowed for user variables. (Bug#9286)

mysqld_safe will create the directory where the UNIX socket file is to be located if the directory does not exist. This applies only to the last component of the directory path name. (Bug#8513)

ONLY_FULL_GROUP_BY no longer is included in the ANSI composite SQL mode. (Bug#8510)

Added sql_notes session variable to cause Note-level warnings not to be recorded. (Bug#6662)

Modified the parser to allow SELECT statements following the UNION keyword to be subqueries in parentheses. (Bug#2435)

InnoDB: Commit after every 10,000 copied rows when executing CREATE INDEX, DROP INDEX or OPTIMIZE TABLE, which are internally implemented as ALTER TABLE. This makes it much faster to recover from an aborted operation.

User variable coercibility has been changed from “coercible” to “implicit.” That is, user variables have the same coercibility as column values.

Added --with-big-tables compilation option to configure. (Previously it was necessary to pass -DBIG_TABLES to the compiler manually in order to enable large table support.) See Section 2.9.2, “Typical configure Options”, for details.

The coercibility for the return value of functions such as USER() or VERSION() now is “system constant” rather than “implicit.” This makes these functions more coercible than column values so that comparisons of the two do not result in Illegal mix of collations errors. COERCIBILITY() was modified to accommodate this new coercibility value. See Section 11.10.3, “Information Functions”.

Added configuration directives !include and !includedir for including option files and searching directories for option files. See Section 4.2.3.3, “Using Option Files”, for usage.

NULL now is considered more coercible than string constants. This resolves some Illegal mix of collations conflicts.

Replication: If multiple semicolon-separated statements were received in a single packet, they were written to the binary log as a single event rather than as separate per-statement events. For a server serving as a replication master, this caused replication to fail when the event was sent to slave servers. (Bug#8436)

Replication: A replication master stamped a generated statement (such as a SET statement) with an error code intended only for another statement. This could happen, for example, when a statement generated a duplicate key error on the master but still had be to replicated to the slave. (Bug#8412)

Replication: If the slave was running with --replicate-*-table options which excluded one temporary table and included another, and the two tables were used in a single DROP TEMPORARY TABLE IF EXISTS statement, as the ones the master automatically writes to its binary log upon client's disconnection when client has not explicitly dropped these, the slave could forget to delete the included replicated temporary table. Only the slave needs to be upgraded. (Bug#8055)

Replication: Treat user variables as having IMPLICIT derivation (coercibility) to avoid “Illegal mix of collations” errors when replicating user variables. (Bug#6676)

For MyISAM tables, REPAIR TABLE no longer discard rows that have incorrect checksum. (Bug#9824)

Fixed option-parsing code for the embedded server to understand K, M, and G suffixes for the net_buffer_length and max_allowed_packet options. (Bug#9472)

Some user variables were not being handled with “implicit” coercibility. (Bug#9425)

Using TIMESTAMP columns with no minute or second parts in GROUP BY clauses with the new system variable set to 1 caused the server to crash. (Bug#9401)

Allow extra HKSCS and cp950 characters (big5 extension characters) to be accepted in big5 columns. (Bug#9357)

Subqueries using ALLor ANY that contained a HAVING clause did not work correctly. (Bug#9350)

BLOB(M) and TEXT(M) columns, with M less than 256, were being created as BLOB and TEXT columns rather than TINYBLOB or TINYTEXT columns. (Bug#9303)

InnoDB: If MySQL wrote to its binlog, but for some reason trx->update_undo and trx->insert_undo were NULL in InnoDB, then trx->commit_lsn was garbage, and InnoDB could assert in the log flush of trx_commit_complete_for_mysql(). (Bug#9277)

The utf8_spanish2_ci and ucs2_spanish2_ci collations no longer consider r equal to rr . If you upgrade to this version from an earlier version, you should rebuild the indexes of any affected tables. (Bug#9269)

Depending on index statistics, GROUP BY col1, col2, ... could return incorrect results if the first table processed for a join had several indexes that cover the grouped columns. (Bug#9213)

OPTIMIZE TABLE was written twice to the binary log when used on InnoDB tables. (Bug#9149)

The CHARSET(), COLLATION(), and COERCIBILITY() functions sometimes returned NULL. CHARSET() and COLLATION() returned NULL when given any of these arguments that evaluated to NULL: A system function such as DATABASE(); a column value; and a user variable. Now CHARSET() and COLLATION() return the system character set and collation; the column character set and collation; and binary. COERCIBILITY(NULL) now returns “ignorable” coercibility rather than NULL. (Bug#9129)

mysqldump misinterpreted “_” and “%” characters in the names of tables to be dumped as wildcard characters. (Bug#9123)

Bundled zlib in the source distribution was upgraded to 1.2.2. (Bug#9118)

Setting the max_error_count system variable to 0 resulted in a setting of 1. (Bug#9072)

The use of XOR together with NOT ISNULL() erroneously resulted in some outer joins being converted to inner joins by the optimizer. (Bug#9017)

MIN(col_name) and MAX(col_name) could fail to produce the correct result if col_name was contained in multiple indexes and the optimizer did not choose the first index that contained the column. (Bug#8893)

Using a compariosn where the left expression of IN, ALL, or ANY was a subquery caused the server to crash (Bug#8888)

When using the cp1250_czech_cs collation, empty literal strings were not regarded as equal to empty character columns. (Bug#8840)

Binary data stored in BLOB or BINARY columns would be erroneously dumped if mysqldump was invoked with --hex-blob and --skip-extended-insert arguments. This happened if data contained characters larger then 0x7F . (Bug#8830)

Using NOW() in a subquery caused the server to crash. (Bug#8824)

In prepared statements, subqueries containing parameters were erroneously treated as const tables during preparation, resulting in a server crash. (Bug#8807)

Killing a filesort could cause an assertion failure. (Bug#8799)

Expressions involving nested CONCAT() calls and character set conversion of string constants could return an incorrect result. (Bug#8785)

Do not try to space-pad BLOB columns containing ucs2 characters. (Bug#8771)

This regression was introduced by Bug#7350.

If max_join_size was set, a query containing a subquery that exceeded the examined-rows limit could hang. (Bug#8726)

Table creation for a MyISAM table failed if DATA DIRECTORY or INDEX DIRECTORY options were given that specified the path name to the database directory where the table files would be created by default. (Bug#8707)

A rare race condition could cause FLUSH TABLES WITH READ LOCK to hang. (Bug#8682)

InnoDB: If one used LOCK TABLES, created an InnoDB temp table, and did a multiple-table update where a MyISAM table was the update table and the temp table was a read table, then InnoDB asserted in row0sel.c because n_mysql_tables_in_use was 0. Also, we remove the assertion altogether and just print an error to the .err log if this important consistency check fails. (Bug#8677)

AES_DECRYPT(col_name,key) could fail to return NULL for invalid values in col_name, if col_name was declared as NOT NULL. (Bug#8669)

Use of GROUP_CONCAT(x) in a subquery, where x was an alias to a column in the outer query, resulted in a server crash. (Bug#8656)

Too many rows were returned from queries that combined ROLLUP and LIMIT if SQL_CALC_FOUND_ROWS was given. (Bug#8617)

Incorrect results were returned from queries that combined SELECT DISTINCT, GROUP BY , and ROLLUP. (Bug#8616)

Queries that combined SELECT DISTINCT, SUM(), and ROLLUP could cause the MySQL server to crash. (Bug#8615)

A join on two tables failed when each contained a BIGINT UNSIGNED column that were compared when their values exceeded 2⁶³ – 1. The match failed and the join returned no rows. (Bug#8562)

LIKE pattern-matching for strings did not work correctly with the cp1251_bin collation. (Bug#8560)

MATCH ... AGAINST in natural language mode could cause a server crash if the FULLTEXT index was not used in a join (that is, EXPLAIN did not show fulltext join mode) and the search query matched no rows in the table. (Bug#8522)

perror was printing a spurious extra line of output ("Error code ###: Unknown error ###" printed directly before the correct line with the error message). (Bug#8517)

The tee command could sometimes cause the mysql client to crash. (Bug#8499)

The MEMORY storage engine did not properly increment an AUTO_INCREMENT column if there was a second composite index that included the column. (Bug#8489)

REPAIR TABLE did not invalidate query results in the query cache that were generated from the table. (Bug#8480)

For a statement string that contained multiple slow queries, only the last one would be written to the slow query log. (Bug#8475)

When the server was started with --skip-name-resolve, specifying host name values that included netmasks in GRANT statements did not work. (Bug#8471)

The Cyrillic letters I (И) and SHORT I (Й) were treated as being the same character by the utf8_general_ci collation. (Bug#8385)

With a database was dropped with lower_case_table_names = 2, tables in the database also were dropped but not being flushed properly from the table cache. If the database was re-created, the tables also would appear to have been re-created. (Bug#8355)

When performing boolean full-text searches on utf8 columns, a double-quote character in the search string caused the server to crash. (Bug#8351)

The MAX_CONNECTIONS_PER_HOUR resource limit was not being reset hourly and thus imposed an absolute limit on number of connections per account until the server is restarted or the limits flushed. (Bug#8350)

The --set-character-set option for myisamchk was changed to --set-collation. The value needed for specifying how to sort indexes is a collation name, not a character set name. (Bug#8349)

Corruption of MyISAM table indexes could occur with TRUNCATE TABLE if the table had already been opened. For example, this was possible if the table had been opened implicitly by selecting from a MERGE table that mapped to the MyISAM table. The server now issues an error message for TRUNCATE TABLE under these conditions. (Bug#8306)

In string literals with an escape character (“\”) followed by a multi-byte character that has a second byte of “\”, the literal was not interpreted correctly. The next character now is escaped, not just the next byte. (Bug#8303)

On Windows, create shared memory objects with the proper access rights to make them usable when the client and server are running under different accounts. (Bug#8226)

For a query with both GROUP BY and COUNT(DISTINCT) clauses and a FROM clause with a subquery, NULL was returned for any VARCHAR column selected by the subquery. (Bug#8218)

With lower_case_table_names set to 1, mysqldump on Windows could write the same table name in different lettercase for different SQL statements. (Bug#8216)

Matching of table names by mysqlhotcopy now accommodates DBD::mysql versions 2.9003 and up, which implement identifier quoting. (Bug#8136)

Creating a table using a name containing a character that is illegal in character_set_client resulted in the character being stripped from the name and no error. The character now is considered an error. (Bug#8041)

mysqldump now avoids writing SET NAMES to the dump output if the server is older than version 4.1 and would not understand that statement. (Bug#7997)

Mixed-case database and table names in the grant tables were ignored for authentication if the lower_case_table_names system variable was set. GRANT will not create such privileges when lower_case_table_names is set, but it is possible to create them by direct manipulation of the grant tables, or that old grant records were present before setting the variable. (Bug#7989)

Accented letters were improperly treated as distinct by the utf_general_ci collation. (Bug#7878)

Conversion of strings to doubles is now more accurate for floating point values that can be represented by integers, such as 123.45E+02. (Bug#7840)

ENUM and SET columns in privilege tables incorrectly had a case-sensitive collation, resulting in failure of assignments of values that did not have the same lettercase as given in the column definitions. The collation was changed to be case insensitive. (Bug#7617)

The output of the STATUS (\s) command in mysql had the values for the server and client character sets reversed. (Bug#7571)

A problem in index cost calculation caused a USE INDEX or FORCE INDEX hint not to be used properly for a LEFT JOIN across indexed BLOB columns. (Bug#7520)

Ordering by an unsigned expression (more complex than a column reference) was treating the value as signed, producing incorrectly sorted results. HAVING was also treating unsigned columns as signed. (Bug#7425)

Changed mysql_server_end() C API function to restore more variables to their initial state so that a subsequent call to mysql_server_init() would not cause a client program crash. (Bug#7344)

Made the relay_log_space_limit system variable show up in the output of SHOW VARIABLES. (Bug#7100)

An expression that tested a case-insensitive character column against string constants that differed in lettercase could fail because the constants were treated as having a binary collation. (For example, WHERE city='London' AND city='london' could fail.) (Bug#7098, Bug#8690)

When setting integer system variables to a negative value with SET VARIABLES, the value was treated as a positive value modulo 2³². (Bug#6958)

If a MyISAM table on Windows had INDEX DIRECTORY or DATA DIRECTORY table options, mysqldump dumped the directory path names with single-backslash path name separators. This would cause syntax errors when importing the dump file. mysqldump now changes “\” to “/” in the path names on Windows. (Bug#6660)

Incorrectly ordered results were returned from a query using a FULLTEXT index to retrieve rows and there was another index that was usable for ORDER BY. For such a query, EXPLAIN showed the fulltext join type, but showed the other (not FULLTEXT) index in the Key column. (Bug#6635)

Neither SHOW ERRORS nor SHOW WARNINGS were displaying Error-level messages. (Bug#6572)

Retrieving from a view defined as a SELECT that mixed UNION ALL and UNION DISTINCT resulted in a different result than retrieving from the original SELECT. (Bug#6565)

A problem with static variables did not allow building the server on Fedora Core 3. (Bug#6554)

Unions between binary and nonbinary columns failed due to a collation coercibility problem. (Bug#6519)

The CHAR() function was not ignoring NULL arguments, contrary to the documentation. (Bug#6317)

FOUND_ROWS() returned an incorrect value for preceding SELECT statements that used no table or view. (Bug#6089)

InnoDB: Honor the --tmpdir startup option when creating temporary files. Previously, InnoDB temporary files were always created in the temporary directory of the operating system. On Netware, InnoDB will continue to ignore --tmpdir. (Bug#5822)

The bundled readline library caused a segmentation fault in mysql when the user entered Shift+Enter. (Bug#5672)

The data type for MAX(datetime_col) was returned as VARCHAR rather than DATETIME if the query included a GROUP BY clause. (Bug#5615)

Worked around a bug in support for NSS support in glibc when static linking is used and LDAP is one of the NSS sources. The workaround is to detect when the bug causes a segmentation fault and issue a diagnostic message with information about the problem. (Bug#4872, Bug#3037)

If the mysql prompt was configured to display the default database name, and that database was dropped, mysql did not update the prompt. (Bug#4802)

Platform and architecture information in version information produced for --version option on Windows was always Win95/Win98 (i32). More accurately determine platform as Win32 or Win64 for 32-bit or 64-bit Windows, and architecture as ia32 for x86, ia64 for Itanium, and axp for Alpha. (Bug#4445)

Host name matching didn't work if a netmask was specified for table-specific privileges. (Bug#3309)

InnoDB: An error in mysqld caused InnoDB in MySQL 4.1.8 to 4.1.10 InnoDB to refuse to use a table created with MySQL 3.23.49 or earlier if it was in the new compact InnoDB table format of 5.0.3 or later.

Workaround.  Upgrade to 4.1.11 or newer, or dump the table and re-create it with MySQL 3.23.50 or newer before upgrading.

InnoDB: A table with a primary key that contained at least two column prefixes was prone to memory corruption. An example of an affected CREATE TABLE statement is shown here:

CREATE TABLE t  (
    a CHAR(100),
    b TINYBLOB,
    PRIMARY KEY(a(5), b(10))
);

InnoDB: If InnoDB cannot allocate memory, keep retrying for 60 seconds before we intentionally crash mysqld; maybe the memory shortage is just temporary.

my_print_defaults ignored the --defaults-extra-file and --defaults-file options.

Privileges could be escalated via database wildcards in GRANT statements. (CVE-2004-0957)

InnoDB: Work around a problem in AIX 5.1 patched with ML7 security patch: InnoDB would refuse to open its ibdata files, complaining about an operating system error 0.