MySQL Community Server 5.0.33がリリースされました。


最も普及しているオープンソースデータベース管理システムであるMySQL Community Server 5.0.33がリリースされました。このリリースは、http://dev.mysql.com/downloads/とミラーサイトのダウンロード・ページから、ソースコードで現在利用可能です。


MySQL Community Server 5.0.33は、ソースコードのみのリリースです。これは、バイナリのビルドを今後提供しないという意味です。



以下のセクションは、MySQL Community Serverの最新のリリースされたバージョンであるMySQL Community Server 5.0.27リリース以降の、MySQLソースコードにおけるバージョン間の変更をリストアップしています。これは、以下のオンラインからも閲覧できます。


Functionality added or changed:
  * NDB Cluster: Setting the configuration parameter
    LockPagesInMainMemory had no effect.
  * NDB Cluster: It is now possible to create a unique hashed
    index on a column that is not defined as NOT NULL. Note
    that this change applies only to tables using the NDB
    storage engine.
    Unique indexes on columns in NDB tables do not store null
    values because they are mapped to primary keys in an
    internal index table (and primary keys cannot contain
    Normally, an additional ordered index is created when one
    creates unique indexes on NDB table columns; this can be
    used to search for NULL values. However, if USING HASH is
    specified when such an index is created, no ordered index
    is created.
    The reason for permitting unique hash indexes with null
    values is that, in some cases, the user wants to save
    space if a large number of records are pre-allocated but
    not fully initialized. This also assumes that the user
    will not try to search for null values. Since MySQL does
    not support indexes that are not allowed to be searched
    in some cases, the NDB storage engine uses a full table
    scan with pushed conditions for the referenced index
    columns to return the correct result.
    Note that a warning is returned if one creates a unique
    nullable hash index, since the query optimizer should be
    provided a hint not to use it with NULL values if this
    can be avoided.
  * DROP TRIGGER now supports an IF EXISTS clause.
  * The Com_create_user status variable was added (for
    counting CREATE USER statements).
  * The --memlock option relies on system calls that are
    unreliable on some operating systems. If a crash occurs,
    the server now checks whether --memlock was specified and
    if so issues some information about possible workarounds.
  * The bundled yaSSL library was upgraded to version 1.5.0.
  * If the user specified the server options
    --max-connections=N or --table-open-cache=M, a warning
    would be given in some cases that some values were
    recalculated, with the result that --table-open-cache
    could be assigned greater value.
    It should be noted that, in such cases, both the warning
    and the increase in the --table-open-cache value were
    completely harmless. Note also that it is not possible
    for the MySQL Server to predict or to control limitations
    on the maximum number of open files, since this is
    determined by the operating system.
    The recalculation code has now been fixed to ensure that
    the value of --table-open-cache is no longer increased
    automatically, and that a warning is now given only if
    some values had to be decreased due to operating system
  * NDB Cluster: The HELP command in the Cluster management
    client now provides command-specific help. For example,
    HELP RESTART in ndb_mgm provides detailed information
    about the START command.
  * NDB Cluster: Added the --bind-address option for ndbd.
    This allows a data node process to be bound to a specific
    network interface.
  * NDB Cluster: The Ndb_number_of_storage_nodes system
    variable was renamed to Ndb_number_of_data_nodes.
  * NDB Cluster: The ndb_config utility now accepts -c as a
    short form of the --ndb-connectstring option.
  * SHOW STATUS is no longer logged to the slow query log.
  * mysqldump --single-transaction now uses START TRANSACTION
    /*!40100 WITH CONSISTENT SNAPSHOT */ rather than BEGIN to
    start a transaction, so that a consistent snapshot will
    be used on those servers that support it.
  * mysql_upgrade now passes all the parameters specified on
    the command line to both mysqlcheck and mysql using the
    upgrade_defaults file.
  * For the CALL statement, stored procedures that take no
    arguments now can be invoked without parentheses. That
    is, CALL p() and CALL p are equivalent.

Bugs fixed:
  * NDB Cluster: Sudden disconnection of an SQL or data node
    could lead to shutdown of data nodes with the error
    failed ndbrequire.
  * NDB Cluster: ndb_config failed when trying to use 2
    management servers and node IDs.
  * NDB Cluster (Cluster APIs): Using BIT values with any of
    the comparison methods of the NdbScanFilter class caused
    the cluster's data nodes to fail.
  * NDB Cluster: The failure of a data node failure during a
    schema operation could lead to additional node failures.
  * NDB Cluster: A committed read could be attempted before a
    data node had time to connect, causing a timeout error.
  * NDB Cluster (Cluster APIs): Some MGM API function calls
    could yield incorrect return values in certain cases
    where the cluster was operating under a very high load,
    or experienced timeouts in inter-node communications.
  * NDB Cluster: A unique constraint violation was not
    ignored by an UPDATE IGNORE statement when the constraint
    violation occurred on a non-primary key.
  * The stack size for NetWare binaries was increased to
    128KB to prevent problems caused by insufficient stack
    size. (Bug#23504:http://bugs.mysql.com/23504)
  * Attempting to use a view containing DEFINER information
    for a non-existent user resulted in an error message that
    revealed the definer account. Now the definer is revealed
    only to superusers. Other users receive only an access
    denied message. (Bug#17254:http://bugs.mysql.com/17254)
  * mysql_upgrade failed if the --password (or -p) option was
    given. (Bug#24896:http://bugs.mysql.com/24896)
  * For a nonexistent table, DROP TEMPORARY TABLE failed with
    an incorrect error message if read_only was enabled.
  * The code for generating USE statements for binary logging
    of CREATE PROCEDURE statements resulted in confusing
    output from mysqlbinlog for DROP PROCEDURE statements.
  * The InnoDB mutex structure was simplified to reduce
    memory load. (Bug#24386:http://bugs.mysql.com/24386)
  * The REPEAT() function could return NULL when passed a
    column for the count argument.
  * Accuracy was improved for comparisons between DECIMAL
    columns and numbers represented as strings.
  * InnoDB crashed while performing XA recovery of prepared
    transactions. (Bug#21468:http://bugs.mysql.com/21468)
  * ROW_COUNT() did not work properly as an argument to a
    stored procedure.
  * The size of MEMORY tables and internal temporary tables
    was limited to 4GB on 64-bit Windows systems.
  * For queries that select from a view, the server was
    returning MYSQL_FIELD metadata inconsistently for view
    names and table names. For view columns, the server now
    returns the view name in the table field and, if the
    column selects from an underlying table, the table name
    in the org_table field.
  * It was possible to use DATETIME values whose year, month,
    and day parts were all zeroes but whose hour, minute, and
    second parts contained nonzero values, an example of such
    an illegal DATETIME being '0000-00-00 11:23:45'.
  * It was possible to set the backslash character ("\") as
    the delimiter character using DELIMITER, but not actually
    possible to use it as the delimiter.
  * The loose index scan optimization for GROUP BY with MIN
    or MAX was not applied within other queries, such as
    CREATE TABLE ... SELECT ..., INSERT ... SELECT ..., or in
    the FROM clauses of subqueries.
    with another ALTER TABLE option other than RENAME TO did
    nothing. In addition, if ALTER TABLE was used on a table
    having disabled keys, the keys of the resulting table
    were enabled. (Bug#24395:http://bugs.mysql.com/24395)
  * Queries using a column alias in an expression as part of
    an ORDER BY clause failed, an example of such a query
    being SELECT mycol + 1 AS mynum FROM mytable ORDER BY 30
    - mynum. (Bug#22457:http://bugs.mysql.com/22457)
  * Trailing spaces were not removed from Unicode CHAR column
    values when used in indexes. This resulted in excessive
    usage of storage space, and could affect the results of
    some ORDER BY queries that made use of such indexes.
    Note: When upgrading, it is necessary to re-create any
    existing indexes on Unicode CHAR columns in order to take
    advantage of the fix. This can be done by using a REPAIR
    TABLE statement on each affected table.
  * Warnings were generated when explicitly casting a
    character to a number (for example, CAST('x' AS SIGNED)),
    but not for implicit conversions in simple arithmetic
    operations (such as 'x' + 0). Now warnings are generated
    in all cases. (Bug#11927:http://bugs.mysql.com/11927)
  * STR_TO_DATE() returned NULL if the format string
    contained a space following a non-format character.
  * yaSSL crashed on pre-Pentium Intel CPUs.
  * Selecting into variables sometimes returned incorrect
    wrong results. (Bug#20836:http://bugs.mysql.com/20836)
  * Inserting DEFAULT into a column with no default value
    could result in garbage in the column. Now the same
    result occurs as when inserting NULL into a NOT NULL
    column. (Bug#20691:http://bugs.mysql.com/20691)
  * mysql_fix_privilege_tables.sql altered the
    table_privs.table_priv column to contain too few
    privileges, causing loss of the CREATE VIEW and SHOW VIEW
    privileges. (Bug#20589:http://bugs.mysql.com/20589)
  * A query with a subquery that references columns of a view
    from the outer SELECT could return an incorrect result if
    used from a prepared statement.
  * A server crash occurred when using LOAD DATA to load a
    table containing a NOT NULL spatial column, when the
    statement did not load the spatial column. Now a NULL
    supplied to NOT NULL column error occurs.
  * Unsigned BIGINT values treated as signed values by the
    MOD() function. (Bug#19955:http://bugs.mysql.com/19955)
  * Compiling PHP 5.1 with the MySQL static libraries failed
    on some versions of Linux.
  * The DELIMITER statement did not work correctly when used
    in an SQL file run using the SOURCE statement.
  * VARBINARY column values inserted on a MySQL 4.1 server
    had trailing zeroes following upgrade to MySQL 5.0 or
    later. (Bug#19371:http://bugs.mysql.com/19371)
  * Constant expressions and some numeric constants used as
    input parameters to user-defined functions were not
    treated as constants.
  * Subqueries of the form NULL IN (SELECT ...) returned
    invalid results. (Bug#8804:http://bugs.mysql.com/8804,
  * The --extern option for mysql-test-run.pl did not
    function correctly.
  * INET_ATON() returned a signed BIGINT value, not an
    unsigned value. (Bug#21466:http://bugs.mysql.com/21466)
  * ALTER TABLE statements that performed both RENAME TO and
    {ENABLE|DISABLE} KEYS operations caused a server crash.
  * myisampack wrote to unallocated memory, causing a crash.
  * Some small double precision numbers (such as
    1.00000001e-300) that should have been accepted were
    truncated to zero.
  * The mysql.server script used the source command, which is
    less portable than the . command; it now uses . instead.
  * DATE_ADD() requires complete dates with no "zero" parts,
    but sometimes did not return NULL when given such a date.
  * Using FLUSH TABLES in one connection while another
    connection is using HANDLER statements caused a server
    crash. (Bug#21587:http://bugs.mysql.com/21587)
  * FLUSH LOGS or mysqladmin flush-logs caused a server crash
    if the binary log was not open.
  * Subqueries for which a pushed-down condition did not
    produce exactly one key field could cause a server crash.
  * LAST_DAY('0000-00-00') could cause a server crash.
  * Through the C API, the member strings in MYSQL_FIELD for
    a query that contains expressions may return incorrect
    results. (Bug#21635:http://bugs.mysql.com/21635)
  * mysql_affected_rows() could return values different from
    mysql_stmt_affected_rows() for the same sequence of
    statements. (Bug#23383:http://bugs.mysql.com/23383)
  * IN() and CHAR() can return NULL, but did not signal that
    to the query processor, causing incorrect results for IS
    NULL operations. (Bug#17047:http://bugs.mysql.com/17047)
  * A trigger that invoked a stored function could cause a
    server crash when activated by different client
    connections. (Bug#23651:http://bugs.mysql.com/23651)
  * CONCURRENT did not work correctly for LOAD DATA INFILE.
  * Inserting a default or invalid value into a spatial
    column could fail with Unknown error rather than a more
    appropriate error.
  * The server could send incorrect column count information
    to the client for queries that produce a larger number of
    columns than can fit in a two-byte number.
  * Evaluation of subqueries that require the filesort
    algorithm were allocating and freeing the
    sort_buffer_size buffer many times, resulting in slow
    performance. Now the buffer is allocated once and reused.
  * SQL statements close to the size of max_allowed_packet
    could produce binary log events larger than
    max_allowed_packet that could not be read by slave
    servers. (Bug#19402:http://bugs.mysql.com/19402)
  * View columns were always handled as having implicit
    derivation, leading to illegal mix of collation errors
    for some views in UNION operations. Now view column
    column derivation comes from the original expression
    given in the view definition.
  * If elements in a non-top-level IN subquery were accessed
    by an index and the subquery result set included a NULL
    value, the quantified predicate that contained the
    subquery was evaluated to NULL when it should return a
    non-NULL value. (Bug#23478:http://bugs.mysql.com/23478)
  * Calculation of COUNT(DISTINCT), AVG(DISTINCT), or
    SUM(DISTINCT) when they are referenced more than once in
    a single query with GROUP BY could cause a server crash.
  * For a cast of a DATETIME value containing microseconds to
    DECIMAL, the microseconds part was truncated without
    generating a warning. Now the microseconds part is
    preserved. (Bug#19491:http://bugs.mysql.com/19491)
  * Metadata for columns calculated from scalar subqueries
    was limited to integer, double, or string, even if the
    actual type of the column was different.
  * The result for CAST() when casting a value to UNSIGNED
    was limited to the maximum signed BIGINT value, not the
    maximum unsigned value.
  * Using EXPLAIN caused a server crash for queries that
    selected from INFORMATION_SCHEMA in a subquery in the
    FROM clause. (Bug#22413:http://bugs.mysql.com/22413)
  * Invalidating the query cache caused a server crash for
    INSERT INTO ... SELECT statements that selected from a
    view. (Bug#20045:http://bugs.mysql.com/20045)
  * Slave servers would retry the execution of a SQL
    statement an infinite number of times, ignoring the value
    SLAVE_TRANSACTION_RETRIES when using the NDB engine.
  * On slave servers, transactions that exceeded the lock
    wait timeout failed to roll back properly.
  * Changes to character set variables prior to an action on
    a replication-ignored table were forgotten by slave
    servers. (Bug#22877:http://bugs.mysql.com/22877)
  * With lower_case_table_names set to 1, SHOW CREATE TABLE
    printed incorrect output for table names containing
  * When applying the group_concat_max_len limit,
    GROUP_CONCAT() could truncate multi-byte characters in
    the middle. (Bug#23451:http://bugs.mysql.com/23451)
  * For some problems relating to character set conversion or
    incorrect string values for INSERT or UPDATE, the server
    was reporting truncation or length errors instead.
  * NDB Cluster: Data nodes added while the cluster was
    running in single user mode were all assigned node ID 0,
    which could later cause multiple node failures. Adding of
    nodes in single user mode is no longer possible.
  * NDB Cluster: Attempting to create an NDB table on a MySQL
    with an existing non-Cluster table with the same name in
    the same database could result in data loss or
    corruption. MySQL now issues a warning when a SHOW TABLES
    or other statement causing table discovery finds such a
    table. (Bug#21378:http://bugs.mysql.com/21378)
  * NDB Cluster (NDB API): Inacivity timeouts for scans were
    not correctly handled.
  * NDB Cluster (NDB API): Attempting to read a nonexistent
    tuple using Commit mode for NdbTransaction::execute()
    caused node failures.
  * NDB Cluster (NDB API): Scans closed before being executed
    were still placed in the send queue.
  * NDB Cluster (NDB API): The NdbOperation::getBlobHandle()
    method, when called with the name of a nonexistent
    column, caused a segmentation fault.
  * NDB Cluster: A problem with takeover during a system
    restart caused ordered indexes to be rebuilt incorrectly.
  * NDB Cluster: The ndb_config utility did not perform host
    lookups correctly when using the --host option.
  * NDB Cluster: The ndb_config utility did not perform host
    lookups correctly when using the --host option
  * NDB Cluster: The error returned by the cluster when too
    many nodes were defined did not make clear the nature of
    the problem. (Bug#19045:http://bugs.mysql.com/19045)
  * NDB Cluster: ndb_mgm -e show | head would hang after
    displaying the first 10 lines of output.
  * NDB Cluster: In rare situations with resource shortages,
    a crash could result from insufficient
  * NDB Cluster: ndb_restore did not always make clear that
    it had recovered successfully from temporary errors while
    restoring a cluster backup.
  * NDB Cluster: Error messages given when trying to make
    online changes parameters such as NoOfReplicas thast can
    only be changed via a complete shutdown and restart of
    the cluster did not indicate the true nature of the
    problem. (Bug#19787:http://bugs.mysql.com/19787)
  * NDB Cluster: Following the restart of an MGM node, the
    Cluster management client did not automatically
    reconnect. (Bug#19873:http://bugs.mysql.com/19873)
  * NDB Cluster: In some cases where SELECT COUNT(*) from an
    NDB table should have yielded an error, MAX_INT was
    returned instead.
  * NDB Cluster (NDB API): When multiple processes or threads
    in parallel performed the same ordered scan with
    exclusive lock and updating the retrieved records, the
    scan could skip some records, which were not updated as a
    result. (Bug#20446:http://bugs.mysql.com/20446)
  * NDB Cluster: Using an invalid node ID with the management
    client STOP command could cause ndb_mgm to hang.
  * NDB Cluster: Under some circumstances, local
    checkpointing would hang, keeping any unstarted nodes
    from being started.
  * NDB Cluster: Condition pushdown did not work correctly
    with DATETIME columns.
  * NDB Cluster: When inserting a row into an NDB table with
    a duplicate value for a non-primary unique key, the error
    issued would reference the wrong key.
  * NDB Cluster: Cluster logs were not rotated following the
    first rotation cycle.
  * NDB Cluster: The ndb_mgm management client did not set
    the exit status on errors, always returning 0 instead.
  * NDB Cluster: Partition distribution keys were updated
    only for the primary and starting replicas during node
    recovery. This could lead to node failure recovery for
    clusters having an odd number of replicas.
    Note: We recommend values for NumberOfReplicas that are
    even powers of 2, for best results.
  * NDB Cluster: The output for the --help option used with
    NDB executable programs (ndbd, ndb_mgm, ndb_restore,
    ndb_config, and so on) referred to the Ndb.cfg file,
    instead of my.cnf.
  * NDB Cluster: The node recovery algorithm was missing a
    version check for tables in the ALTER_TABLE_COMMITTED
    state (as opposed to the TABLE_ADD_COMMITTED state, which
    has the version check). This could cause inconsistent
    schemas across nodes following node recovery.
  * NDB Cluster: A scan timeout returned Error 4028 (Node
    failure caused abort of transaction) instead of Error
    4008 (Node failure caused abort of transaction...).
  * NDB Cluster: The --help output from NDB binaries did not
    include file-related options.
  * NDB Cluster: Multiple node restarts in rapid succession
    could cause a system restart to fail
    (Bug#22892:http://bugs.mysql.com/22892), or induce a
    race condition (Bug#23210:http://bugs.mysql.com/23210).
  * NDB Cluster: If a node restart could not be performed
    from the REDO log, no node takeover took place. This
    could cause partitions to be left empty during a system
    restart. (Bug#22893:http://bugs.mysql.com/22893)
    table could lead to deadlocks and memory leaks.
  * NDB Cluster: The management client command ALL DUMP 1000
    would cause the cluster to crash if data nodes were
    connected to the cluster but not yret fully started.
  * NDB Cluster: Cluster backups would fail when there were
    more than 2048 schema objects in the cluster.
  * NDB Cluster: Restoring a cluster failed if there were any
    tables with 128 or more columns.
  * The internal SQL interpreter of InnoDB placed an
    unnecessary lock on the supremum record when
    innodb_locks_unsafe_for_binlog=1. This caused an
    assertion failure when InnoDB was built with debugging
    enabled. (Bug#23769:http://bugs.mysql.com/23769)
  * If a table contains an AUTO_INCREMENT column, inserting
    into an insertable view on the table that does not
    include the AUTO_INCREMENT column should not change the
    value of LAST_INSERT_ID(), because the side effects of
    inserting default values into columns not part of the
    view should not be visible. MySQL was incorrectly setting
    LAST_INSERT_ID() to zero.
  * M % 0 returns NULL, but (M % 0) IS NULL evaluated to
    false. (Bug#23411:http://bugs.mysql.com/23411)
  * Within a stored routine, a view definition cannot refer
    to routine parameters or local variables. However, an
    error did not occur until the routine was called. Now it
    occurs during parsing of the routine creation statement.
    Note: A side effect of this fix is that if you have
    already created such routines, and error will occur if
    FUNCTION. You should drop these routines because they are
  * A client library crash was caused by executing a
    statement such as SELECT * FROM t1 PROCEDURE ANALYSE()
    using a server side cursor on a table t1 that does not
    have the same number of columns as the output from
  * mysql did not check for errors when fetching data during
    result set printing.
  * Adding a day, month, or year interval to a DATE value
    produced a DATE, but adding a week interval produced a
    DATETIME value. Now all produce a DATE value.
  * The column default value in the output from SHOW COLUMNS
    to 64 characters.
  * For not-yet-authenticated connections, the Time column in
    SHOW PROCESSLIST was a random value rather than NULL.
  * The Host column in SHOW PROCESSLIST output was blank when
    the server was started with the --skip-grant-tables
    option. (Bug#22728:http://bugs.mysql.com/22728)
  * The Handler_rollback status variable sometimes was
    incremented when no rollback had taken place.
  * Within a prepared statement, SELECT (COUNT(*) = 1) (or
    similar use of other aggregate functions) did not return
    the correct result for statement re-execution.
  * Lack of validation for input and output TIME values
    resulted in several problems: SEC_TO_TIME() within
    subqueries incorrectly clipped large values;
    SEC_TO_TIME() treated BIGINT UNSIGNED values as signed;
    only truncation warnings were produced when both
    truncation and out-of-range TIME values occurred.
  * Range searches on columns with an index prefix could miss
    records. (Bug#20732:http://bugs.mysql.com/20732)
  * With SQL_MODE=TRADITIONAL, MySQL incorrectly aborted on
    warnings within stored routines and triggers.
  * In mysql, invoking connect or \r with very long db_name
    or host_name parameters caused buffer overflow.
  * mysqldump --xml produced invalid XML for BLOB data.
  * For a debug server, a reference to an undefined user
    variable in a prepared statment executed with EXECUTE
    caused an assertion failure.
  * Within a trigger for a base table, selecting from a view
    on that base table failed.
  * DELETE IGNORE could hang for foreign key parent deletes.
  * Transient errors in replication from master to slave may
    trigger multiple Got fatal error 1236: 'binlog truncated
    in the middle of event' errors on the slave.
  * The value of the warning_count system variable was not
    being calculated correctly (also affecting SHOW COUNT(*)
    WARNINGS). (Bug#19024:http://bugs.mysql.com/19024)
  * InnoDB exhibited thread thrashing with more than 50
    concurrent connections under an update-intensive
    workload. (Bug#22868:http://bugs.mysql.com/22868)
  * InnoDB showed substandard performance with multiple
    queries running concurrently.
  * There was a race condition in the InnoDB
    fil_flush_file_spaces() function.
  * FROM_UNIXTIME() did not accept arguments up to
    POWER(2,31)-1, which it had previously.
  * Some yaSSL-related memory leaks detected by Valgrind were
    fixed. (Bug#23981:http://bugs.mysql.com/23981)
  * If COMPRESS() returned NULL, subsequent invocations of
    COMPRESS() within a result set or within a trigger also
    returned NULL. (Bug#23254:http://bugs.mysql.com/23254)
  * mysql would lose its connection to the server if its
    standard output was not writable.
  * mysql-test-run did not work correctly for RPM-based
    installations. (Bug#17194:http://bugs.mysql.com/17194)
  * The return value from my_seek() was ignored.
  * Use of PREPARE with a CREATE PROCEDURE statement that
    contained a syntax error caused a server crash.
  * Use of a DES-encrypted SSL certificate file caused a
    server crash. (Bug#21868:http://bugs.mysql.com/21868)
  * Column names were not quoted properly for replicated
    views. (Bug#19736:http://bugs.mysql.com/19736)
  * InnoDB used table locks (not row locks) within stored
    functions. (Bug#18077:http://bugs.mysql.com/18077)
  * Statements such as DROP PROCEDURE and DROP VIEW were
    written to the binary log too late due to a race
    condition. (Bug#14262:http://bugs.mysql.com/14262)
  * MySQL would fail to build on the Alpha platform.
  * The optimizer failed to use equality propagation for
    BETWEEN and IN predicates with string arguments.
  * The optimizer used the ref join type rather than eq_ref
    for a simple join on strings.
  * The WITH CHECK OPTION for a view failed to prevent
    storing invalid column values for UPDATE statements.
  * A literal string in a GROUP BY clause could be
    interpreted as a column name.
  * Some queries that used MAX() and GROUP BY could
    incorrectly return an empty result.
  * WITH ROLLUP could group unequal values.
  * Use of a subquery that invoked a function in the column
    list of the outer query resulted in a memory leak.
  * LIKE searches failed for indexed utf8 character columns.
  * FLUSH INSTANCES in Instance Manager triggered an
    assertion failure.
  * ALTER TABLE was not able to rename a view.
  * Entries in the slow query log could have an incorrect
    Rows_examined value.
  * Insufficient memory (myisam_sort_buffer_size) could cause
    a server crash for several operations on MyISAM tables:
    repair table, create index by sort, repair by sort,
    parallel repair, bulk insert.
  * OPTIMIZE TABLE with myisam_repair_threads > 1 could
    result in MyISAM table corruption.
  * Selecting from a MERGE table could result in a server
    crash if the underlying tables had fewer indexes than the
    MERGE table itself.
  * A locking safety check in InnoDB reported a spurious
    error stored_select_lock_type is 0 inside ::start_stmt()
    for INSERT ... SELECT statements in
    innodb_locks_unsafe_for_binlog mode. The safety check was
    removed. (Bug#10746:http://bugs.mysql.com/10746)
  * For multiple-table UPDATE statements, storage engines
    were not notified of duplicate-key errors.
  * Incorrect results could be obtained from re-execution of
    a parametrized prepared statement or a stored routine
    with a SELECT that uses LEFT JOIN with a second table
    having only one row.
  * An UPDATE that referred to a key column in the WHERE
    clause and activated a trigger that modified the column
    resulted in a loop.
  * Creating a TEMPORARY table with the same name as an
    existing table that was locked by another client could
    result in a lock conflict for DROP TEMPORARY TABLE
    because the server unnecessarily tried to acquire a name
    lock. (Bug#21096:http://bugs.mysql.com/21096)
    TABLES, attempts to drop or alter a stored routine failed
    with an error that the routine did not exist, and
    attempts to execute the routine failed with a lock
    conflict error. (Bug#21414:http://bugs.mysql.com/21414)
  * SHOW VARIABLES truncated the Value field to 256
    characters. (Bug#20862:http://bugs.mysql.com/20862)
  * Instance Manager didn't close the client socket file when
    starting a new mysqld instance. mysqld inherited the
    socket, causing clients connected to Instance Manager to
    hang. (Bug#12751:http://bugs.mysql.com/12751)
  * Instance Manager had a race condition involving mysqld
    PID file removal.
  * It was possible for a stored routine with a non-latin1
    name to cause a stack overrun.
  * MySQL 5.0.26 introduced an ABI incompatibility, which
    this release reverts. Programs compiled against 5.0.26
    are not compatible with any other version and must be
    recompiled. (Bug#23427:http://bugs.mysql.com/23427)
  * InnoDB: Reduced optimization level for Windows 64 builds
    to handle possible memory overrun.