Functionality added or changed:
Important Change: Partitioning: Security Fix:
It was possible, by creating a partitioned table using the
DATA DIRECTORY and
DIRECTORY options to gain privileges on other tables
having the same name as the partitioned table. As a result of
this fix, any table-level
DATA DIRECTORY or
INDEX DIRECTORY options are now ignored for
In MySQL 5.1.6, when log tables were implemented, the default
log destination for the general query and slow query log was
TABLE. This default has been changed to
FILE, which is compatible with MySQL 5.0, but
incompatible with earlier releases of MySQL 5.1 from 5.1.6 to
5.1.20. If you are upgrading from MySQL 5.0 to 5.1.21 or higher,
no logging option changes should be necessary. However, if you
are upgrading from 5.1.6 through 5.1.20 to 5.1.21 or higher and
TABLE logging, use the
explicitly to preserve your server's table-logging behavior.
The MySQL 5.1.23 fix is in addition to a fix in 5.1.21 because it turned out that the default was set in two places, only one of which was fixed the first time. (Bug#29993)
The parser accepted statements that contained
*/ that were not properly closed with
*/, such as
SELECT 1 /* +
2. Statements that contain unclosed
/*-comments now are rejected with a syntax
This fix has the potential to cause incompatibilities. Because
of Bug#26302, which caused the trailing
be truncated from comments in views, stored routines, triggers,
and events, it is possible that objects of those types may have
been stored with definitions that now will be rejected as
syntactically invalid. Such objects should be dropped and
re-created so that their definitions do not contain truncated
MySQL Cluster: The following improvements have been made in the ndb_size.pl utility:
The script can now be used with multiple databases; lists of databases and tables can also be excluded from analysis.
Schema name information has been added to index table calculations.
The database name is now an optional parameter, the exclusion of which causes all databases to be examined.
If selecting from
fails, the script now attempts to fall back to
--real_table_name option has been added;
this designates a table to handle unique index size
The report title has been amended to cover cases where more than one database is being analyzed.
Support for a
--socket option was also added.
For more information, see Section 17.4.21, “ndb_size.pl — NDBCLUSTER Size Requirement Estimator”. (Bug#28683, Bug#28253)
NDB error codes to MySQL
storage engine error codes has been improved.
The output of the ndb_mgm client
now indicates when the cluster is in single user mode.
MySQL Cluster: The output from the cluster management client showing the progress of data node starts has been improved. (Bug#23354)
Partitioning: Error messages for partitioning syntax errors have been made more descriptive. (Bug#29368)
Replication of the following SQL functions now switches to
row-based logging in
MIXED mode, and
generates a warning in
mysqldump information at the top of the
output now shows the same information as
mysqldump invoked with the
-V option, namely the
mysqldump version number, the MySQL server
version, and the distribution.
mysql-test-run.pl now allows a suite name
prefix to be specified in command-line arguments that name test
cases. The test name syntax now is
For example, mysql-test-run.pl binlog.mytest
mytest.test test in the
binlog test suite.
mysqldump produces a
completed on comment
at the end of the dump if
--comments is given. The date
causes dump files for identical data take at different times to
appear to be different. The new options
control whether the date is added to the comment.
suppresses date printing. The default is
--dump-date (include the date
in the comment).
Server parser performance was improved for expression parsing by lowering the number of state transitions and reductions needed. (Bug#30625)
Server parser performance was improved for identifier lists, expression lists, and UDF expression lists. (Bug#30333)
Server parser performance was improved for boolean expressions. (Bug#30237)
LAST_EXECUTED column of the
INFORMATION_SCHEMA.EVENTS table now
indicates when the event started executing rather than when it
finished executing. As a result, the
column is never less than
mysql_odbc_escape_string() C API
function has been removed. It has multi-byte character escaping
issues, doesn't honor the
NO_BACKSLASH_ESCAPES SQL mode
and is not needed anymore by Connector/ODBC as of 3.51.17.
See also Bug#41728.
MyISAM table is created with no
DATA DIRECTORY option, the
.MYD file is created in the database
directory. By default, if
MyISAM finds an
.MYD file in this case, it
overwrites it. The same applies to
files for tables created with no
DIRECTORY option. To suppress this behavior, start the
server with the new
option, in which case
MyISAM will not
overwrite existing files and returns an error instead.
The default value of the
connect_timeout system variable
was increased from 5 to 10 seconds. This might help in cases
where clients frequently encounter errors of the form
Lost connection to MySQL server at
XXX', system error:
MySQL now can be compiled with gcc 4.2.x.
There was a problem involving a conflict with the
mysql-test-run.pl now supports a
--combination option for specifying options to
the mysqld server. This option is similar to
--mysqld but should be given two or more times.
mysql-test-run.pl executes multiple test
runs, using the options for each instance of
--combination in successive runs.
For test runs specific to a given test suite, an alternative to
the use of
--combination is to create a
combinations file in the suite directory.
The file should contain a section of options for each test run.
The argument for the mysql-test-run.pl
options is now interpreted as a Perl regular expression if there
is a pattern metacharacter in the argument value. This allows
more flexible specification of which tests to perform or skip.
LIMIT clause was present, the server
could fail to consider indexes that could be used for
ORDER BY or
Security Fix: Replication:
It was possible for any connected user to issue a
BINLOG statement, which could be
used to escalate privileges.
Security Fix: Three vulnerabilities in yaSSL versions 1.7.5 and earlier were discovered that could lead to a server crash or execution of unauthorized code. The exploit requires a server with yaSSL enabled and TCP/IP connections enabled, but does not require valid MySQL account credentials. The exploit does not apply to OpenSSL.
The proof-of-concept exploit is freely available on the Internet. Everyone with a vulnerable MySQL configuration is advised to upgrade immediately.
RENAME TABLE against a
table with explicit
DATA DIRECTORY and
INDEX DIRECTORY options can be used to
overwrite system table information by replacing the symbolic
link points. the file to which the symlink points.
ALTER VIEW retained the original
DEFINER value, even when altered by another
user, which could allow that user to gain the access rights of
the view. Now
ALTER VIEW is
allowed only to the original definer or users with the
When using a
FEDERATED table, the local
server could be forced to crash if the remote server returned a
result with fewer columns than expected.
Security Enhancement: It was possible to force an error message of excessive length which could lead to a buffer overflow. This has been made no longer possible as a security precaution. (Bug#32707)
It is no longer possible to create
NULL columns. However, for backwards
compatibility, you can continue to use such tables that were
created in previous MySQL releases.
mode enabled, queries such as
SELECT a FROM t1 HAVING
COUNT(*)>2 were not being rejected as they should
This fix results in the following behavior:
There is a check against mixing group and nongroup columns
This check is done both for the select list and for the
HAVING clause if there is one.
This behavior differs from previous versions as follows:
Inserting a row with a
NULL value for a
DATETIME column results in a
CSV file that the storage engine cannot read.
mysql_install_db script could fail to
locate some components (including resolveip)
during execution if the
--basedir option was
specified on the command-line or within the
my.cnf file. This was due to a conflict
when comparing the compiled-in values and the supplied values.
--source-install command-line option to the
script has been removed and replaced with the
mysql_install_db now locates components
either using the compiled-in options, the
--basedir option or
containing ambiguous aliases could have unintended side effects
such as deleting rows from the wrong table. Example:
DELETE FROM t1 AS a2 USING t1 AS a1 INNER JOIN t2 AS a2;
Now alias declarations can be declared only in the
table_references part. Elsewhere in
the statement, alias references are allowed but not alias
declarations. Statements containing aliases that are no longer
allowed must be rewritten.
See also Bug#27525.
Within a stored routine, it is no longer allowable to declare a
cursor for a
DESCRIBE statement. This happened
to work in some instances, but is no longer supported. In many
cases, a workaround for this change is to use the cursor with a
SELECT query to read from an
INFORMATION_SCHEMA table that produces the
same information as the
Incompatible Change: It was possible to create a view having a column whose name consisted of an empty string or space characters only.
One result of this bug fix is that aliases for columns in the
SELECT statement are checked to ensure
that they are legal column names. In particular, the length must
be within the maximum column length of 64 characters, not the
maximum alias length of 256 characters. This can cause problems
for replication or loading dump files. For additional
information and workarounds, see
Section D.4, “Restrictions on Views”.
See also Bug#31202.
Several type-preserving functions and operators returned an
incorrect result type that does not match their argument types:
CASE. These now aggregate using the
precise SQL types of their arguments rather than the internal
type. In addition, the result type of the
STR_TO_DATE() function is now
DATETIME by default.
Incompatible Change: It was possible for option files to be read twice at program startup, if some of the standard option file locations turned out to be the same directory. Now duplicates are removed from the list of files to be read.
Also, users could not override system-wide settings using
was read last. The latter file now is read earlier so that
~/.my.cnf can override system-wide
The fix for this problem had a side effect such that on Unix,
MySQL programs looked for options in
~/my.cnf rather than the standard location
~/.my.cnf. That problem was addressed as
A number of problems existed in the implementation of
MERGE tables that could cause problems. The
problems are summarized below:
A thread trying to lock a
performs busy waiting while
TABLE or a similar table administration task
is ongoing on one or more of its
A thread trying to lock a
performs busy waiting until all threads that did
REPAIR TABLE or similar
table administration tasks on one or more of its
MyISAM tables in
LOCK TABLES segments do
TABLES. The difference against problem #1 is
that the busy waiting takes place after the
administration task. It is terminated by
TABLES within a
TABLES segment can invalidate the lock. This
does not require a
MERGE table. The
TABLES can be replaced by any statement that
requires other threads to reopen the table. In 5.0 and
5.1 a single
TABLES can provoke the problem.
Trying DML on a
MERGE table, which has a
child locked and repaired by another thread, made an
infinite loop in the server.
MERGE table and its children in
parent-child order and flushing the child deadlocked the
MERGE child, while the
MERGE table was in use, let the truncate
fail instead of waiting for the table to become free.
Repairing a child of an open
corrupted the child. It was necessary to
FLUSH the child first.
Flushing and optimizing locked
children crashed the server.
Bug#19627 - temporary merge table locking
Use of a temporary
MERGE table with
nontemporary children could corrupt the children.
Temporary tables are never locked. Creation of tables with
nontemporary children of a temporary
MERGE table is now prohibited.
MERGE table possible
It was possible to create a
MERGE tables: Can't lock file
This was a Windows-only bug. Table administration statements sometimes failed with "Can't lock file (errno: 155)".
The fix introduces the following changes in behavior:
This patch changes the behavior of temporary
MERGE tables. Temporary
MERGE must have temporary children. The
old behavior was wrong. A temporary table is not locked.
Hence even nontemporary children were not locked. See Bug#19627.
You cannot change the union list of a nontemporary
MERGE table when
LOCK TABLES is in effect. The
following does not work:
CREATE TABLE m1 ... ENGINE=MRG_MYISAM ...; LOCK TABLES t1 WRITE, t2 WRITE, m1 WRITE; ALTER TABLE m1 ... UNION=(t1,t2) ...;
However, you can do this with a temporary
You cannot create a
MERGE table with
CREATE ... SELECT, neither as a temporary
MERGE table, nor as a nontemporary
MERGE table. For example,
TABLE m1 ... ENGINE=MRG_MYISAM ... SELECT ...;
causes the error message: table is not BASE
Important Change: MySQL Cluster:
AUTO_INCREMENT columns had the following
problems when used in
AUTO_INCREMENT counter was not
updated correctly when such a column was updated.
AUTO_INCREMENT values were not
prefetched beyond statement boundaries.
AUTO_INCREMENT values were not handled
After being set,
showed a value of 1, regardless of the value it had
actually been set to.
As part of this fix, the behavior of
has changed. Setting this to less than 32 no longer has any
effect on prefetching within statements (where IDs are now
always obtained in batches of 32 or more), but only between
statements. The default value for this variable has also
changed, and is now
(Bug#25176, Bug#31956, Bug#32055)
Partitioning: Important Note:
An apostrophe or single quote character
') used in the
INDEX DIRECTORY, or
COMMENT for a
clause caused the server to crash. When used as part of a
CREATE TABLE statement, the crash
was immediate. When used in an
TABLE statement, the crash did not occur until trying
to perform a
SELECT or DML
statement on the table. In either case, the server could not be
completely restarted until the
corresponding to the newly created or altered table was deleted.
Upgrading to the current (or later) release solves this
problem only for tables that are newly created or altered.
Tables created or altered in previous versions of the server
' characters in
PARTITION options must still be removed by
deleting the corresponding
.frm files and
re-creating them afterwards.
RENAME DATABASE statement was removed and
db_name UPGRADE DATA DIRECTORY
RENAME DATABASE statement
was intended for upgrading database directory names to the
encoding format used in 5.1 for representing identifiers in the
file system (see Section 8.2.3, “Mapping of Identifiers to File Names”). However,
the statement was found to be dangerous because it could result
in loss of database contents. See
Section 12.1.32, “
RENAME DATABASE Syntax”, and
Section 12.1.1, “
ALTER DATABASE Syntax”.
(Bug#17565, Bug#21741, Bug#28360)
Replication: MySQL Cluster:
Row-based replication from or to a big-endian machine where the
table used the
NDB storage engine
failed, if the same table on the other machine was either
NDB or the other machine was
An improperly reset internal signal was observed as a hang when
using events in the
NDB API but
could result in various errors.
MySQL Cluster: Incorrectly handled parameters could lead to a crash in the Transaction Coordinator during a node failure, causing other data nodes to fail. (Bug#33168)
MySQL Cluster: A memory leak occurred if a subscription start request was received by the subscription manager before the node making the request was fully connected to the cluster. (Bug#32652)
MySQL Cluster: A local checkpoint could sometimes be started before the previous LCP was restorable from a global checkpoint. (Bug#32519)
MySQL Cluster: High numbers of API nodes on a slow or congested network could cause connection negotiation to time out prematurely, leading to the following issues:
Excessive CPU usage
Partially connected API nodes
When a mysqld acting as a cluster SQL node
engine, there is a delay during which some necessary data
structures cannot be initialized until after it has connected to
the cluster, and all MySQL Cluster tables should be opened as
read-only. This worked correctly when the
NDB binlog thread was running, but
when it was not running, Cluster tables were not opened as
read-only even when the data structures had not yet been set up.
MySQL Cluster: The failure of a master node could lead to subsequent failures in local checkpointing. (Bug#32160)
An error with an
if statement in
sql/ha_ndbcluster.cc could potentially lead
to an infinite loop in case of failure when working with
AUTO_INCREMENT columns in
NDB storage engine code was not
safe for strict-alias optimization in gcc
MySQL Cluster: It was possible in some cases for a node group to be “lost” due to missed local checkpoints following a system restart. (Bug#31525)
A query against a table with
BLOB columns that would return
more than a certain amount of data failed with Got
error 4350 'Transaction already aborted' from
This regression was introduced by Bug#29102.
NDB tables having names containing
nonalphanumeric characters (such as
$”) were not discovered
MySQL Cluster: A node failure during a local checkpoint could lead to a subsequent failure of the cluster during a system restart. (Bug#31257)
MySQL Cluster: A cluster restart could sometimes fail due to an issue with table IDs. (Bug#30975)
NDB tables did not
work correctly with
MySQL Cluster: A file system close operation could fail during a node or system restart. (Bug#30646)
MySQL Cluster: Transaction timeouts were not handled well in some circumstances, leading to excessive number of transactions being aborted unnecessarily. (Bug#30379)
MySQL Cluster: The cluster management client could not connect, and would hang instead. This issue affected Mac OS X 64-bit only. (Bug#30366)
MySQL Cluster: Attempting to restore a backup made on a cluster host using one endian to a machine using the other endian could cause the cluster to fail. (Bug#29674)
MySQL Cluster: Log event requests to ndb_mgmd could time out, causing it to fail. (Bug#29621)
MySQL Cluster: In some cases, the cluster managment server logged entries multiple times following a restart of mgmd. (Bug#29565)
--help did not
display any information about the
MySQL Cluster: An interpreted program of sufficient size and complexity could cause all cluster data nodes to shut down due to buffer overruns. (Bug#29390)
The error message for
code 275 (Out of transaction records for complete
phase) was missing.
See also Bug#31482.
A query using joins between several large tables and requiring
unique index lookups failed to complete, eventually returning
Uknown Error after a very long period of
time. This occurred due to inadequate handling of instances
where the Transaction Coordinator ran out of
TransactionBufferMemory, when the cluster
should have returned NDB error code 4012 (Request
MySQL Cluster: There was a short interval during the startup process prior to the beginning of heartbeat detection such that, were an API or management node to reboot or a network failure to occur, data nodes could not detect this, with the result that there could be a lingering connection. (Bug#28445)
The description of the
An attempt to perform a
SELECT ... FROM
INFORMATION_SCHEMA.TABLES whose result included
NDB tables for
which the user had no privileges crashed the MySQL Server on
which the query was performed.
UPDATE IGNORE could sometimes fail on
NDB tables due to the use of
unitialized data when checking for duplicate keys to be ignored.
MySQL Cluster: The cluster log was formatted inconsistently and contained extraneous newline characters. (Bug#25064)
MySQL Cluster: A restart of the cluster failed when more than 1 REDO phase was in use. (Bug#22696)
When inserting a row into an
table with a duplicate value for a nonprimary unique key, the
error issued would reference the wrong key.
This improves on an initial fix for this issue made in MySQL 5.1.13. (Bug#21072)
MySQL Cluster: An insufficiently descriptive and potentially misleading Error 4006 (Connect failure - out of connection objects...) was produced when either of the following two conditions occurred:
There were no more transaction records in the transaction coordinator
NDB object in the NDB API
was initialized with insufficient parallelism
Separate error messages are now generated for each of these two cases. (Bug#11313)
Partitioning: It was possible to partition a table to which a foreign key referred. (Bug#32948)
A query of the form
table GROUP BY (SELECT
table LIMIT 1);
table having a
SET column crashed the server.
An error in the internal function
mysql_unpack_partition() led to a fatal
error in subsequent calls to
Partitioning: Changing the storage engine used by a table having subpartitions led to a server crash. (Bug#31893)
ORDER BY ... DESC did not always work
correctly when selecting from partitioned tables.
See also Bug#31001.
This regression was introduced by Bug#27405.
ALTER TABLE ... COALESCE PARTITION on a table
[LINEAR] HASH or
[LINEAR] KEY caused the server to crash.
LIKE queries on tables partitioned by
KEY and using third-party storage engines
could return incomplete results.
Partitioning: It was not possible to insert the greatest possible value for a given data type into a partitioned table. For example, consider a table defined as shown here:
CREATE TABLE t (c BIGINT UNSIGNED) PARTITION BY RANGE(c) ( PARTITION p0 VALUES LESS THAN MAXVALUE );
The largest possible value for a
UNSIGNED column is 18446744073709551615, but the
INSERT INTO t VALUES
(18446744073709551615); would fail, even though the
same statement succeeded were
t not a
In other words,
MAXVALUE was treated as being
equal to the greatest possible value, rather than as a least
Cluster Replication: Replication: A node failure during replication could lead to buckets out of order; now active subscribers are checked for, rather than empty buckets. (Bug#31701)
When dropping a database containing a stored procedure while
using row-cased replication, the delete of the stored procedure
mysql.proc table was recorded in the
binary log following the
DATABASE statement. To correct this issue,
DROP DATABASE now uses
Replication: It was possible for the name of the relay log file to exceed the amount of memory reserved for it, possibly leading to a crash of the server. (Bug#31836)
See also Bug#28597.
Replication: Corruption of log events caused the server to crash on 64-bit Linux systems having 4 GB or more of memory. (Bug#31793)
Replication: Trying to replicate an update of a row that was missing on the slave led to a failure on the slave. (Bug#31702)
Use of the
@@hostname system variable in
not replicate. The workaround is to select its value into a user
variable (which does replicate) and insert that.
Replication: Table names were displayed as binary “garbage” characters in slave error messages. The issue was observed on 64-bit Windows but may have effected other platforms. (Bug#30854)
It was possible to set
such that the slave would jump into the middle of a transaction.
This fix improves on one made for this bug in MySQL 5.1.20; the
previous fix insured that the slave could not be made to jump
into the middle of an event group, but the slave failed to
statements could begin or end an event group.
See also Bug#12691.
See also Bug#31836.
This regression was introduced by Bug#20166.
Replication: If a temporary error occured inside an event group on an event that was not the first event of the group, the slave could get caught in an endless loop because the retry counter was reset whenever an event was executed successfully. (Bug#24860)
UPDATE statement using a
stored function that modified a nontransactional table was not
logged if it failed. This caused the copy of the
nontransactional table on the master have a row that the copy on
the slave did not.
In addition, when an
ON DUPLICATE KEY UPDATE statement encountered a
duplicate key constraint, but the
UPDATE did not actually change
any data, the statement was not logged. As a result of this fix,
such statements are now treated the same for logging purposes as
UPDATE statements, and so
are written to the binary log.
See also Bug#12713.
A replication slave sometimes failed to reconnect because it was
unable to run
SHOW SLAVE HOSTS.
It was not necessary to run this statement on slaves (since the
master should track connection IDs), and the execution of this
statement by slaves was removed.
Replication: A replication slave sometimes stopped for changes that were idempotent (that is, such changes should have been considered “safe”), even though it should have simply noted that the change was already done, and continued operation. (Bug#19958)
Replicating from a master table to a slave table where the size
VARCHAR column was a different
size would cause mysqld to crash. For more
information on replicating with different column definitions,
see Section 22.214.171.124, “Replication with Differing Table Definitions on Master and Slave”.
A replication slave could return “garbage” data
that was not in recognizable row format due to a problem with
See also Bug#18106.
Cluster Replication: Under certain conditions, the slave stopped processing relay logs. This resulted in the logs never being cleared and the slave eventually running out of disk space. (Bug#31958)
See also Bug#29549.
See also Bug#21494.
An issue with the
table could cause
autodiscovery to fail in certain rare circumstances.
A call to
mgmapi.cpp should have been a call to
Corrected a typecast involving
bool on Mac OS
X 10.5 (Leopard), which evaluated differently from earlier Mac
OS X versions.
Use of uninitialized memory for
filesort in a
subquery caused a server crash.
The correct data type for a
resulting from a
UNION could be
determined incorrectly in some cases: 1) Not correctly inferred
NULL depending on the number of selects;
2) Not inferred correctly as
NULL if one
select used a subquery.
For queries containing
, there was a
limitation that the
col_list ORDER BY
DISTINCT columns had to
be the same as
ORDER BY columns. Incorrect
results could be returned if this was not true.
A subquery using an
IS NULL check of a column
NOT NULL in a table used in the
FROM clause of the outer query produced an
An issue with the
sql_mode database can cause the
creation of stored routines to fail. If you are having problems
with creating stored routines while using this
sql_mode value, remove this
value from your
libmysqld to crash.
Use of a
BY expression in conjunction with
ROLLUP could cause a server crash.
See also Bug#31095.
SELECT INTO OUTFILE with 8-bit
ENCLOSED BY characters led to corrupted data
when the data was reloaded using LOAD DATA INFILE. This was
SELECT INTO OUTFILE failed to escape
the 8-bit characters.
The rules for valid column names were being applied differently for base tables and views. (Bug#32496)
A query of the form
table GROUP BY
alias WITH ROLLUP
ssl-cipher values in option files were not
being read by
Repeated execution of a query containing a
expression and numerous
OR relations could crash the server. The root
cause of the issue was determined to be that the internal
SEL_ARG structure was not properly
initialized when created.
ORDER BY query on a view created using a
FEDERATED table as a base table caused the
server to crash.
Comparison of a
BIGINT NOT NULL column with a
constant arithmetic expression that evaluated to NULL mistakenly
caused the error Column '...' cannot be
null (error 1048).
Some uses of user variables in a query could result in a server crash. (Bug#32260)
Memory corruption could occur due to large index map in
Range checked for each record status reported
SELECT. The problem was based in an incorrectly
calculated length of the buffer used to store a hexadecimal
representation of an index map, which could result in buffer
overrun and stack corruption under some circumstances.
Various test program cleanups were made: 1)
were removed. 2) bug25714 displays an error
message when invoked with incorrect arguments or the
--help option. 3)
mysql_client_test exits cleanly with a proper
The default grant tables on Windows contained information for
production.mysql.com, which should not
Under certain conditions, the presence of a
BY clause could cause an
clause to be ignored.
For comparisons of the form
the comparison is done using
DATETIME values, per the fix for
Bug#27590. However that fix caused any index on
date_col not to be used and
compromised performance. Now the index is used again.
Killing a statement could lead to a race condition in the server. (Bug#32148)
InnoDB does not support
SPATIAL indexes, but could crash when asked
to handle one. Now an error is returned.
On 64-bit platforms, assignments of values to enumeration-valued storage engine-specific system variables were not validated and could result in unexpected values. (Bug#32034)
Using dates in the range
'0000-00-99' range in the
WHERE clause could result in an incorrect
result set. (These dates are not in the supported range for
DATE, but different results for a
given query could occur depending on position of records
containing the dates within a table.)
User-defined functions are not loaded if the server is started
option, but the server did not properly handle this case and
issued an Out of memory error message
If a user-defined function was used in a
SELECT statement, and an error
occurred during UDF initialization, the error did not terminate
execution of the
rather was converted to a warning.
Changing the SQL mode to cause dates with “zero”
parts to be considered invalid (such as
'1000-00-00') could result in indexed and
nonindexed searches returning different results for a column
that contained such dates.
In debug builds, testing the result of an
NULL caused an assertion
mysql-test-run.pl sometimes set up test scenarios in which the same port number was passed to multiple servers, causing one of them to be unable to start. (Bug#31880)
The server crashed after insertion of a negative value into an
AUTO_INCREMENT column of an
ORDER BY led to the wrong result when
ARCHIVE on a table with a
BLOB when the table cache was
full. The table could also be reported as crashed after the
query had completed, even though the table data was intact.
Comparison results for
different from those for operators like
DATETIME-like values with
trailing extra characters such as
the values as
the other operators performed a binary-string comparison. Now
they all uniformly use a
comparison, but generate warnings for values with trailing
Name resolution for correlated subqueries and
HAVING clauses failed to distinguish which of
two was being performed when there was a reference to an outer
aliased field. This could result in error messages about a
HAVING clause for queries that had no such
For a fatal error during a filesort in
find_all_keys(), the error was returned
without the necessary handler uninitialization, causing an
SELECT ... INTO
OUTFILE, if the
ENCLOSED BY string
is empty and the
FIELDS TERMINATED BY string
started with a special character (one of
N), every occurrence
of the character within field values would be duplicated.
Queries that include a comparison of an
INFORMATION_SCHEMA table column to
NULL caused a server crash.
ucs2 does not work as a client character set,
but attempts to use it as such were not rejected. Now
character_set_client cannot be
ucs2. This also affects statements
SET NAMES and
A buffer used when setting variables was not dimensioned to
accommodate the trailing
'\0' byte, so a
single-byte buffer overrun was possible.
Spurious duplicate-key errors could occur for multiple-row
inserts into an
InnoDB table that activate a
InnoDB now tracks locking and use of tables
by MySQL only after a table has been successfully locked on
behalf of a transaction. Previously, the locked flag was set and
the table in-use counter was updated before checking whether the
lock on the table succeeded. A subsequent failure in obtaining a
lock on the table led to an inconsistent state as the table was
neither locked nor in use.
SELECT 1 REGEX NULL caused an assertion
failure for debug servers.
NDB libraries and include files were missing from some binary tar file distributions. (Bug#31414)
mysql-test-run.pl tried to create files in a
directory where it could not be expected to have write
permission. mysqltest created
.reject files in a directory other than the
one where test results go.
Automatically allocated memory for string options associated with a plugin was not freed if the plugin did not get installed. (Bug#31382)
InnoDB tables with
READ COMMITTED isolation
level, semi-consistent reads used for
UPDATE statements skipped rows
locked by another transaction, rather than waiting for the locks
to be released. Consequently, rows that possibly should have
been updated were never examined.
For an almost-full
MyISAM table, an insert
that failed could leave the table in a corrupt state.
Allocation of an insufficiently large group-by buffer following creation of a temporary table could lead to a server crash. (Bug#31249)
TIMESTAMP with a nonzero
time part was converted to a
value, no warning was generated. This caused index lookups to
assume that this is a valid conversion and was returning rows
that match a comparison between a
TIMESTAMP value and a
DATE keypart. Now a warning is
generated so that
a nonzero time part will not match
Server variables could not be set to their current values on Linux platforms. (Bug#31177)
See also Bug#6958.
An assertion designed to detect a bug in the
ROLLUP implementation would incorrectly be
triggered when used in a subquery context with noncacheable
The server crashed in the parser when running out of memory. Memory handling in the parser has been improved to gracefully return an error when out-of-memory conditions occur in the parser. (Bug#31153)
MySQL declares a
UNIQUE key as a
PRIMARY key if it doesn't have
NULL columns and is not a partial key, and
PRIMARY key must alway be the first key.
However, in some cases, a nonfirst key could be reported as
PRIMARY, leading to an assert failure by
InnoDB. This is fixed by correcting the key
Starting the server using
--read-only and with the Event
Scheduler enabled caused it to crash.
This issue occurred only when the server had been built with certain nonstandard combinations of configure options.
GROUP BY NULL WITH ROLLUP could cause a
See also Bug#32558.
A rule to prefer
filesort over an indexed
ORDER BY when accessing all rows of a table
was being used even if a
LIMIT clause was
REGEXP operations could cause a
server crash for character sets such as
Now the arguments are converted to
possible, to allow correct results to be produced if the
resulting strings contain only 8-bit characters.
Expressions of the form
, where the same
column was named both times, could cause a server crash in the
col NOT IN
Many nested subqueries in a single query could led to excessive memory consumption and possibly a crash of the server. (Bug#31048)
ORDER BY with
ARCHIVE tables caused a server crash.
A server crash could occur when a
DETERMINISTIC stored function was used in
GROUP BY clause.
On Windows, the
implementation was incorrect.
A character set introducer followed by a hexadecimal or bit-value literal did not check its argument and could return an ill-formed result for invalid input. (Bug#30986)
MEMORY tables, lookups for
NULL values in
indexes could return incorrect results.
For a spatial column with a regular
SPATIAL) index, queries failed if the
optimizer tried to use the index.
Values for the
incorrectly were treated as values for the
The optimizer incorrectly optimized conditions out of the
WHERE clause in some queries involving
subqueries and indexed columns.
On Windows, the
implementation was incorrect. One symptom was that invalidating
the query cache could cause a server crash.
User-supplied names foreign key names might not be set to the right key, leading to foreign keys with no name. (Bug#30747)
The embedded server did not properly check column-level privileges. (Bug#30710)
incorrect for views that were defined to select from other
The optimizer could ignore
ORDER BY in cases
when the result set is ordered by
resulting in rows being returned in incorrect order.
A different execution plan was displayed for
EXPLAIN than would actually have
been used for the
the test of sort keys for
ORDER BY did not
consider keys mentioned in
IGNORE KEYS FOR ORDER
LIMIT arguments greater than
232 did not work correctly.
MyISAM tables could not exceed 4294967295
(232 – 1) rows on Windows.
HANDLER ... READ operation could
leave the table in a locked state.
The optimization that uses a unique index to remove
GROUP BY did not ensure that the index was
actually used, thus violating the
that is implied by
SHOW STATUS LIKE 'Ssl_cipher_list' from a
MySQL client connected via SSL returned an empty string rather
than a list of available ciphers.
GROUP BY on an expression of the form
caused a server
crash due to incorrect calculation of number of decimals.
The options available to the
TABLE statement were also allowed in
OPTIMIZE TABLE and
ANALYZE TABLE statements, but
caused corruption during their execution. These options were
never supported for these statements, and an error is now raised
if you try to apply these options to these statements.
A self-referencing trigger on a partitioned table caused the server to crash instead of failing with an error. (Bug#30484)
When expanding a
* in a
NATURAL join, the
check for table access for both tables in the join was done
using only the grant information of the first table.
When casting a string value to an integer, cases where the input
string contained a decimal point and was long enough to overrun
unsigned long long type were not handled
correctly. The position of the decimal point was not taken into
account which resulted in miscalculated numbers and incorrect
truncation to appropriate SQL data type limits.
Versions of mysqldump from MySQL 4.1 or
higher tried to use
START TRANSACTION WITH CONSISTENT
SNAPSHOT if the
--master-data options were
given, even with servers older than 4.1 that do not support
libmysqld, use of prepared statements
and the query cache at the same time caused problems.
variable value can be computed accurately only for simple
“flat” queries, not complex queries such as those
with subqueries or
However, the value was not consistently being set to 0 for
The optimizer made incorrect assumptions about the value of the
is_member value for user-defined functions,
sometimes resulting in incorrect ordering of UDF results.
euc-kr characters having the
second byte in the ranges
[0x61..0x7A] were rejected.
When loading a dynamic plugin on FreeBSD, the plugin would fail to load. This was due to a build error where the required symbols would be not exported correctly. (Bug#30296)
Setting certain values on a table using a spatial index could cause the server to crash. (Bug#30286)
Tables with a
GEOMETRY column could be marked
as corrupt if you added a non-
Flushing a merge table between the time it was opened and its child table were actually attached caused the server to crash. (Bug#30273)
This regression was introduced by Bug#26379.
The query cache does not support retrieval of statements for which column level access control applies, but the server was still caching such statements, thus wasting memory. (Bug#30269)
Short-format mysql commands embedded within
/*! ... */ comments were parsed incorrectly
by mysql, which discarded the rest of the
comment including the terminating
characters. The result was a malformed (unclosed) comment. Now
mysql does not discard the
If the server crashed during an
TABLE statement, leaving a temporary file in the
database directory, a subsequent
DATABASE statement failed due to the presence of the
When mysqldump wrote
DROP DATABASE statements within
version-specific comments, it included the terminating semicolon
in the wrong place, causing following statements to fail when
the dump file was reloaded.
It was not possible for client applications to distinguish
between auto-set and auto-updated
TIMESTAMP column values.
To rectify this problem, a new
ON_UPDATE_NOW_FLAG flag is set by
Field_timestamp constructors whenever a column should be set to
now reports whether a timestamp column is set to
In addition, such columns now display
CURRENT_TIMESTAMP in the
column in the output from
--without-geometry option for
configure caused server compilation to fail.
See also Bug#26162.
With auto-reconnect enabled, row fetching for a prepared statement could crash after reconnect occurred because loss of the statement handler was not accounted for. (Bug#29948)
configure did not find
on some Linux platforms.
It was possible when creating a partitioned table using
CREATE TABLE ...
SELECT to refer in the
clause to columns in the table being selected from, which could
cause the server to crash. An example of such a statement is:
CREATE TABLE t1 (b INT) PARTITION BY RANGE(t2.b) ( PARTITION p1 VALUES LESS THAN (10), PARTITION p2 VALUES LESS THAN (20) ) SELECT * FROM t2;
The fix is to disallow references in
BY clauses to columns not in the table being created.
The mysql client program now ignores Unicode byte order mark (BOM) characters at the beginning of input files. Previously, it read them and sent them to the server, resulting in a syntax error.
variables were displayed by
VARIABLES but could not be accessed in expressions as
@@log_slow_queries. Also, attempting to set
SET produced an incorrect
Unknown system variable message. Now these
variables are treated as synonyms for
slow_query_log, which means
that they can be accessed in expressions and their values can be
Denormalized double-precision numbers cannot be handled properly by old MIPS pocessors. For IRIX, this is now handled by enabling a mode to use a software workaround. (Bug#29085)
When doing a
DELETE on a table
that involved a
MERGE tables and
JOIN referred to the same table, the
operation could fail reporting
ERROR 1030 (HY000): Got
error 134 from storage engine. This was because scans
on the table contents would change because of rows that had
already been deleted.
Killing an SSL connection on platforms where MySQL is compiled
-DSIGNAL_WITH_VIO_CLOSE (Windows, Mac OS
X, and some others) could crash the server.
Index hints specified in view definitions were ignored when using the view to select from the base table. (Bug#28702)
Views do not have indexes, so index hints do not apply. Use of index hints when selecting from a view is now disallowed. (Bug#28701)
After changing the SQL mode to a restrictive value that would make already-inserted dates in a column be considered invalid, searches returned different results depending on whether the column was indexed. (Bug#28687)
When running the MySQL Instance Configuration Wizard, a race condition could exist that would fail to connect to a newly configured instance. This was because mysqld had not completed the startup process before the next stage of the installation process. (Bug#28628)
The parser confused user-defined function (UDF) and stored
function creation for
FUNCTION and required that there be a default database
when creating UDFs, although there is no such requirement.
Fast-mutex locking was not thread-safe and optimization-safe on some platforms, which could cause program failures such as out-of-memory errors. (Bug#28284)
The metadata in some
could be incorrect when a temporary table was used to evaluate a
The anonymous accounts were not being created during MySQL installation. (Bug#27692)
ALTER TABLE did
not cause the table to be rebuilt.
A race condition between killing a statement and the thread executing the statement could lead to a situation such that the binary log contained an event indicating that the statement was killed, whereas the statement actually executed to completion. (Bug#27571)
Some character mappings in the
file were incorrect.
As a result of this bug fix, indexes must be rebuilt for columns
that use the
ascii_general_ci collation for
columns that contain any of these characters:
Section 2.4.3, “Checking Whether Tables or Indexes Must Be Rebuilt”.
Some queries using the
NAME_CONST() function failed to
return either a result or an error to the client, causing it to
hang. This was due to the fact that there was no check to insure
that both arguments to this function were constant expressions.
MySQL failed to generate or retrieve an
AUTO_INCREMENT primary key for
InnoDB tables with user-defined partitioning.
UpdateXML() functions performed
extremely slowly for large amounts of XML data (greater than 64
KB). These functions now execute approximately 2000 times faster
On Windows, writes to the debug log were using
freopen() instead of
fflush(), resulting in slower performance.
Symbolic links on Windows could fail to work. (Bug#26811)
mysqld sometimes miscalculated the number of
digits required when storing a floating-point number in a
CHAR column. This caused the
value to be truncated, or (when using a debug build) caused the
server to crash.
See also Bug#12860.
It makes no sense to attempt to use
ALTER TABLE ...
ORDER BY to order an
if there is a user-defined clustered index, because rows are
always ordered by the clustered index. Such attempts now are
ignored and produce a warning.
Also, in some cases,
InnoDB incorrectly used
a secondary index when the clustered index would produce a
EXPLAIN output now
indicates use of the clustered index (for tables that have one)
as lines with a
type value of
key value of
Using index in the
See also Bug#35850.
For a prepared statement
changing the default database following
stmt to be recorded
incorrectly in the binary log.
CREATE TABLE LIKE did not work when the
source table was an
Threads that were calculating the estimated number of records
for a range scan did not respond to the
KILL statement. That is, if a
range join type is possible
(even if not selected by the optimizer as a join type of choice
and thus not shown by
the query in the
statistics state (shown by
SHOW PROCESSLIST) did not
respond to the
CREATE TABLE a AS
SELECT * FROM A would fail.
returns column of the
mysql.proc table was
CHAR(64), which is not long enough to store
long data types such as
types. The column has been changed to
LONGBLOB and a warning is
generated if truncation occurs when storing a row into the
If the expected precision of an arithmetic expression exceeded the maximum precision supported by MySQL, the precision of the result was reduced by an unpredictable or arbitrary amount, rather than to the maximum precision. In some cases, exceeding the maximum supported precision could also lead to a crash of the server. (Bug#24907)
For Vista installs, MySQLInstanceConfig.exe did not add the default MySQL port to the firewall exceptions. It now provides a checkbox that enables the user a choice of whether to do this. (Bug#24853)
For storage engines that do not redefine
handler::index_next_same() and are capable
of indexes, statements that include a
clause might select incorrect data.
Entries in the general query log were truncated at 1000 characters. (Bug#21557)
Worked around an icc problem with an incorrect machine instruction being generated in the context of software pre-fetching after a subroutine got in-lined. (Upgrading to icc 10.0.026 makes the workaround unnecessary.) (Bug#20803)
See also Bug#40363.
-lmtmalloc library was removed from the
output of mysql_config on Solaris, as it
caused problems when building
possibly other applications) on that platform that tried to use
dlopen() to access the client library.
MySQLInstanceConfig.exe failed to grant
certain privileges to the
Clients were ignoring the TCP/IP port number specified as the default port via the --with-tcp-port configuration option. (Bug#15327)
Zero-padding of exponent values was not the same across platforms. (Bug#12860)
Values of types
ZEROFILL, were not zero-filled when converted to a
character representation in the C prepared statement API.
mysql stripped comments from statements sent
to the server. Now the
--skip-comments option can be
used to control whether to retain or strip comments. The default
Several buffer-size system variables were either being handled incorrectly for large values (for settings larger than 4GB, they were truncated to values less than 4GB without a warning), or were limited unnecessarily to 4GB even on 64-bit systems. The following changes were made:
DISABLE KEYS and
KEYS on a nonempty table would cause the size of the
index file for the table to grow considerable. This was because
DISABLE KEYS operation would only mark
the existing index, without deleting the index blocks. The
ENABLE KEYS operation would re-create the
index, adding new blocks, while the previous index blocks would
remain. Existing indexes are now dropped and recreated when the
ENABLE KEYS statement is executed.
Grant table checks failed in