Functionality Added or Changed
MySQL now supports pluggable authentication, such that the server uses plugins to authenticate incoming client connections. Client programs load an authentication plugin that interacts properly with the corresponding server plugin.
Pluggable authentication enables two important capabilities, external authentication and proxy users:
Pluggable authentication makes it possible for clients to connect to the MySQL server with credentials that are appropriate for authentication methods other than native authentication based on passwords stored in the
mysql.user table. For example, plugins can be created to use external authentication methods such as PAM, Windows login IDs, LDAP, or Kerberos.
If a user is permitted to connect, an authentication plugin can return to the server a user name different from the name of the connecting user, to indicate that the connecting user is a proxy for another user. While the connection lasts, the proxy user is treated, for purposes of access control, as having the privileges of a different user. In effect, one user impersonates another.
Pluggable authentication entails these changes:
For user specifications in the
CREATE USER and
GRANT statements, there is a new
IDENTIFIED WITH clause for specifying the authentication plugin.
mysql.user table, there are new columns that specify plugin information. The
plugin column, if nonempty, indicates which plugin authenticates connections for an account. The
authentication_string column is a string that the server passes to the plugin for connections by clients that authenticate using the plugin.
mysql_options() C API function, there are new
MYSQL_PLUGIN_DIR options that enable client programs to load authentication plugins.
For the mysql client, there are new
--plugin-dir options for specifying which authentication plugin and plugin directory to use. These options will be added to other clients in future releases.
For the mysqltest client, there is a new
--plugin-dir option for specifying which plugin directory to use, and a new
connect() command argument to specify an authentication plugin.
For the server plugin API, there is a new
MYSQL_AUTHENTICATION_PLUGIN plugin type.
A new client plugin API enables client programs to manage plugins.
The native authentication methods previously supported in MySQL have been reimplemented as plugins. These methods provide against the current password format and pre-MySQL 4.1.1 format that uses shorter password hash values. This change reimplements the native methods as plugins that cannot be unloaded. Existing clients authenticate as before with no changes needed. In particular, starting the server with the
--secure-auth option still prevents clients that have pre-4.1.1 password hashes from connecting, and
--skip-grant-tables still disables all password checking.
Proxy user capability entails these changes:
There is a new
PROXY privilege that can be managed with the
external_user system variables indicate whether the current session uses proxying.
mysql.proxies_priv grant table records proxy information for MySQL accounts.
Due to these changes, the server requires that a new grant table,
proxies_priv, be present in the
mysql database. If you are upgrading to MySQL 5.5.7 from a previous MySQL release rather than performing a new installation, the server will find that this table is missing and exit during startup with the following message:
Table 'mysql.proxies_priv' doesn't exist
To create the
proxies_priv table, start the server with the
--skip-grant-tables option to cause it to skip the normal grant table checks, then run mysql_upgrade. For example:
mysqld --skip-grant-tables &shell>
Then stop the server and restart it normally.
You can specify other options on the mysqld command line if necessary. Alternatively, if your installation is configured so that the server normally reads options from an option file, use the
--defaults-file option to specify the file (enter each command on a single line):
--skip-grant-tables option, the server does no password or privilege checking, so any client can connect and effectively have all privileges. For additional security, use the
--skip-networking option as well to prevent remote clients from connecting.
The upgrade problem just described is fixed in MySQL 5.5.8. The server treats a missing
proxies_priv table as equivalent to an empty table.
For additional information, consult these references:
Information about pluggable authentication, including installation and usage instructions: Pluggable Authentication.
Information about proxy users: Proxy Users.
Information about the server and client plugin API: Writing Plugins.
Information about the C API functions for managing client plugins: See C API Client Plugin Functions.
Information about current restrictions on the use of pluggable authentication, including which connectors support which plugins: See Restrictions on Pluggable Authentication. Third-party connector developers should read that section to determine the extent to which a connector can take advantage of pluggable authentication capabilities and what steps to take to become more compliant.
MySQL releases now are built using CMake rather than the GNU autotools. Accordingly, the instructions for installing MySQL from source have been updated to discuss how to build MySQL using CMake. See Installing MySQL from Source. If you are familiar with autotools but not CMake, you might find these transition instructions helpful: Autotools to CMake Transition Guide
The build process is now similar enough on all platforms, including Windows, that there are no longer sections dedicated to notes for specific platforms.
The default installation layout when compiling from source now matches that used for binary distributions. You will notice these differences for installations from source distributions:
mysqld is installed in
mysql_install_db is installed in
The data directory is
The make_binary_distribution and
make_win_bin_dist scripts are now obsolete. To create a binary distribution, use make package.
Functionality Added or Changed
Incompatible Change: Previously, if you flushed the logs using
FLUSH LOGS or mysqladmin flush-logs and mysqld was writing the error log to a file (for example, if it was started with the
--log-error option), it renamed the current log file with the suffix
-old, then created a new empty log file. This had the problem that a second log-flushing operation thus caused the original error log file to be lost unless you saved it under a different name. For example, you could use the following commands to save the file:
To avoid the preceding file-loss problem, renaming no longer occurs. The server merely closes and reopens the log file. To rename the file, you can do so manually before flushing. Then flushing the logs reopens a new file with the original file name. For example, you can rename the file and create a new one using the following commands:
References: See also Bug #56821.
The unused and undocumented
thread_pool_size system variable was removed. (Bug #57338)
pstack library was nonfunctional and has been removed, along with the
--with-pstack option for configure and the
--enable-pstack option for mysqld. (Bug #57210)
Added a new
SHOW PROCESSLIST state,
Waiting for query cache lock. This indicates that a session is waiting to take the query cache lock while it performs some query cache operation. (Bug #56822)
A new status variable,
Handler_read_last, displays the number of requests to read the last key in an index. With
ORDER BY, the server issues a first-key request followed by several next-key requests, whereas with
ORDER BY DESC, the server issues a last-key request followed by several previous-key requests. (Bug #52312)
Previously, the server supported values of
FORCE for the
-- option format for controlling plugin loading using an option named after the plugin. Such options now support a
FORCE_PLUS_PERMANENT value. This value is like
FORCE, but in addition prevents the plugin from being unloaded at runtime. If a user attempts to do so with
UNINSTALL PLUGIN, an error occurs. See Installing and Uninstalling Plugins.
In addition, the
INFORMATION_SCHEMA.PLUGINS table now has a
LOAD_OPTION column that indicates the plugin loading value (
FORCE_PLUS_PERMANENT). See The INFORMATION_SCHEMA PLUGINS Table.
Security Fix; Incompatible Change; InnoDB: Issuing
TRUNCATE TABLE and examining the same table's information in the
INFORMATION_SCHEMA database at the same time could cause a crash in the debug version of the server.
As a result of this change,
InnoDB always uses the fast truncation technique, equivalent to
DROP TABLE and
CREATE TABLE. It no longer performs a row-by-row delete for tables with parent-child foreign key relationships.
TRUNCATE TABLE returns an error for such tables. Modify your SQL to issue
DELETE FROM for such tables instead. (Bug #54678)
Security Fix: The server crashed for assignment of values of types other than
Geometry to items of type
MultiSurface). Now the server checks the value type and fails with
bad geometry value if it detects incorrect parameters. (Bug #55531)
Security Fix: The
CONVERT_TZ() function crashed the server when the timezone argument was an empty
SET column value. (Bug #55424)
EXPLAIN EXTENDED caused a server crash with some prepared statements. (Bug #54494)
Security Fix: In prepared-statement mode,
EXPLAIN for a
SELECT from a derived table caused a server crash. (Bug #54488)
Security Fix: The
PolyFromWKB() function could crash the server when improper WKB data was passed to the function. (Bug #51875, Bug #11759554, CVE-2010-3840)
Performance; InnoDB: The master
InnoDB background thread could sometimes cause transient performance drops due to excessive flushing of modified pages. (Bug #56933)
Incompatible Change; Replication: The behavior of
INSERT DELAYED statements when using statement-based replication has changed as follows:
Previously, when using
binlog_format=STATEMENT, a warning was issued in the client when executing
INSERT DELAYED; now, no warning is issued in such cases.
Previously, when using
INSERT DELAYED was logged as
INSERT DELAYED; now, it is logged as an
INSERT, without the
INSERT DELAYED continues to be executed as
INSERT (without the
DELAYED option). The behavior of
INSERT DELAYED remains unchanged when using
INSERT DELAYED generates no warnings, is executed as
INSERT DELAYED, and is logged using the row-based format.
This change also affects
INSERT DELAYED is no longer considered unsafe. Now, when the logging format is
MIXED, no switch to row-based logging occurs. This means that the statement is logged as a simple
INSERT (that is, without the
DELAYED option), using the statement-based logging format. (Bug #54579, Bug #11762035)
References: See also Bug #56678, Bug #11763907, Bug #57666. This bug was introduced by Bug #39934, Bug #11749859.
HANDLER ... READ statements that invoke stored functions can cause replication errors. Such statements are now disallowed and result in an
ER_NOT_SUPPORTED_YET error. (Bug #54920)
Important Change; InnoDB: The server could crash with an assertion, possibly leading to data corruption, while updating the primary key of an
InnoDB table containing
BLOB or other columns requiring off-page storage. This fix applies to the
InnoDB Plugin in MySQL 5.1, and to
InnoDB 1.1 in MySQL 5.5. (Bug #55543)
InnoDB; Replication: If the master had
innodb_strict_mode=OFF), or both, certain
CREATE TABLE options, such as
KEY_BLOCK_SIZE, were ignored. This could permit the master to avoid raising
However, the ignored
CREATE TABLE options were still written into the binary log, so that, if the slave had
innodb_file_format=Barracuda, it could encounter an
ER_TOO_BIG_ROWSIZE error while executing the record from the log, causing the slave SQL thread to abort and replication to fail.
In the case where the master was running MySQL 5.1 and the slave was MySQL 5.5 (or later), the failure occurred when both master and slave were running with default values for
innodb_file_format. This could cause problems during upgrades.
To address this issue, the default values for
innodb_file_format are reverted to the MySQL 5.1 default values—that is,
Antelope, respectively. (Bug #56318, Bug #11763590)
InnoDB: The server could crash with a high volume of concurrent
LOCK TABLES and
UNLOCK TABLES statements. (Bug #57345)
InnoDB incorrectly reported an error when a cascading foreign key constraint deleted more than 250 rows. (Bug #57255)
InnoDB: If the server crashed during an
ALTER TABLE operation on an
InnoDB table, examining the table through
SHOW CREATE TABLE or querying the
INFORMATION_SCHEMA tables could cause the server to stop with an assertion error. (Bug #56982)
InnoDB: The output from the
SHOW ENGINE INNODB STATUS command can now be up to 1MB. Formerly, it was truncated at 64KB. Monitoring applications that parse that output can check whether it exceeds this new, larger limit by testing the
Innodb_truncated_status_writes status variable. (Bug #56922)
InnoDB: For debug builds, a
SELECT ... FOR UPDATE statement affecting a range of rows in an
InnoDB table could cause a server crash. (Bug #56716)
InnoDB: Improved the performance of
UPDATE operations on
InnoDB tables, when only non-indexed columns are changed. (Bug #56340)
InnoDB: When MySQL was restarted after a crash with the option
innodb_force_recovery=6, certain queries against
InnoDB tables could fail, depending on
ORDER BY clauses.
Usually in such a disaster recovery situation, you dump the entire table using a query without these clauses. During advanced troubleshooting, you might use queries with these clauses to diagnose the position of the corrupted data, or to recover data following the corrupted part. (Bug #55832)
CHECK TABLE command could cause a time-consuming verification of the
InnoDB adaptive hash index memory structure. Now this extra checking is only performed in binaries built for debugging. (Bug #55716)
InnoDB: A heavy workload with a large number of threads could cause a crash in the debug version of the server. (Bug #55699)
InnoDB: The server could crash on shutdown, if started with
--innodb-use-system-malloc=0. (Bug #55627)
InnoDB: If the server crashed during a
RENAME TABLE operation on an
InnoDB table, subsequent crash recovery could fail. This problem could also affect an
ALTER TABLE statement that caused a rename operation internally. (Bug #55027)
InnoDB: Setting the
PACK_KEYS=0 table option for an
InnoDB table prevented new indexes from being added to the table. (Bug #54606)
InnoDB: The server could crash when opening an
InnoDB table linked through foreign keys to a long chain of child tables. (Bug #54582, Bug #11762038)
InnoDB: Changed the locking mechanism for the
InnoDB data dictionary during
ROLLBACK operations, to improve concurrency for
REPLACE statements. (Bug #54538)
InnoDB: With multiple buffer pools enabled,
InnoDB could flush more data from the buffer pool than necessary, causing extra I/O overhead. (Bug #54346)
InnoDB transactions could be incorrectly committed during recovery, rather than rolled back, if the server crashed and was restarted after performing
ALTER TABLE ... ADD PRIMARY KEY on an
InnoDB table, or some other operation that involves copying the entire table. (Bug #53756)
InnoDB startup messages now include the start and end times for buffer pool initialization, and the total buffer pool size. (Bug #48026)
ALTER TABLE statement acting on table partitions that failed while the affected table was locked could cause the server to crash. (Bug #56172)
UPDATE statements involving a partitioned
MyISAM table could cause this table to become corrupted. Not all tables affected by the
UPDATE needed to be partitioned for this issue to be observed. (Bug #55458)
EXPLAIN PARTITIONS returned bad estimates for range queries on partitioned
MyISAM tables. In addition, values in the
rows column of
EXPLAIN PARTITIONS output did not take partition pruning into account. (Bug #53806, Bug #46754)
SET PASSWORD caused failure of row-based replication between a MySQL 5.1 master and a MySQL 5.5 slave.
This fix makes it possible to replicate
SET PASSWORD correctly, using row-based replication between a master running MySQL 5.1.53 or a later MySQL 5.1 release to a slave running MySQL 5.5.7 or a later MySQL 5.5 release. (Bug #57098)
References: See also Bug #55452, Bug #57357.
Replication: Prepared multiple-row
INSERT DELAYED statements were written to the binary log without
DELAYED. (Bug #56678, Bug #11763907)
References: This bug is a regression of Bug #54579, Bug #11762035.
Replication: Backticks used to enclose identifiers for savepoints were not preserved in the binary log, which could lead to replication failure when the identifier, stripped of backticks, could be misinterpreted, causing a syntax or other error.
This could cause problems with MySQL application programs making use of generated savepoint IDs. If, for instance,
java.sql.Connection.setSavepoint() is called without any parameters, Connector/J automatically generates a savepoint identifier consisting of a string of hexadecimal digits
F encased in backtick (
`) characters. If such an ID took the form
N represents a string of the decimal digits
e is a literal uppercase or lowercase “E” character). Removing the backticks when writing the identifier into the binary log left behind a substring which the slave MySQL server tried to interpret as a floating point number, rather than as an identifier. The resulting syntax error caused loss of replication. (Bug #55961)
References: See also Bug #55962.
Replication: When a slave tried to execute a transaction larger than the slave's value for
max_binlog_cache_size, it crashed. This was caused by an assertion that the server should roll back only the statement but not the entire transaction when the error
ER_TRANS_CACHE_FULL occurred. However, the slave SQL thread always rolled back the entire transaction whenever any error occurred, regardless of the type of error. (Bug #55375)
Replication: The error message for
ER_SLAVE_HEARTBEAT_VALUE_OUT_OF_RANGE was hard coded in English in
sql_yacc.yy, so that it could not be translated in
errmsg.txt for other languages.
Additionally, this same error message was used for three separate error conditions:
When the heartbeat period exceeded the value of
When the heartbeat period was nonnegative but shorter than 1 millisecond.
When the value for the heartbeat period was either negative or greater than the maximum permitted.
These issues have been addressed as follows:
By using three distinct error messages for each of the conditions listed previously.
By moving the sources for these error messages into the
errmsg-utf8.txt file to facilitate translations into languages other than English.
mysqld segfaulted if compiled with gcc 4.6. (Bug #61509, Bug #14548064)
A buffer overrun could occur when formatting
DBL_MAX numbers. (Bug #57209)
COALESCE() in MySQL 5.5 could return a result different from MySQL 5.1 for some arguments. (Bug #57095)
SUBTIME() expressions could return incorrect results. (Bug #57039)
When mysqld was started as a service on Windows and mysqld was writing the error log to a file (for example, if it was started with the
--log-error option), the server reassigned the file descriptors of the
stderr streams to the file descriptor of the log file. On Windows, if
stderr is not associated with an output stream, the file descriptor returns a negative value. Previously, this caused the file descriptor reassignment to fail and the server to abort. To avoid this problem on Windows, the server now first assigns the
stderr streams to the log file stream by opening this file. This causes the
stderr file descriptors to be nonzero and the server can successfully reassign them to the file descriptor of the log file. (Bug #56821)
References: This bug is a regression of Bug #29751.
The server could crash inside
memcpy() when reading certain Performance Schema tables. (Bug #56761, Bug #58003)
Deadlock could occur for heavily concurrent workloads consisting of a mix of DML, DDL, and
FLUSH TABLES statements affecting the same set of tables. (Bug #56715, Bug #56404, Bug #56405)
Memory leaks detected by Valgrind were corrected. (Bug #56709)
On Mac OS X,
RENAME TABLE raised an assertion if the
lower_case_table_names system variable was 2 and the old table name was specified in uppercase. (Bug #56595)
Performance for certain read-only queries, in particular
point_select, had deteriorated compared to previous versions. (Bug #56585)
It was possible to compile mysqld with Performance Schema support but with a dummy atomic-operations implementation, which caused a server crash. This problem does not affect binary distributions. It is helpful as a safety measure for users who build MySQL from source. (Bug #56521)
XA END after an XA transaction was already ended raised an assertion. (Bug #56448)
SELECT statement could produce a number of rows different from a
CREATE TABLE ... SELECT that was supposed to select the same rows. (Bug #56423)
References: This bug is a regression of Bug #38999.
The server crashed if a table maintenance statement such as
ANALYZE TABLE or
REPAIR TABLE was executed on a
MERGE table and opening and locking a child table failed. For example, this could happen if a child table did not exist or if a lock timeout happened while waiting for a conflicting metadata lock to disappear.
As a consequence of this bug fix, it is now possible to use
CHECK TABLE for log tables without producing an error. (Bug #56422, Bug #56494)
ALTER TABLE on a
MERGE table could result in deadlock with other connections. (Bug #56292, Bug #57002)
Comparison of one
STR_TO_DATE() result with another could return incorrect results. (Bug #56271)
tcmalloc library was missing from binary MySQL packages for Linux. (Bug #56267)
INSERT DELAYED statement for a
MERGE table could cause deadlock if it occurred as part of a transaction or under
LOCK TABLES, and there was a concurrent DDL or
LOCK TABLES ... WRITE statement that tried to lock one of its underlying tables. (Bug #56251)
In debug builds, the server raised an assertion for
DROP DATABASE in installations that had an outdated or corrupted
mysql.proc table. For example, this affected mysql_upgrade when run as part of a MySQL 5.1 to 5.5 upgrade. (Bug #56137)
TIME argument to
MAX() could raise an assertion. (Bug #56120)
The ordering for supplementary characters in the
utf32_bin collations was incorrect. (Bug #55980)
On Solaris with gcc 3.4.6,
ha_example.so was built with DTrace support even if the server was not, causing plugin loading problems. (Bug #55966)
Short (single-letter) command-line options did not work. (Bug #55873)
If a query specified a
DATETIME value in a format different from
'YYYY-MM-DD HH:MM:SS', a greater-than-or-equal (
>=) condition matched only greater-than values in an indexed
TIMESTAMP column. (Bug #55779, Bug #50774, Bug #11758558)
If a view was named as the destination table for
CREATE TABLE ... SELECT, the server produced a warning whether or not
IF NOT EXISTS was used. Now it produces a warning only when
IF NOT EXISTS is used, and an error otherwise. (Bug #55777)
CASE expressions with a mix of operands in different character sets sometimes returned incorrect results. (Bug #55744)
After the fix for Bug #39653, the shortest available secondary index was used for full table scans. The primary clustered key was used only if no secondary index could be used. However, when the chosen secondary index includes all columns of the table being scanned, it is better to use the primary index because the amount of data to scan is the same but the primary index is clustered. This is now taken into account. (Bug #55656)
The server entered an infinite loop with high CPU utilization after an error occurred during flushing of the I/O cache. (Bug #55629)
For the Performance Schema, the default number of rwlock classes was increased to 30, and the default number of rwlock and mutex instances was increased to 1 million. These changes were made to account for the volume of data instrumented when the
InnoDB storage engine is used (because of the
InnoDB buffer pool). (Bug #55576)
If there was an active
SELECT statement, an error arising during trigger execution could cause a server crash. (Bug #55421)
InnoDB scalar subquery results to a variable resulted in unexpected
S locks in
READ COMMITTED transaction isolation level. (Bug #55382)
In debug builds,
FLUSH TABLE for a
table_list WITH READ LOCK
MERGE table led to an assertion failure if one of the table's children was not present in the list of tables to be flushed. (Bug #55273)
The server could crash during shutdown due to a race condition relating to Performance Schema cleanup. (Bug #55105, Bug #56324)
Queries involving predicates of the form
could return incorrect data due to incorrect handling by the range optimizer. (Bug #54802)
const NOT BETWEEN
UPDATE IGNORE statement including a subquery that was evaluated using a temporary table, an error transferring the data from the temporary was ignored, causing an assertion to be raised. (Bug #54543)
DBUG_PRINT statement in
fill_schema_schemata() caused server crashes on Solaris. (Bug #54478)
MAX() with a subquery argument could raise a debug assertion for debug builds or return incorrect data for nondebug builds. (Bug #54465)
If one session attempted to drop a database containing a table which another session had opened with
HANDLER, any instance of
CREATE DATABASE, or
DROP DATABASE issued by the latter session produced a deadlock. (Bug #54360)
INFORMATION_SCHEMA plugins with no
deinit() method resulted in a memory leak. (Bug #54253)
Row subqueries producing no rows were not handled as
UNKNOWN values in row-comparison expressions. (Bug #54190)
SETUP_INSTRUMENTS.TIMER = 'NO' caused
TIMER_WAIT values for aggregations to be
NULL rather than 0. (Bug #53874)
max_length metadata value of
MEDIUMBLOB types was reported as 1 byte greater than the correct value. (Bug #53296)
If an application using the embedded server called
mysql_library_init() a second time after calling
mysql_library_end() to start and stop the server, the application crashed when reading option files. (Bug #53251)
The fix for Bug #30234 caused the server to reject the
DELETE Access compatibility syntax for multiple-table
DELETE statements. (Bug #53034)
plugin_audit.h include files are part of the public API/ABI, but were not tested by the ABI check. (Bug #52821)
An atomic “compare and swap” operation using x86 assembly code (32 bit) could access incorrect data, which would make it work incorrectly and lose the intended atomicity. This in turn caused the MySQL server to work on inconsistent data structures and return incorrect data. That code affected only 32-bit builds; the effect has been observed when icc was used to build binaries. With gcc, no incorrect results have been observed during tests, so this fix is a proactive one. Other compilers do not use this assembly code. (Bug #52419)
LOAD DATA INFILE, using a
SET clause to set a column equal to itself caused a server crash. (Bug #51850)
An assertion could be raised by
DELETE on a view that referenced another view which in turn (directly or indirectly) referenced more than one table. (Bug #51099)
In some cases, when the left part of a
NOT IN subquery predicate was a row and contained
NULL values, the query result was incorrect. (Bug #51070)
CHECKSUM TABLE for Performance Schema tables could cause a server crash due to uninitialized memory reads. (Bug #50557)
For some queries, the optimizer produced incorrect results using the Index Merge access method with
InnoDB tables. (Bug #50402)
EXPLAIN produced an incorrect
rows value for queries evaluated using an index scan and that included
GROUP BY, and
ORDER BY on a computed column. (Bug #50394)
mysql_use_result() are not for use with prepared statements and are not intended to be called following
mysql_stmt_execute(), but failed to return an error when invoked that way. (Bug #47485)
REPAIR TABLE on a
MERGE table caused the server to crash. (Bug #46339)
If the global and session
debug system variables had the same value, the debug trace file could be closed twice, leading to freeing already freed memory and a server crash. (Bug #46165)
ALTER EVENT failed to load an event after altering it, an assertion could be raised. This could occur, for example, if
ALTER EVENT was killed with
KILL QUERY. (Bug #44171)
Many type-punning warnings during compilation were silenced. (Bug #42733, Bug #11751755)
Trailing space removal for
utf32 strings was done with non-multibyte-safe code, leading to incorrect result length and assertion failure. (Bug #42511)
A malformed packet sent by the server when the query cache was in use resulted in lost-connection errors. (Bug #42503)
Multiple-statement execution could fail. (Bug #40877)
CREATE TABLE failed if a column referred to in an index definition and foreign key definition had different lettercases in the two definitions. (Bug #39932)
mysqlcheck behaved differently depending on the order in which options were given on the command line. (Bug #35269)
When invoked to display a help message, mysqld also displayed spurious warning or error messages. (Bug #30025)