Functionality added or changed:
  * If you use SSL for a client connection, you can tell the
    client not to authenticate the server certificate by
    specifying neither --ssl-ca nor --ssl-capath. The server
    still verifies the client according to any applicable
    requirements established via GRANT statements for the
    client, and it still uses any --ssl-ca/--ssl-capath
    values that were passed to server at startup time.
  * Prefix lengths for columns in SPATIAL indexes are no
    longer displayed in SHOW CREATE TABLE output. mysqldump
    uses that statement, so if a table with SPATIAL indexes
    containing prefixed columns is dumped and reloaded, the
    index is created with no prefixes. (The full column width
    of each column is indexed.)
  * The output of mysql --xml and mysqldump --xml now
    includes a valid XML namespace.
  * The mysql_create_system_tables script was removed because
    mysql_install_db no longer uses it in MySQL 5.0.
  * The syntax for index hints has been extended to enable
    explicit specification that the hint applies only to join
    processing. See Section, "Index Hint Syntax."
  * Binary distributions for some platforms did not include
    shared libraries; now shared libraries are shipped for
    all platforms except AIX 5.2 64-bit.
  * NDB Cluster: It is now possible to restore selected
    databases or tables using ndb_restore.
  * NDB Cluster: Several options have been added for use with
    ndb_restore --print_data to facilitate the creation of
    data dump files. (Bug#26900:http://bugs.mysql.com/26900)
  * To satisfy different user requirements, we provide
    several servers. mysqld is an optimized server that is a
    smaller, faster binary. Each package now also includes
    mysqld-debug, which is compiled with debugging support
    but is otherwise configured identically to the non-debug
  * Added the --secure-file-priv option for
    mysql-test-run.pl, which limits the effect of the
    load_file command for mysqltest and for the LOAD DATA and
    SELECT ... INTO OUTFILE statements to work with files in
    a given directory.
  * Added the hostname system variable, which the server sets
    at startup to the server hostname.
  * The server now includes a timestamp in error messages
    that are logged as a result of unhandled signals (such as
    mysqld got signal 11 messages).

Bugs fixed:
  * The patches for Bug#19370:http://bugs.mysql.com/19370
    and Bug#21789:http://bugs.mysql.com/21789were reverted.
  * NDB Cluster: NDB tables having MEDIUMINT AUTO_INCREMENT
    columns were not restored correctly by ndb_restore,
    causing spurious duplicate key errors. This issue did not
    affect TINYINT, INT, or BIGINT columns with
    AUTO_INCREMENT. (Bug#27775:http://bugs.mysql.com/27775)
  * NDB Cluster: NDB tables with indexes whose names
    contained space characters were not restored correctly by
    ndb_restore (the index names were truncated).
  * NDB Cluster: Some queries that updated multiple tables
    were not backed up correctly.
  * NDB Cluster: Joins on multiple tables containing BLOB
    columns could cause data nodes run out of memory, and to
    crash with the error NdbObjectIdMap::expand unable to
    expand. (Bug#26176:http://bugs.mysql.com/26176)
  * NDB Cluster (APIs): Using NdbBlob::writeData() to write
    data in the middle of an existing blob value (that is,
    updating the value) could overwrite some data past the
    end of the data to be changed.
  * NDB Cluster: Under certain rare circumstances, DROP TABLE
    or TRUNCATE of an NDB table could cause a node failure or
    forced cluster shutdown.
  * NDB Cluster: Memory usage of a mysqld process grew even
    while idle. (Bug#27560:http://bugs.mysql.com/27560)
  * NDB Cluster: In some cases, AFTER UPDATE and AFTER DELETE
    triggers on NDB tables that referenced subject table did
    not see the results of operation which caused invocation
    of the trigger, but rather saw the row as it was prior to
    the update or delete operation.
    This was most noticeable when an update operation used a
    subquery to obtain the rows to be updated. An example
    would be UPDATE tbl1 SET col2 = val1 WHERE tbl1.col1 IN
    (SELECT col3 FROM tbl2 WHERE c4 = val2) where there was
    an AFTER UPDATE trigger on table tbl1. In such cases, the
    trigger would fail to execute.
    The problem occurred because the actual update or delete
    operations were deferred to be able to perform them later
    as one batch. The fix for this bug solves the problem by
    disabling this optimization for a given update or delete
    if the table has an AFTER trigger defined for this
    operation. (Bug#26242:http://bugs.mysql.com/26242)
  * NDB Cluster: Condition pushdown did not work with
    prepared statements.
  * NDB Cluster: When trying to create tables on an SQL node
    not connected to the cluster, a misleading error message
    Table 'tbl_name' already exists was generated. The error
    now generated is Could not connect to storage engine.
  * NDB Cluster: Error messages displayed when running in
    single user mode were inconsistent.
  * NDB Cluster: On Solaris, the value of an NDB table column
    declared as BIT(33) was always displayed as 0.
  * NDB Cluster: The output from ndb_restore --print_data was
    incorrect for a backup made of a database containing
    tables with TINYINT or SMALLINT columns.
  * NDB Cluster: After entering single user mode it was not
    possible to alter non-NDB tables on any SQL nodes other
    than the one having sole access to the cluster.
  * NDB Cluster: The failure of a data node while restarting
    could cause other data nodes to hang or crash.
  * NDB Cluster: The management client command node_id STATUS
    displayed the message Node node_id: not connected when
    node_id was not the node ID of a data node.
    The ALL STATUS command in the cluster management client
    still displays status information for data nodes only.
    This is by design. See Section 15.7.2, "Commands in the
    MySQL Cluster Management Client," for more information.
  * NDB Cluster: It was not possible to set
    LockPagesInMainMemory equal to 0.
  * NDB Cluster: A race condition could sometimes occur if
    the node acting as master failed while node IDs were
    still being allocated during startup.
  * NDB Cluster: When a data node was taking over as the
    master node, a race condition could sometimes occur as
    the node was assuming responsibility for handling of
    global checkpoints.
  * NDB Cluster: mysqld processes would sometimes crash under
    high load. (Bug#26825:http://bugs.mysql.com/26825)
  * NDB Cluster: Some values of MaxNoOfTables caused the
    error Job buffer congestion to occur.
  * Some equi-joins containing a WHERE clause that included a
    NOT IN subquery caused a server crash.
  * Windows binaries contained no debug symbol file. Now .map
    and .pdb files are included in 32-bit builds for
    mysqld-nt.exe, mysqld-debug.exe, and mysqlmanager.exe.
  * The test for the MYSQL_OPT_SSL_VERIFY_SERVER_CERT option
    for mysql_options() was performed incorrectly. Also
    changed as a result of this bugfix: The arg option for
    the mysql_options() C API function was changed from char
    * to void *. (Bug#24121:http://bugs.mysql.com/24121)
  * The range optimizer could consume a combinatorial amount
    of memory for certain classes of WHERE clauses.
  * Conversion of DATETIME values in numeric contexts
    sometimes did not produce a double
    (YYYYMMDDHHMMSS.uuuuuu) value.
  * Passing nested row expressions with different structures
    to an IN predicate caused a server crash.
  * SELECT DISTINCT could return incorrect results if the
    select list contained duplicated columns.
  * A subquery could get incorrect values for references to
    outer query columns when it contained aggregate functions
    that were aggregated in outer context.
  * In some cases, the optimizer preferred a range or full
    index scan access method over lookup access methods when
    the latter were much cheaper.
  * Duplicates were not properly identified among
    (potentially) long strings used as arguments for
  * For InnoDB, fixed consistent-read behavior of the first
    read statement, if the read was served from the query
    cache, for the READ COMMITTED isolation level.
  * The decimal.h header file was incorrectly omitted from
    binary distributions.
  * Duplicate members in SET definitions were not detected.
    Now they result in a warning; if strict SQL mode is
    enabled, an error occurs instead.
  * For INSERT INTO ... SELECT where index searches used
    column prefixes, insert errors could occur when key value
    type conversion was done.
    INFORMATION was not always cleared properly.
  * mysqldump could crash or exhibit incorrect behavior when
    some options were given very long values, such as
    --fields-terminated-by="some very long string". The code
    has been cleaned up to remove a number of fixed-sized
    buffers and to be more careful about error conditions in
    memory allocation.
  * Setting a column to NOT NULL with an ON DELETE SET NULL
    clause foreign key crashes the server.
  * The values displayed for the Innodb_row_lock_time,
    Innodb_row_lock_time_avg, and Innodb_row_lock_time_max
    status variables were incorrect.
  * COUNT(decimal_expr) sometimes generated a spurious
    truncation warning.
  * With NO_AUTO_VALUE_ON_ZERO SQL mode enabled, LOAD DATA
    operations could assign incorrect AUTO_INCREMENT values.
  * Incorrect results could be returned for some queries that
    contained a select list expression with IN or BETWEEN
    together with an ORDER BY or GROUP BY on the same
    expression using NOT IN or NOT BETWEEN.
  * Queries containing subqueries with COUNT(*) aggregated in
    an outer context returned incorrect results. This
    happened only if the subquery did not contain any
    references to outer columns.
  * Use of an aggregate function from an outer context as an
    argument to GROUP_CONCAT() caused a server crash.
  * REPAIR TABLE ... USE_FRM with an ARCHIVE table deleted
    all records from the table.
  * On Windows, debug builds of mysqld could fail with heap
    assertions. (Bug#25765:http://bugs.mysql.com/25765)
  * On Windows, debug builds of mysqlbinlog could fail with a
    memory error. (Bug#23736:http://bugs.mysql.com/23736)
  * String truncation upon insertion into an integer or year
    column did not generate a warning (or an error in strict
    mode). (Bug#26359:http://bugs.mysql.com/26359,
  * In out-of-memory conditions, the server might crash or
    otherwise not report an error to the Windows event log.
  * The temporary file-creation code was cleaned up on
    Windows to improve server stability.
  * Out-of-memory errors for slave I/O threads were not
    reported. Now they are written to the error log.
  * mysqldump crashed for MERGE tables if the
    --complete-insert (-c) option was given.
  * In certain situations, MATCH ... AGAINST returned false
    hits for NULL values produced by LEFT JOIN when no
    full-text index was available.
  * OPTIMIZE TABLE might fail on Windows when it attempts to
    rename a temporary file to the original name if the
    original file had been opened, resulting in loss of the
    .MYD file. (Bug#25521:http://bugs.mysql.com/25521)
  * GRANT statements were not replicated if the server was
    started with the --replicate-ignore-table or
    --replicate-wild-ignore-table option.
  * A problem in handling of aggregate functions in
    subqueries caused predicates containing aggregate
    functions to be ignored during query execution.
  * Improved out-of-memory detection when sending logs from a
    master server to slaves, and log a message when
    allocation fails.
  * MBROverlaps() returned incorrect values in some cases.
  * SHOW CREATE VIEW qualified references to stored functions
    in the view definition with the function's database name,
    even when the database was the default database. This
    affected mysqldump (which uses SHOW CREATE VIEW to dump
    views) because the resulting dump file could not be used
    to reload the database into a different database. SHOW
    CREATE VIEW now suppresses the database name for
    references to functions in the default database.
  * With innodb_file_per_table enabled, attempting to rename
    an InnoDB table to a non-existent database caused the
    server to exit. (Bug#27381:http://bugs.mysql.com/27381)
  * mysql_install_db could terminate with an error after
    failing to determine that a system table already existed.
  * For InnoDB tables having a clustered index that began
    with a CHAR or VARCHAR column, deleting a record and then
    inserting another before the deleted record was purged
    could result in table corruption.
  * Selecting the result of AVG() within a UNION could
    produce incorrect values.
  * An INTO OUTFILE clause is allowed only for the final
    SELECT of a UNION, but this restriction was not being
    enforced correctly.
  * Duplicate entries were not assessed correctly in a MEMORY
    table with a BTREE primary key on a utf8 ENUM column.
  * For MyISAM tables, COUNT(*) could return an incorrect
    value if the WHERE clause compared an indexed TEXT column
    to the empty string (''). This happened if the column
    contained empty strings and also strings starting with
    control characters such as tab or newline.
  * For DELETE FROM tbl_name ORDER BY col_name (with no WHERE
    or LIMIT clause), the server did not check whether
    col_name was a valid column in the table.
  * ALTER VIEW requires the CREATE VIEW and DROP privileges
    for the view. However, if the view was created by another
    user, the server erroneously required the SUPER
    privilege. (Bug#26813:http://bugs.mysql.com/26813)
  * In a view, a column that was defined using a GEOMETRY
    function was treated as having the LONGBLOB data type
    rather than the GEOMETRY type.
  * With the NO_AUTO_VALUE_ON_ZERO SQL mode enabled,
    LAST_INSERT_ID() could return 0 after INSERT ... ON
    DUPLICATE KEY UPDATE. Additionally, the next rows
    inserted (by the same INSERT, or the following INSERT
    with or without ON DUPLICATE KEY UPDATE), would insert 0
    for the auto-generated value if the value for the
    AUTO_INCREMENT column was NULL or missing.
  * For a stored procedure containing a SELECT statement that
    used a complicated join with an ON expression, the
    expression could be ignored during re-execution of the
    procedure, yielding an incorrect result.
  * When RAND() was called multiple times inside a stored
    procedure, the server did not write the correct random
    seed values to the binary log, resulting in incorrect
    replication. (Bug#25543:http://bugs.mysql.com/25543)
  * SOUNDEX() returned an invalid string for international
    characters in multi-byte character sets.
  * Row equalities in WHERE clauses could cause memory
    corruption. (Bug#27154:http://bugs.mysql.com/27154)
  * GROUP BY on a ucs2 column caused a server crash when
    there was at least one empty string in the column.
  * Evaluation of an IN() predicate containing a
    decimal-valued argument caused a server crash.
  * Storing NULL values in spatial fields caused excessive
    memory allocation and crashes on some systems.
  * mysql_stmt_fetch() did an invalid memory deallocation
    when used with the embedded server.
  * In a MEMORY table, using a BTREE index to scan for
    updatable rows could lead to an infinite loop.
  * The range optimizer could cause the server to run out of
    memory. (Bug#26625:http://bugs.mysql.com/26625)
  * The parser accepted illegal code in SQL exception
    handlers, leading to a crash at runtime when executing
    the code. (Bug#26503:http://bugs.mysql.com/26503)
  * Difficult repair or optimization operations could cause
    an assertion failure, resulting in a server crash.
  * Increasing the width of a DECIMAL column could cause
    column values to be changed.
  * Replication between master and slave would infinitely
    retry binary log transmission where the
    max_allowed_packet on the master was larger than that on
    the slave if the size of the transfer was between these
    two values. (Bug#23775:http://bugs.mysql.com/23775)
  * Invalid optimization of pushdown conditions for queries
    where an outer join was guaranteed to read only one row
    from the outer table led to results with too few rows.
  * For INSERT ... ON DUPLICATE KEY UPDATE statements on
    tables containing AUTO_INCREMENT columns,
    LAST_INSERT_ID() was reset to 0 if no rows were
    successfully inserted or changed. "Not changed" includes
    the case where a row was updated to its current values,
    but in that case, LAST_INSERT_ID() should not be reset to
    0. Now LAST_INSERT_ID() is reset to 0 only if no rows
    were successfully inserted or touched, whether or not
    touched rows were changed.
    This bug was introduced by the fix for
  * For an INSERT statement that should fail due to a column
    with no default value not being assigned a value, the
    statement succeeded with no error if the column was
    assigned a value in an ON DUPLICATE KEY UPDATE clause,
    even if that clause was not used.
  * A result set column formed by concatention of string
    literals was incomplete when the column was produced by a
    subquery in the FROM clause.
  * When using the result of SEC_TO_TIME() for time value
    greater than 24 hours in an ORDER BY clause, either
    directly or through a column alias, the rows were sorted
    incorrectly as strings.
  * If the server was started with --skip-grant-tables,
    Selecting from INFORMATION_SCHEMA tables causes a server
    crash. (Bug#26285:http://bugs.mysql.com/26285)
  * Incompatible change: INSERT DELAYED statements are not
    supported for MERGE tables, but the MERGE storage engine
    was not rejecting such statements, resulting in table
    corruption. Applications previously using INSERT DELAYED
    into MERGE table will break when upgrading to versions
    with this fix. To avoid the problem, remove DELAYED from
    such statements. (Bug#26464:http://bugs.mysql.com/26464)
  * NDB Cluster: An invalid pointer was returned following a
    FSCLOSECONF signal when accessing the REDO logs during a
    node restart or system restart.
  * NDB Cluster: An inadvertent use of unaligned data caused
    ndb_restore to fail on some 64-bit platforms, including
    Sparc and Itanium-2.
  * NDB Cluster: An infinite loop in an internal logging
    function could cause trace logs to fill up with Unknown
    Signal type error messages and thus grow to unreasonable
    sizes. (Bug#26720:http://bugs.mysql.com/26720)
  * NDB Cluster: The failure of a data node when restarting
    it with --initial could lead to failures of subsequent
    data node restarts.
  * NDB Cluster: Takeover for local checkpointing due to
    multiple failures of master nodes was sometimes incorrect
    handled. (Bug#26457:http://bugs.mysql.com/26457)
  * NDB Cluster: The LockPagesInMemory parameter was not read
    until after distributed communication had already started
    between cluster nodes. When the value of this parameter
    was 1, this could sometimes result in data node failure
    due to missed heartbeats.
  * NDB Cluster: Under some circumstances, following the
    restart of a management, all cluster data nodes would
    connect to it normally, but some of them subsequently
    failed to log any events to the management node.
  * NDB Cluster: An error was produced when SHOW TABLE STATUS
    was used on an NDB table that had no AUTO_INCREMENT
    column. (Bug#21033:http://bugs.mysql.com/21033)
    value could crash the server.
  * DOUBLE values such as 20070202191048.000000 were being
    treated as illegal arguments by WEEK().
  * AFTER UPDATE triggers were not activated by the update
    part of INSERT ... ON DUPLICATE KEY UPDATE statements.
    This bug was introduced by the fix for
  * For INSERT ... ON DUPLICATE KEY UPDATE statements where
    some AUTO_INCREMENT values were generated automatically
    for inserts and some rows were updated, one
    auto-generated value was lost per updated row, leading to
    faster exhaustion of the range of the AUTO_INCREMENT
    column. (Bug#24432:http://bugs.mysql.com/24432)
    Because the original problem can affect replication
    (different values on master and slave), it is recommended
    that the master and its slaves be upgraded to the current
  * IN ((subquery)), IN (((subquery))), and so forth, are
    equivalent to IN (subquery), which is always interpreted
    as a table subquery (so that it is allowed to return more
    than one row). MySQL was treating the
    "over-parenthesized" subquery as a single-row subquery
    and rejecting it if it returned more than one row. This
    bug primarily affected automatically generated code (such
    as queries generated by Hibernate), because humans rarely
    write the over-parenthesized forms.
  * For MERGE tables defined on underlying tables that
    contained a short VARCHAR column (shorter than four
    characters), using ALTER TABLE on at least one but not
    all of the underlying tables caused the table definitions
    to be considered different from that of the MERGE table,
    even if the ALTER TABLE did not change the definition.
  * If a thread previously serviced a connection that was
    killed, excessive memory and CPU use by the thread
    occurred if it later serviced a connection that had to
    wait for a table lock.
  * CURDATE() is less than NOW(), either when comparing
    CURDATE() directly (CURDATE() < NOW() is true) or when
    casting CURDATE() to DATE (CAST(CURDATE() AS DATE) <
    NOW() is true). However, storing CURDATE() in a DATE
    column and comparing col_name < NOW() incorrectly yielded
    false. This is fixed by comparing a DATE column as
    DATETIME for comparisons to a DATETIME constant.
  * A view on a join is insertable for INSERT statements that
    store values into only one table of the join. However,
    inserts were being rejected if the inserted-into table
    was used in a self-join because MySQL incorrectly was
    considering the insert to modify multiple tables of the
    view. (Bug#25122:http://bugs.mysql.com/25122)
  * Expressions involving SUM(), when used in an ORDER BY
    clause, could lead to out-of-order results.
  * LOAD DATA INFILE sent an okay to the client before
    writing the binary log and committing the changes to the
    table had finished, thus violating ACID requirements.
  * Views that used a scalar correlated subquery returned
    incorrect results.
  * IF(expr, unsigned_expr, unsigned_expr) was evaluated to a
    signed result, not unsigned. This has been corrected. The
    fix also affects constructs of the form IS [NOT]
    {TRUE|FALSE}, which were transformed internally into IF()
    expressions that evaluated to a signed result.
    For existing views that were defined using IS [NOT]
    {TRUE|FALSE} constructs, there is a related implication.
    The definitions of such views were stored using the IF()
    expression, not the original construct. This is manifest
    in that SHOW CREATE VIEW shows the transformed IF()
    expression, not the original one. Existing views will
    evaluate correctly after the fix, but if you want SHOW
    CREATE VIEW to display the original construct, you must
    drop the view and re-create it using its original
    definition. New views will retain the construct in their
  * BENCHMARK() did not work correctly for expressions that
    produced a DECIMAL result.
  * For some values of the position argument, the INSERT()
    function could insert a NUL byte into the result.
  * Inserting utf8 data into a TEXT column that used a
    single-byte character set could result in spurious
    warnings about truncated data.
  * EXPLAIN EXTENDED did not show WHERE conditions that were
    optimized away. (Bug#22331:http://bugs.mysql.com/22331)
  * INSERT DELAYED statements inserted incorrect values into
    BIT columns. (Bug#26238:http://bugs.mysql.com/26238)
  * For expr IN(value_list), the result could be incorrect if
    BIGINT UNSIGNED values were used for expr or in the value
    list. (Bug#19342:http://bugs.mysql.com/19342)
  * When a TIME_FORMAT() expression was used as a column in a
    GROUP BY clause, the expression result was truncated.
  * For SUBSTRING() evaluation using a temporary table, when
    SUBSTRING() was used on a LONGTEXT column, the max_length
    metadata value of the result was incorrectly calculated
    and set to 0. Consequently, an empty string was returned
    instead of the correct result.
  * Use of a GROUP BY clause that referred to a stored
    function result together with WITH ROLLUP caused
    incorrect results.
  * Use of a subquery containing GROUP BY and WITH ROLLUP
    caused a server crash.
  * Use of a subquery containing a UNION with an invalid
    ORDER BY clause caused a server crash.
  * In certain cases it could happen that deleting a row
    corrupted an RTREE index. This affected indexes on
    spatial columns. (Bug#25673:http://bugs.mysql.com/25673)
  * SSL connections failed on Windows.
  * Added support for --debugger=dbx for mysql-test-run.pl
    and fixed support for --debugger=devenv,
    --debugger=DevEnv, and --debugger=/path/to/devenv.
  * X() IS NULL and Y() IS NULL comparisons failed when X()
    and Y() returned NULL.
  * UNHEX() IS NULL comparisons failed when UNHEX() returned
    NULL. (Bug#26537:http://bugs.mysql.com/26537)
  * The REPEAT() function did not allow a column name as the
    count parameter. (Bug#25197:http://bugs.mysql.com/25197)
  * On 64-bit Windows, large timestamp values could be
    handled incorrectly.
  * In some error messages, inconsistent format specifiers
    were used for the translations in different languages.
    comp_err (the error message compiler) now checks for
    mismatches. (Bug#26571:http://bugs.mysql.com/26571)
  * On Windows, the server exhibited a file-handle leak after
    reaching the limit on the number of open file
    descriptors. (Bug#25222:http://bugs.mysql.com/25222)
  * A reference to a non-existent column in the ORDER BY
    clause of an UPDATE ... ORDER BY statement could cause a
    server crash. (Bug#25126:http://bugs.mysql.com/25126)
  * A multiple-row delayed insert with an auto increment
    column could cause duplicate entries to be created on the
    slave in a replication environment.
  * Duplicating the usage of a user variable in a stored
    procedure or trigger would not be replicated correctly to
    the slave. (Bug#25167:http://bugs.mysql.com/25167)
  * User defined variables used within stored procedures and
    triggers are not replicated correctly when operating in
    statement-based replication mode.
  * Loading data using LOAD DATA INFILE may not replicate
    correctly (due to character set incompatibilities) if the
    character_set_database variable is set before the data is
    loaded. (Bug#15126:http://bugs.mysql.com/15126)
  * DROP TRIGGER statements would not be filtered on the
    slave when using the replication-wild-do-table option.
  * MySQL would not compile when configured using
  * When using certain server SQL modes, the mysql.proc table
    was not created by mysql_install_db. In addition, the
    creation of this and other MySQL system tables was not
    checked for by mysql-test-run.pl.
  * VIEW restrictions were applied to SELECT statements after
    a CREATE VIEW statement failed, as though the CREATE had
    succeeded. (Bug#25897:http://bugs.mysql.com/25897)
  * An INSERT trigger invoking a stored routine that inserted
    into a table other than the one on which the trigger was
    defined would fail with a Table '...' doesn't exist
    referring to the second table when attempting to delete
    records from the first table.
  * A stored procedure that made use of cursors failed when
    the procedure was invoked from a stored function.
  * When nesting stored procedures within a trigger on a
    table, a false dependency error was thrown when one of
    the nested procedures contained a DROP TABLE statement.
  * When attempting to call a stored procedure creating a
    table from a trigger on a table tbl in a database db, the
    trigger failed with ERROR 1146 (42S02): Table 'db.tbl'
    doesn't exist. However, the actual reason that such a
    trigger fails is due to the fact that CREATE TABLE causes
    an implicit COMMIT, and so a trigger cannot invoke a
    stored routine containing this statement. A trigger which
    does so now fails with ERROR 1422 (HY000): Explicit or
    implicit commit is not allowed in stored function or
    trigger, which makes clear the reason for the trigger's
    failure. (Bug#18914:http://bugs.mysql.com/18914)
  * Local variables in stored routines or triggers, when
    declared as the BIT type, were interpreted as strings.
  * When a stored routine attempted to execute a statement
    accessing a nonexistent table, the error was not caught
    by the routine's exception handler.
  * NOW() returned the wrong value in statements executed at
    server startup with the --init-file option.
  * Instance Manager did not remove the angel PID file on a
    clean shutdown. (Bug#22511:http://bugs.mysql.com/22511)
  * The server could crash if two or more threads initiated
    query cache resize operation at moments very close in
    time. (Bug#23527:http://bugs.mysql.com/23527)
  * The conditions checked by the optimizer to allow use of
    indexes in IN predicate calculations were unnecessarily
    tight and were relaxed.
  * Several deficiencies in resolution of column names for
    INSERT ... SELECT statements were corrected.
  * Indexes on TEXT columns were ignored when ref accesses
    were evaluated. (Bug#25971:http://bugs.mysql.com/25971)
  * The update columns for INSERT ... SELECT ... ON DUPLICATE
    KEY UPDATE could be assigned incorrect values if a
    temporary table was used to evaluate the SELECT.
  * CONNECTION is no longer treated as a reserved word.
  * A user-defined variable could be assigned an incorrect
    value if a temporary table was employed in obtaining the
    result of the query used to determine its value.
  * Queries that used a temporary table for the outer query
    when evaluating a correlated subquery could return
    incorrect results.
  * For index reads, the BLACKHOLE engine did not return
    end-of-file (which it must because BLACKHOLE tables
    contain no rows), causing some queries to crash.