MySQL 5.7.9

What's new in this version:

Audit Log Plugin Notes:
- Some events cannot be terminated. Previously, if an audit plugin returned nonzero status for a nonterminable event, the server ignored the status and continued processing the event. However, if an audit plugin used the my_message() function to terminate a nonterminable event, a server exit occurred. Now the server correctly handles termination of nonterminable events using my_message(). (Bug #21458066)

C API Notes:
- Incompatible Change: The mysql_options() C API function has two new options, MYSQL_OPT_MAX_ALLOWED_PACKET and MYSQL_OPT_NET_BUFFER_LENGTH, that set the max_allowed_packet and net_buffer_length system variables, respectively. Each option name also now can be passed to the mysql_get_option() C API function to retrieve its value. For more information, see mysql_options(), and mysql_get_option().
- The (undocumented) mysql_get_parameters() function has been removed. Applications that attempt to use it will get link errors and should be modified to use mysql_options() and mysql_get_option() instead.

Previously, it was necessary to call mysql_thread_end() for each mysql_thread_init() call to avoid a memory leak. C API internals have been reimplemented to reduce the amount of information allocated by mysql_thread_init() that must be freed by mysql_thread_end():
- For release/production builds without debugging support enabled, mysql_thread_end() need not be called.
- For debug builds, mysql_thread_init() allocates debugging information for the DBUG package (see The DBUG Package). mysql_thread_end() must be called for each mysql_thread_init() call to avoid a memory leak.

Compilation Notes:
- MySQL distributions now contain a mysqlclient.pc file that provides information about MySQL configuration for use by the pkg-config command. This enables pkg-config to be used as an alternative to mysql_config for obtaining information such as compiler flags or link libraries required to compile MySQL applications. For more information, see Building C API Client Programs Using pkg-config.
- A new INSTALL_PKGCONFIGDIR CMake option is available to specify the directory in which to install the mysqlclient.pc file. The default value is INSTALL_LIBDIR/pkgconfig, unless INSTALL_LIBDIR ends with /mysql, in which case that is removed first.

Packaging Notes:
- The shell and Perl versions of mysql_install_db have been removed from MySQL distributions. The executable C++ version of mysql_install_db implemented in MySQL 5.7.5 is still present, but remains deprecated (use mysqld --initialize instead) and will be removed in a future MySQL release.
- The deprecated _r versions of the libmysqlclient libraries are no longer installed.

Performance Schema Notes:
- Previously, the transaction instrument in the setup_instruments table was disabled by default, and the events_transactions_current and events_transactions_history consumers in the setup_consumers table were enabled by default. This setup is inconsistent, and having the consumers enabled could lead to the impression that transactions were instrumented by default. Now, the consumers are also disabled by default. To monitor transactions, enable the instrument and the applicable consumers.

With the show_compatibility_56 system variable disabled, SHOW VARIABLES and SHOW STATUS statements failed if MySQL was compiled without Performance Schema support. Consequently, it is no longer possible to compile without the Performance Schema. If it is desired to compile without particular types of instrumentation, that can be done with the following CMake options:
- For example, to compile without mutex instrumentation, configure MySQL using the -DDISABLE_PSI_MUTEX=1 option. (Bug #21669500)

Two changes were made regarding the effect of show_compatibility_56:
- Previously, when show_compatibility_56=OFF, selecting from the following INFORMATION_SCHEMA system and status variable tables returned an empty result and a deprecation warning: INFORMATION_SCHEMA.GLOBAL_VARIABLES, INFORMATION_SCHEMA.SESSION_VARIABLES, INFORMATION_SCHEMA.GLOBAL_STATUS, INFORMATION_SCHEMA.SESSION_STATUS
- This caused confusion for applications that were not aware that such selects could be empty: An empty result and a warning was not sufficient notice to signal the need to migrate to the corresponding Performance Schema system and status variable tables.
- To address this issue, selecting from the INFORMATION_SCHEMA system and status tables now produces an error, to make it more evident that an application is operating under conditions that require modification, as well as where the problem lies. The error code is ER_FEATURE_DISABLED_SEE_DOC. The error message indicates which table is disabled and that the show_compatibility_56 documentation should be consulted.

Previously, when show_compatibility_56=ON, selecting from the following Performance Schema status variable tables returned an empty result:
- performance_schema.global_status
- performance_schema.session_status
- This made it more difficult to migrate applications from the INFORMATION_SCHEMA status variable tables to the corresponding Performance Schema tables: Successfully selecting from the Performance Schema tables required knowing both that the server is from MySQL 5.7 and that show_compatibility_56=OFF.
- To address this issue, selecting from the Performance Schema status variable tables now produces the same result regardless of the value of show_compatibility_56. Thus, it is necessary to know only that the server is from MySQL 5.7.9 or higher. (If so, select from the Performance Schema tables. Otherwise, select from the INFORMATION_SCHEMA tables.)
- For additional information about the effects of show_compatibility_56 and migration issues, see Server System Variables, and Migrating to Performance Schema System and Status Variable Tables. (Bug #21606701)
- These Performance Schema tables now are world readable and accessible without the SELECT privilege: global_variables, session_variables, global_status, and session_status. An implication of this change is that SHOW VARIABLES and SHOW STATUS no longer require privileges on the underlying Performance Schema tables from which their output is produced when show_compatibility_56=OFF. (Bug #21251297)
- With the show_compatibility_56 system variable enabled, the reported values of the Last_query_cost and Last_query_partial_plans status variables were incorrect.
- With the show_compatibility_56 system variable disabled, the reported values of the Created_tmp_tables, Handler_external_lock, and Table_open_cache_% status variables were incorrect. (Bug #20483278, Bug #21788549, Bug #21788887)
- The Performance Schema session_account_connect_attrs had mistakenly been changed to require the SELECT privilege. It requires no special privileges again. (Bug #77702, Bug #21436364)
- In the setup_timers table, a CYCLE timer for ARM64 platforms is now available.

The Performance Schema now includes these instruments for monitoring I/O on binary log and relay log cache files:
- wait/io/file/sql/binlog_cache
- wait/io/file/sql/binlog_index_cache
- wait/io/file/sql/relaylog_cache
- wait/io/file/sql/relaylog_index_cache
- In addition, the default value of performance_schema_max_file_classes has been increased from 50 to 80.
- The Performance Schema threads table now contains a THREAD_OS_ID column that indicates the thread or task identifier as defined by the underlying operating system. For example, the column value corresponds to the Process Explorer thread ID on Windows and the gettid() value on Linux. For more information, see The threads Table.
- If you upgrade to this MySQL release from an earlier version, you must run mysql_upgrade (and restart the server) to incorporate this change into the performance_schema database.

Plugin Notes:
- The audit plugin API has been extensively revised to support a finer breakdown of the general event type (MYSQL_AUDIT_GENERAL_CLASS) into more specific events. This enables audit plugins to more precisely indicate the types of events in which they are interested and reduces overhead for plugins that have use for only a few event types. The API also now permits early termination of event execution. For more information, see Writing Audit Plugins. The general event type is still available but is deprecated and will be removed in a future MySQL release.
- In addition, a security_context plugin service is now available. Audit plugins can use this service to examine or modify the security context of threads associated with audited events. See MySQL Services for Plugins.

Security Notes:
- For RPM-based packages, the permissions used to create the data directory (/var/lib/mysql) have been changed from 755 to 751. This tightens the data directory permissions while still permitting world access to the mysql.sock file in that directory. (Bug #21066592)

Spatial Data Support:
- The required version of the Boost library for server builds has been raised from 1.58.0 to 1.59.0.

Spatial functions such as ST_MPointFromText() and ST_GeomFromText() that accept WKT-format representations of MultiPoint values now permit individual points within values to be surrounded by parentheses. For example, both of the following function calls are valid, whereas previously the second one produced an error:
- ST_MPointFromText('MULTIPOINT (1 1, 2 2, 3 3)')
- ST_MPointFromText('MULTIPOINT ((1 1), (2 2), (3 3))')
- In addition, functions such as ST_AsText() and ST_AsWKT() that produce WKT-format results now display MultiPoint values with parentheses surrounding each point.

Schema Notes:

The sys schema included in MySQL distributions was updated to version 1.5.0. This includes new features and fixes for several issues:
- sys schema objects now have a DEFINER of 'mysql.sys'@'localhost'. (Previously, the DEFINER was 'root'@'localhost'.) Use of the dedicated mysql.sys account avoids problems that occur if a DBA renames or removes the root account.
- The following sys schema session views now provide progress reporting for long-running transactions.
- The progress column of these views shows the percentage of work completed for stages that support progress reporting.
- A new diagnostics() stored procedure enables DBAs and other support personnel to collect diagnostic information for investigating MySQL instances. A new metrics view and statement_performance_analyzer() stored procedure provide supporting infrastructure for the diagnostics() procedure.
- The sys schema ps_is_instrument_default_enabled() and ps_is_instrument_default_timed() stored functions returned incorrect results in some cases.
- The ENABLED and HISTORY columns that were added to the Performance Schema setup_actors table in earlier MySQL 5.7 releases caused the sys schema ps_setup_reset_to_default() stored procedure not to work.
- Handing of event-timing information in the sys schema was updated to handle changes to Performance Schema event-timing columns in MySQL 5.7.8.
- mysql_upgrade previously checked for an exact object-count value in the sys schema to determine whether an upgrade was needed. If local objects had been added, the resulting reinstallation removed those objects. Now it checks for at least the expected number of objects.
- If you upgrade to this MySQL release from an earlier version, you must run mysql_upgrade to incorporate these changes into the sys schema.
- Thanks to Daniël van Eeden, Jesper Wisborg Krogh, Shlomi Noach, and Morgan Tocker for their contributions to this update. (Bug #78115, Bug #21647101, Bug #77927, Bug #21550271, Bug #78720, Bug #21966366)

Functionality Added or Changed:
- Incompatible Change; InnoDB: To better manage redo log format changes, the redo log header of the first redo log file (ib_logfile0) now includes a format version identifier and a text string that identifies the MySQL version that created the redo log files.
- A new boolean configuration option, innodb_log_checksums, replaces the innodb_log_checksum_algorithm option. innodb_log_checksums=ON enables a CRC-32C checksum, making it the only supported checksum for redo log pages.
- This patch also removes unused fields from the redo log header and checkpoint pages.
- Due to redo log format changes introduced by this patch, upgrading to or downgrading from MySQL 5.7.9 and higher requires a clean shutdown and, in some cases, removal of existing redo log files. For instructions related to this change, see Changes Affecting Upgrades to MySQL 5.7, and Changes Affecting Downgrades from MySQL 5.7.
- Important Change; InnoDB: DYNAMIC replaces COMPACT as the implicit default row format for InnoDB tables. A new configuration option, innodb_default_row_format, specifies the default InnoDB row format. Permitted values include DYNAMIC (the default), COMPACT, and REDUNDANT.
- The COMPACT row format remained the default row format until this release to ensure compatibility with older versions of InnoDB in MySQL 5.1 and earlier. Now that MySQL 5.1 has reached the end of its product lifecycle, the newer DYNAMIC row format becomes the default. For information about advantages of the DYNAMIC row format, see DYNAMIC and COMPRESSED Row Formats.
- Newly created tables use the row format defined by innodb_default_row_format when a ROW_FORMAT option is not specified explicitly or when ROW_FORMAT=DEFAULT is used.
- Existing tables retain their current row format if a ROW_FORMAT option was specified explicitly. If a ROW_FORMAT option was not specified explicitly or if ROW_FORMAT=DEFAULT was used, any operation that rebuilds a table also silently changes the row format of the table to the format defined by innodb_default_row_format. For more information, see Specifying the Row Format for a Table.
- Important Change: Introduced the -> JSON column-path operator. column->path is now supported as a synonym of JSON_EXTRACT(column, path), where column is a JSON column, and path is a valid JSON path.
- An expression with ->, like its equivalent that uses JSON_EXTRACT() instead, can be used in place of a column identifier wherever the latter can occur within a valid SQL statement. For example, the following CREATE TABLE and SELECT statements are valid.
- A column-path expression can be used for any column value that is read in a SELECT column list, or in a WHERE, ORDER BY, or GROUP BY clause in any SQL statement; such expressions cannot be used to set values.
- When an SQL statement contains one or more expressions using -> notation, each of these is translated into an equivalent expression that employs the JSON_EXTRACT() function instead. This can be seen in the output from EXPLAIN when used on such a statement.
- Like JSON_EXTRACT(), the -> operator returns as NULL if no matching value for an otherwise valid path is found.
- For more information about -> and JSON_EXTRACT(), see Functions That Search JSON Values. See Searching and Modifying JSON Values, for information about JSON path support. See also Secondary Indexes and Virtual Generated Columns, for additionl information and examples.
- InnoDB: A new INNODB_METRICS server operations counter (innodb_dict_lru_count) counts the number of tables evicted from the table cache LRU list. Thanks to Daniël van Eeden for the patch.
- InnoDB: The new innodb_numa_interleave read-only configuration option allows you to enable the NUMA interleave memory policy for allocation of the InnoDB buffer pool. When innodb_numa_interleave is enabled, the NUMA memory policy is set to MPOL_INTERLEAVE for the mysqld process. After the InnoDB buffer pool is allocated, the NUMA memory policy is set back to MPOL_DEFAULT. This option is only available on NUMA-enabled systems.
- MySQL distributions now include these header files because my_sys.h depends on them: my_thread_local.h, thr_cond.h, thr_mutex.h, thr_rwlock.h.
- MySQL Server RPM packages now obsolete MySQL Connector C. Installing MySQL Server causes older libmysqlclient from any MySQL Connector C packages to be removed and replaces them with the current libmysqlclient.
- yaSS was upgraded to version 2.3.8.
- Insert overhead for the MEMORY storage table was reduced by caching computed hash values.
- RPM .spec files were updated so that MySQL Server builds from source RPM packages will include the proper files to take advantage of operating system NUMA capabilities. This introduces a runtime dependency on libnuma.so.1. RPM and yum detect this and refuse to install if that library is not installed.
- The initial-password message written by mysqld --initialize to the error log has been reduced from a [Warning] to a [Note].
- The JSON_APPEND() function was renamed to JSON_ARRAY_APPEND().
- Unneeded scripts and test files were removed from the tests directory and that directory was renamed to testclients.
- The JSON value-updating functions JSON_APPEND(), JSON_SET(), JSON_REPLACE(), and JSON_INSERT() now treat SQL NULL values as JSON null literals, which is consistent with JSON_OBJECT() and JSON_ARRAY().
- The server now generates a warning when the default_storage_engine or default_tmp_storage_engine system variable is set to a disabled storage engine named in the disabled_storage_engines system variable.
- Metadata locking for tablespaces has been extended so that, for DDL statements that refer to multiple tablespaces, a metadata lock is acquired on all used tablespaces.
- Support for building with Solaris Studio 5.13 was added.
- mysql_ssl_rsa_setup now is less noisy by default. Output from openssl commands is displayed only if --verbose is given.
- yaSSL was upgraded to version 2.3.7d. This fixes a connection-failure issue whenused with the thread pool plugin.
- mysqlpump now supports a --version option.
- Performance Schema digests in DIGEST_TEXT columns have ... appended to the end to indicate when statements exceed the maximum statement size and were truncated. This is also now done for statement text values in SQL_TEXT columns.
- Output from mysql_upgrade is now less noisy and more informative.
- A new SHUTDOWN SQL statement is available. This provides a SQL-level interface to the same functionality previously available using the mysqladmin shutdown command or the mysql_shutdown() C API function. See SHUTDOWN Syntax.
- Work was done to clean up the source code base, including: Removal of unneeded CMake checks; removing unused macros from source files; reorganizing header files to reduce the number of dependencies and make them more modular, removal of function declarations without definitions.

Bugs Fixed:
- Important Change; InnoDB; Partitioning: There was no way to upgrade existing partitioned tables to use the native partitioning implemented for InnoDB tables in MySQL 5.7.6. This fix adds support to both mysql_upgrade and the mysql client for upgrading partitioned InnoDB tables created in previous releases, which used the ha_partition handler, to use InnoDB native partitioning instead.
- mysql_upgrade now checks for all InnoDB tables that were created using the generic ha_partition handler and attempts to upgrade them to InnoDB native partitioning.
- In the mysql client, pre-5.7.6 partitioned InnoDB tables can be upgraded one by one to native partitioning using the ALTER TABLE ... UPGRADE PARTITIONING statement that is implemented in this release.
- Important Change; Replication: The START SLAVE and STOP SLAVE statements can no longer be used for the group_replication_recovery channel. See Replication Channels, for more information.
- Important Change; Replication: In MySQL 5.7, the binlog_max_flush_queue_time system variable introduced in MySQL 5.6 no longer has any effect. This variable is now deprecated, and has been marked for eventual removal in a future MySQL release.
- Important Change; Replication: When using a single-threaded slave, the status of the applier thread is now reported as part of the replication_applier_status_by_worker table, instead of replication_applier_status_by_coordinator. This means that replication_applier_status_by_coordinator is now empty when using a single-threaded slave; it should be noted that such reporting for multi-threaded slaves has not been changed, and continues to be shown in that table.
- InnoDB: The undo log contained insufficient information about virtual columns and virtual column indexes, which could cause a server exit when adding or dropping virtual columns. As a result, a slow shutdown (using innodb_fast_shutdown=0) is required prior to performing an in-place upgrade or downgrade from MySQL 5.7.8. For more information, refer to Upgrading or Downgrading MySQL.
- InnoDB: A table-rebuilding ALTER TABLE operation that created an index on a virtual column raised an assertion.
- InnoDB: A SELECT ... FOR UPDATE operation on a table with virtual generated columns raised an assertion. InnoDB unnecessarily retrieved a non-indexed virtual column for a covered secondary index scan.
- InnoDB: Creating a table with large rows failed when using a ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED due to an incorrect undo log record size calculation. The same operation succeeded with ROW_FORMAT=COMPACT or ROW_FORMAT=REDUNDANT.ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPRESSED now permit a row length violation at DDL time if innodb_strict_mode is disabled.
- InnoDB: Adding an index on a generated virtual column with an index prefix length that exceeded the maximum length caused a server exit.
- InnoDB: A memory leak occurred after crash recovery. Memory allocated in fil_space_read_name_and_filepath() was not freed.
- InnoDB: Altering the data type or computation method of a virtual generated column caused an error.
- InnoDB: An assertion was raised when creating a spatial index. InnoDB failed to count irtual columns that preceded the spatial index column.
- InnoDB: Tablespace discovery modifications in MySQL 5.7.5 included the removal of code related to MLOG_FILE_CREATE2 redo log records. As a result, the redo log did not contain sufficient information about file creation.
- InnoDB: The redo log provided no indication that redo logging is disabled for index page writes during concurrent DDL operations. As a result, external hot backup tools could produce corrupt backups.
- InnoDB: In debug builds, recovery asserted during a transparent page compression test. A torn page from the doublewrite buffer caused an LSN debug check failure.
- InnoDB: Creating a virtual generated column on a partitioned table caused a server exit.
- InnoDB: An old version of numactl headers on the build host caused a compilation error when building a MySQL version that includes NUMA memory policy support.
- InnoDB: A SELECT .. FOR UPDATE operation on an indexed virtual generated column raised an assertion.
- InnoDB: The mysql_system_tables_fix_for_downgrade.sql script, provided to facilitate system table alterations when downgrading from MySQL 5.7.6 or higher, was removed from the MySQL installation directory. The script was no longer valid for all downgrade paths. For more information, see Changes Affecting Downgrades from MySQL 5.7.
- InnoDB: The tablespace identifier (space_id) was logged twice for an MLOG_TRUNCATE redo log record.
- InnoDB: Crash recovery issued an invalid error message indicating that an isl file could not be opened or is not correct.
- InnoDB: The InnoDB Monitor displayed incorrect mutex creation information for a mutex with a long semaphore wait.
- InnoDB: Error message formatting was corrected in the os_file_write_page() function.
- InnoDB: The i_s_dict_fill_sys_tablespaces() function could free memory associated with a file name before printing an error message.
- InnoDB: In debug builds, scanned and applied redo log records are now printed with a string identifier instead of a numerical identifier when --debug=d,ib_log is passed to mysqld. For example, rec MLOG_2BYTES is now printed instead of rec 2.
- InnoDB: The number of system calls made by the InnoDB page compression feature was reduced.
- InnoDB: The recv_parse_log_rec function returned the length of the redo log record instead of 0 when encountering an incomplete MLOG_CHECKPOINT record.


InnoDB: In debug builds, a parenthesis mismatch in a MATCH() ... AGAINST clause raised a full-text parser assertion.
InnoDB: Debug code was added to avoid a doublewrite buffer assertion that was raised during Valgrind testing.
InnoDB: Log messages were improved to help identify out-of-space errors that could occur when adding rollback segments.
InnoDB: In debug builds, enabling the innodb_log_checkpoint_now debug option while a table-rebuilding ALTER TABLE operation is running could result in an infinite loop.
InnoDB: UNIV_INLINE was not defined for the dict_table_has_indexed_v_cols function.
InnoDB: Altering a virtual column data type is not supported as an in-place operation.
InnoDB: Sorting was skipped by an ALTER TABLE statement that changed the primary key and dropped the last column of the previous primary key.
InnoDB: During recovery, an invalid isl file was treated as a missing isl file, resulting in the tablespace being opened using the file location defined in an MLOG_FILE_* record in the redo log. Recovery no longer opens the tablespace if the isl file is invalid.
InnoDB: Code that provided a timeout mechanism intended to reduce adaptive hash index search latch (btr_search_latch) contention was removed. The code became obsolete after the introduction of adaptive hash index search system partitioning in MySQL 5.7.8.
InnoDB: The COMPRESSION option was displayed incorrectly in SHOW CREATE TABLE output.
InnoDB: An innodb_data_file_path mismatch raised an assertion, as did initializing the database with a data file size that was too small for the specified page size.
InnoDB: A check was added to prevent accessing full-text index tables that are in an inconsistent state.
InnoDB: A schema mismatch error occurred when importing a tablespace that was altered by DROP INDEX operation on the source server.
InnoDB: A virtual generated column on a table that uses index condition pushdown (ICP) caused an assertion.
InnoDB: Creating an index on a virtual generated column after adding a full-text index on a preceding column raised an assertion.
InnoDB: An INSERT operation raised a btr_search_enabled assertion. Assertion code was too restrictive.
InnoDB: Reloading a table that was evicted while empty caused an AUTO_INCREMENT value to be reset.
InnoDB: A crash during a TRUNCATE TABLE operation caused the server to exit on startup.
InnoDB: The .isl was not removed when dropping a general tablespace that was created outside of the MySQL data directory.
InnoDB: An inconsistent read occurred under the REPEATABLE READ transaction isolation level. Transactions that operated on the same row were removed from transaction ID list in the incorrect order.
InnoDB: Attempting to create a general tablespace data file on a Windows root drive caused an error.
InnoDB: After disabling the adaptive hash index feature, an adaptive hash index latch was unnecessarily obtained and released.
InnoDB: An ALTER TABLE operation on a table with an index defined on a virtual column incorrectly modified the data of the virtual column.
InnoDB: Resizing the buffer pool online raised an assertion due to a memory full condition.
InnoDB: An ALTER TABLE ... IMPORT TABLESPACE operation with innodb_page_size=4K and ROW_FORMAT=DYNAMIC raised an assertion.
InnoDB: For spatial indexes, InnoDB unnecessarily stored a 3072-byte prefix in undo log records instead of just the maximum bounding rectangle (MBR). For columns with externally stored data, both the prefix and MBR are logged.
InnoDB: Invalid init_ftfuncs() assertion code was removed.
InnoDB: Memory allocation sanity checks were added to the memcached code.
InnoDB: An incorrect reference count caused a hang in the TrxInInnoDB constructor. innobase_close_connection() released the transaction object before destroying the TrxInInnoDB object where the reference count is adjusted.
InnoDB: A MySQL 5.7.8 patch that reintroduced SHOW ENGINE INNODB MUTEX functionality caused a performance regression.
InnoDB: A memcached flush_all command raised an assertion. A function that starts a transaction was called from within assertion code.
InnoDB: A shutdown hang occurred when a high priority transaction waited for a victim transaction to exit while the victim transaction waited for an asynchronous rollback t
InnoDB: A data corruption occurred on ARM64. GCC builtins did not issue the correct fences when setting or unsetting the lock word.
InnoDB: Server shutdown was delayed waiting for the purge thread to exit. To avoid this problem, the number of calls to trx_purge() was reduced, and the trx_purge() batch size was reduced to 20.
InnoDB: In READ COMMITTED mode, a REPLACE operation on a unique secondary index resulted in a constraint violation. Thanks to Alexey Kopytov for the patch.
InnoDB: The IBUF_BITMAP_FREE bit indicated that there was more free space in the leaf page than was actually available.
InnoDB: Moving the data directory before recovering a crashed database caused tablespace discovery to fail for file-per-table tablespaces created outside of the MySQL data directory.
InnoDB: The innodb_buf_flush_list_now debug setting failed to flush all dirty pages to disk.
InnoDB: An ALTER TABLE ... ADD FULLTEXT INDEX operation raised an assertion. A thread attempted to use a lower priority transaction that was being rolled back before the rollback operation completed.
InnoDB: Running an ALTER TABLE operation on a referencing table with a cascading foreign key constraint during a concurrent DML operation on the referenced table caused a loss of referential integrity.
InnoDB: Setting lower_case_table_names=0 on a case-insensitive file system could result in a hang condition when running an INSERT INTO ... SELECT ... FROM tbl_name operation with the wrong tbl_name letter case. An error message is now printed and the server exits when attempting to start the server with --lower_case_table_names=0 on a case-insensitive file system.
InnoDB: The server failed to start with an innodb_force_recovery setting greater than 3. InnoDB was set to read-only mode before redo logs were applied.
DROP TABLE is now supported with an innodb_force_recovery setting greater than 3.
InnoDB: The trx_sys_read_pertable_file_format_id() function reported the wrong file format.
Packaging; OS X: Using user=mysql during installation on OS X did not allow the mysql database to be installed. To fix this problem, OS X packages now use the --no-defaults option when creating this database. This also means that having a my.cnf file on the system no longer affects the installation.
Partitioning: Error handling for failed partitioning-related ALTER TABLE operations against non-partitioned tables was not performed correctly.
Partitioning: ALTER TABLE when executed from a stored procedure did not always work correctly with tables partitioned by RANGE.
Replication: The group replication applier channel does not support DATABASE as the slave_parallel_type; when group replication is started, this is checked for explicitly, and handled correctly. However, it remained possible to change this value indirectly at a later point in time by increasing the value of slave_parallel_workers while the slave SQL thread was stopped, which caused the applier to fail with an error. To fix this problem, the slave_parallel_type for the group_replication_applier is now checked to make sure that it is set to LOGICAL_CLOCK whenever the number of slave_parallel_workers is set greater then 0, and not merely when group replication is first started.
Replication: As binlog_error_action=ABORT_SERVER is the default in MySQL 5.7.7 and later it is being used for more error situations. The behavior has been adjusted to generate a core dump to improve troubleshooting possibilities.
Replication: At runtime, some Gtid_set objects could be instrumented with a performance schema mutex key equal to 0 (which is invalid), due to its use as the effective default value when the mutex key was not actually supplied. This allowed these objects to be created without a valid key, which led to further issues when using them.
Replication: When running the server with gtid_mode=ON, a DELETE from a MEMORY table following a restart was not written to the binary log correctly.
Replication: The locking behavior of replication administration statements has changed to make SHOW SLAVE STATUS more concurrent. This makes the NONBLOCKING clause redundant for SHOW SLAVE STATUS and it has been removed.
Replication: ER_CANT_USE_AUTO_POSITION_WITH_GTID_MODE_OFF errors were not reported using the correct format.
Replication: When the dump thread was killed while dumping an inactive binary log, some events in this log could be skipped and thus not replicated.
Replication: XA transactions could cause an assert condition on XA COMMIT; this was happening because the internal transaction state was not reset between XA PREPARE and XA COMMIT or XA ROLLBACK, due to the fact that these operations constitute separate transactions under XA. In addition, XA ROLLBACK statements were not handled properly in some cases.
Replication: The interface between the Group Replication plugin and the Performance Schema engine made use of a type of memory allocation which was passed to the server, and was a potential source of problems when passing information between the plugin and performance_schema tables. The implementation for this interface has been reworked so as to avoid performing this type of memory allocation when sharing data.
Replication: The MTS submode set for each channel was ignored by the worker threads, which continued to read and use the global flag set for all slave channels. This could lead to errors when the coordinator was of one type and its workers of another.
Replication: Replication slaves could fail for having insufficient privileges when they had been granted only the REPLICATION SLAVE privilege.
Replication: The status variable Slave_open_temp_tables keeps track of the number of temporary tables that are opened by the replication slave. If multi-source replication is enabled, it is the total number of temporary tables for all channels. This fix addresses the following issues relating to this variable: RESET SLAVE FOR CHANNEL channel forced the value of Slave_open_temp_tables to 0; in the event that some other replication channel had open temporary tables which were later dropped, the value wrapped around to a large negative value (1 - 232). This also caused spurious or missed warnings when issuing a STOP SLAVE or CHANGE MASTER TO statement.

The internal function that modifies Slave_open_temp_tables in such cases relied on two incorrect assumptions:
- That the variable is updated by only one thread when multi-threaded slaves are not enabled, which is not true in the case of multi-source replication.
- That non-atomic operations are safe with a single writer and multiple readers, which is not necessarily true for some platforms supported by MySQL.
- Replication: The warning '@@session.gtid_executed' is deprecated and will be removed in a future release. was printed even when the session variable gtid_executed was not included in the result of a query. In addition, the result of SELECT @@session.gtid_executed included a duplicate warning. Both issues occurred because the warning was printed whenever the value of gtid_executed was accessed by a statement, such access occurring as a matter of course, whether or not a given variable is actually included in the result.

To fix this issue, we make handling of @@session.gtid_executed consistent with how the also-deprecated variable @@global.sql_log_bin is treated in such cases, by making the following changes:
- gtid_executed is no longer included in the performnce_schema.session_variables table.
- gtid_executed is still included in the information_schema.session_variables table, but when show_compatibility_56 = ON, the warning is not issued when querying the session_variables table, or when issuing SHOW VARIABLES or SHOW SESSION VARIABLES, even when using a matching LIKE clause with either of the SHOW statements.
- The warning is still issued by a statement such as SELECT @@session.gtid_executed which accesses the value of the variable directly.
- Replication: When a transaction consisting of a single statement with a specified GTID failed in autocommit mode, its GTID was not released when rolling it back when binary logging was disabled.
- Replication: The slave group event parser did not properly register an XA_ROLLBACK event as a transaction boundary.
- Replication: mysqlbinlog printed a ROLLBACK statement at the end of the binary log file, which when played back failed with error 1782 @@SESSION.GTID_NEXT cannot be set to ANONYMOUS when @@GLOBAL.GTID_MODE = ON. This occurred when the binary log file did not include any data related events, or when the relay log file included a Format_description_log_event that had been generated on the master at server startup.
- The fix for this issue causes a relay log's Format_description_log_event to do nothing if it is applied by a BINLOG statement, and stops a ROLLBACK from setting gtid_next to ANONYMOUS when the state of gtid_next has not yet been determined by a subsequent event.
- Replication: SAVEPOINT and ROLLBACK TO SAVEPOINT within a trigger led to an assertion.
- Replication: While a SHOW BINLOG EVENTS statement was executing, any parallel transaction was blocked. The fix ensures that the SHOW BINLOG EVENTS process now only acquires a lock for the duration of calculating the file's end position, therefore parallel transactions are not blocked for long durations.
- Replication: If a CREATE VIEW statement failed, it was being incorrectly written to the binary log even though it did not result in the creation of a partial view. The fix ensures that such statements are not recorded in the binary log. Additionally it was found that when a statement which had failed on a master was received by a slave with an expected error, if the statement was skipped on the slave, for example due to a replication filter, the expected error was being compared with the actual error that happened on the slave. The fix ensures that if a statement with an expected error is received by a slave, if the statement has not been filtered, only then is it compared with the actual error that happened on the slave.
- Replication: The action specified for binlog_error_action was not always honored correctly after a hardware failure occurred during log rotation.
- Replication: When using MySQL 5.7.6 and later with binlog_format=row and gtid_mode=off, if CREATE ... SELECT was killed during execution it could lead to an inconsistent state, breaking replication. The cause was that in MySQL 5.7.6 the way CREATE ... SELECT was logged was changed, so that a commit was introduced between the CREATE TABLE and SELECT steps. The fix ensures that CREATE ... SELECT does not commit in the middle of the transaction when binlog_format=row.
- Replication: Modifying the master_info_repository or relay_log_info_repository inside a transaction and later rolling back that transaction left the repository in an unusable state. We fix this by preventing any modification of these repositories inside a transaction.
- Replication: Transactions added to gtid_executed using SET gtid_purged were not taken into account by WAIT_FOR_EXECUTED_GTID_SET() until a subsequent transaction was committed by a client or slave thread.
- Replication: When the running with --relay-log-info-repository=TABLE, the mysql.slave_relay_log_info table is updated when a transaction is committed or when a flush is performed explicitly, such during relay log rotation. If a transaction that uses any nontransactional tables (such as MyISAM tables) is split across multiple relay logs, it is partially committed on relay log flush. When gtid_mode=ON, this caused the same GTID to be used for the remaining portion of the transaction, which raised an ER_GTID_NEXT_TYPE_UNDEFINED_GROUP error.
- We fix this issue by postponing in such cases the update of the relay log information repository that normally occurs on relay log rotation until the commit for the transaction in question has been executed.
- This issue did not affect tables using transactional storage engines such as InnoDB.
- For an index-only scan over an indexed generated column, the server could do random calculations; the random results were not exposed to the user, but Valgrind warnings could occur, and the server could exit when calculations involved function
- Although the use of JSON values with GREATEST() or LEAST() is not currently supported, the server did not handle attempts to do so correctly, leading to an assert (Linux) or exit() call (Windows) in debug builds. Now when you try to use JSON values with either of these functions, the server emits a suitable warning (ER_NOT_SUPPORTED_YET).
- For debug builds, a DROP TRIGGER statement could raise an assertion if the trigger was defined on a table that contained a generated column.
- A table that included a generated column referencing a JSON column in some cases become corrupted, so that a subsequent access of the table using a different connection caused the server to fail.
- For tables with VIRTUAL generated columns, an INSERT with an empty values list could cause a server exit.
- CMake configuration was adjusted to handle new warnings reported by Clang 3.7.
- Using a materialized view defined over a table containing generated columns could cause a server exit.
- For partitioned InnoDB tables containing a virtual generated column, reads from the table could return random data for the column.
- The CMake checks for NUMA availability could cause compilation problems on platforms without NUMA support.
- The optimizer did not consider nonfunctional expressions such as (a AND b) = 1 when looking for indexed generated columns to substitute for the (a AND b) expression. Now expressions using the AND and OR logical operators are considered.
- For debug builds, when the optimizer tried to clone certain types of keys for a range optimization, an assertion was raised.
- For debug builds, the server could exit when the optimizer attempted to estimate the cost for processing unique values when there were no keys.
- An INSERT into a view with a subquery could fail if executed as a prepared statement.
- mysql-test-run.pl now has an --valgrind-clients option that causes all clients started by .test files to be run with valgrind. This option requires valgrind 3.9 or later.
- In addition, several client memory leak issues were fixed.
- For queries on InnoDB tables for which the optimizer used SPATIAL indexes for full index scans, the result was empty because such indexes do not support a full scan. The optimizer no longer considers SPATIAL indexes as candidates for full index scans.
- For some inputs, ST_Intersection() could return an invalid polygon.
- Added Microsoft Visual Studio 2015 support. Changes include using the native (added in VS 2015) timespec library if it exists, renamed lfind/lsearch and timezone/tzname to avoid redefinition problems, set TMPDIR to "" by default as P_tmpdir no longer exists, deprecated std::hash_map in favor of std::unordered_map, and added Wix Toolset 3.10 support.
- Spatial functions could simplify geometry values in contexts where the value might be used elsewhere in a query, producing incorrect results.
- If ST_ConvexHull() or ST_SRID() were used in a view definition, the resulting definition contained ST_Convex_Hull() (misspelled) or SRID() (deprecated).
- JSON_TYPE() returned OPAQUE for some binary values that it should have identified as BLOB.
- The mysql client parser incorrectly interpreted optimizer hint comments that contained ;, ", ', or ` characters.

These Version Tokens issues were resolved:
- version_tokens_delete() now strips whitespace surrounding token names in its argument, similar to version_tokens_set() and version_tokens_edit().
- Passing NULL to version_tokens_delete() caused a server exit.
- Passing an argument with an empty token name to to version_tokens_set() or version_tokens_edit() caused a server exit.
- Passing NULL as the timeout value to version_tokens_lock_exclusive() or version_tokens_lock_shared() caused a server exit.
- For debug builds, enabling the PAD_CHAR_TO_FULL_LENGTH SQL mode could cause SHOW FUNCTION STATUS to raise an assertion.
- mysqlpump did not exit with a message for some combinations of incompatible options.
- An assertion could be raised if the optimizer tried to create a temporary table based on a prepared statement parameter.
- Executing a prepared statement with multiple nested subqueries could raise an assertion.
- For debug builds, failure of subquery optimization could cause an assertion to be raised due to improper error handling.
- Some table and index optimizer hints were lost early in statement processing, so query rewrite plugins did not have access to them. This could cause incorrect matching between incoming statements and statement pattern templates.
- Queries containing nested subqueries combining grouping and outer references might cause a server exit.
- Passing NULL as the second or third argument to ST_AsGeoJSON() could cause the server to stop responding to the session or (in debug builds) to raise an assertion.
- Giving input to HANDLER READ that could not be converted to the correct type could cause the server to stop responding to the session or (in debug builds) to raise an assertion.
- For debug builds, ST_IsValid(NULL) could raise an assertion.
- For debug builds, an assertion could be raised for negative zero values when converting time values to decimal.
- ST_AsWKB() could cause a server exit if given invalid data.
- If an aggregate function was used over a generated column that was itself part of a multiple-column index, the server could exit.
- A missing error check during column reference resolution could result in an incorrect error message or (in debug builds) an assertion being raised.
- For debug builds, an assertion could be raised in Filesort::make_sortorder() for attempts to sort Item_ref objects.
- JSON functions could return incorrect values if a path argument was passed as a user-defined variable that changed values between result set rows.
- Columns selected from the right-hand table of a left join, which was also a derived table, might produce incorrect NULL value information when used in an IN subquery.
- For debug builds, an assertion was raised for some queries that have a semi-join and use the materialization strategy, if a key length or number of key parts was zero.
- Compilation using gcc 4.9 or 5.1 failed on ARM64 platforms.
- If a multiple-column UPDATE statement failed to update a JSON column that was then referenced in a later update, the server could exit.
- For debug builds, invalid geometry byte strings could cause spatial functions to raise an assertion rather than return an error to the caller.
- For debug builds, a too-strict assertion could be raised by invalid characters for LOAD DATA.
- The server could exit when InnoDB tried to update a secondary index on a VIRTUAL generated column of type BLOB.
- Binary logging of CREATE USER and statements could log the hash of the password hash (rather than the hash itself) when log_backward_compatible_user_definitions was enabled. Binary logging of ALTER USER statements could include attributes not present in the original statements.
- In consequence of the fix for these issues, log_backward_compatible_user_definitions has been replaced by log_builtin_as_identified_by_password. If this variable is enabled, binary logging for CREATE USER statements involving built-in authentication plugins rewrites the statements to include an IDENTIFIED BY PASSWORD clause, and SET PASSWORD statements are logged as SET PASSWORD statements, rather than being rewritten to ALTER USER statements.
- For debug builds, incorrect caching of JSON values could cause an assertion to be raised.
- An empty string (which is not a valid JSON value) normally is parsed and returned as a JSON null literal, but in some cases could raise an assertion for debug builds.
- For temporary tables created to handle UNION statements that selected CHAR or SET columns, the maximum column width could be too long for InnoDB to handle. Now such columns are created as variable-length columns.
- Certain JSON functions could return incorrect results when used in prepared statements which had path expression constants.
- Valgrind errors could occur during partition pruning for tables containing generated columns.
- For builds configured with MAX_INDEXES greater than 64, certain queries for which the server used temporary tables could cause a server exit.
- Adding or dropping a VIRTUAL generated column could cause a server exit.
- An attempt to use a previously unused time zone with CONVERT_TZ() could produce warnings or errors or (in debug builds) raise an assertion if GTIDs were enabled but the binary log was not enabled.
- In debug builds, an attempt to use a previously unused time zone with CONVERT_TZ() or as the value of the time_zone system variable inside a stored program could raise an assertion.
- For plugins that use the audit plugin API, MYSQL_AUDIT_GLOBAL_VARIABLE_SET events passed to the notification function did not include the new variable value.
- Queries on a table containing an indexed generated column could fail if the table name contained special characters.
- When a view was the inner table of an outer join, a JSON column could produce a non-NULL value when NULL was expected.
- If JSON_CONTAINS_PATH() was called with a one_or_all argument of all and a path argument contained a wildcard, the function found all matches per path, even though in this case one match is sufficient.
- JSON_SET() and JSON_REPLACE() sometimes produced an incorrect result if a path expression identified a nonarray value.
- Suppression of JSON conversion errors using non-strict SQL mode or INSERT IGNORE could then cause an assertion to be raised if an empty value inserted into a JSON NOT NULL column was copied to another JSON column.
- For debug builds, some spatial functions that accept raw byte data for spatial arguments (for example, specified as hex values) could raise an assertion if such an argument contained extra garbage following valid data.
- ALTER TABLE could raise an assertion for a table with an indexed virtual column having a column position greater than 64.
- Evaluation of virtual generated columns could fail to evaluate all base columns and result in invalid memory reads.
- For deeply nested JSON input, ST_GeomFromGeoJSON() or JSON_VALID() could produce stack overflow.
- Failure to parse a JSON string that contained a floating-point number with a large, negative exponent could cause a server exit.
- For debug builds, an incorrect assertion could be raised during subquery execution.
- For debug builds, a missing error check in Item_sum_hybrid::fix_fields() caused an assertion to be raised.
- For debug builds, invoking ST_AsGeoJSON() within GROUP BY ... WITH ROLLUP could raise an assertion.
- JSON_SEARCH() could return incorrect results if an invalid escape expression was specified.
- For debug builds, a NULL first argument to JSON_SET() could raise an assertion.
- For expressions of the form (subquery) IN (subquery), where a subquery could return a JSON value, failure to handle a row result could cause a server exit.
- Failure of JSON_APPEND() to handle a legal condition could cause a server exit.
- Calls to ST_Buffer() could hang or raise an assertion.
- The server could exit in unclean fashion if configured to listen on a TCP/IP port number already in use by another server instance.
- Certain subqueries as arguments to PROCEDURE ANALYSE() could cause a server exit.
- A query with a right outer join inside a derived table might return wrong data.
- Starting the server with --skip-grant-tables (or with options such as --initialize for which --skip-grant-tables is implicit) prevented the INSTALL PLUGIN and UNINSTALL PLUGIN statements from working.
- mysql_ssl_rsa_setup could create an unwanted .rnd file in the data directory. (The file is actually created by openssl, which mysql_ssl_ras_setup invokes. mysql_ssl_rsa_setup now cleans up the file.)
- For spatial functions, input polygons were automatically closed if open. However, the Open Geospatial Consortium guidelines require that input polygons already be closed. Unclosed polygons are now rejected as invalid rather than being closed.
- When a VIRTUAL generated column was added to a table, it was not ensured that data being calculated by the generated column expression would not be out of range for the column. This could lead to inconsistent data being returned and unexpectedly failed statements.

ALTER TABLE now supports WITHOUT VALIDATION and WITH VALIDATION clauses to control whether ALTER TABLE validates the data for a VIRTUAL generated column:
- With WITHOUT VALIDATION (the default if neither clause is specified), an in-place operation is performed (if possible), data integrity is not checked, and the statement finishes more quickly. However, later reads from the table might report warnings or errors for the column if values are out of range.
- With WITH VALIDATION, ALTER TABLE copies the table. If an out-of-range or any other error occurs, the statement fails. Because a table copy is performed, the statement takes longer.
- WITHOUT VALIDATION and WITH VALIDATION are permitted only with ADD COLUMN, CHANGE COLUMN, and MODIFY COLUMN operations.
- Some INFORMATION_SCHEMA queries consumed excessive memory due to suboptimal query plans and insufficient materialization.
- With the server configured to send error messages to syslog or a log file, messages generated prior to error log setup were sent to stderr or stdout. These messages are now buffered until error log setup has completed, then logged to the proper destination.
- Statement digests did not include information about optimizer hint comments, causing statements with and without hints to be aggregated.
- Executing a prepared statement using a derived table and an aggregate function in a subquery in the SELECT list could cause a server exit.
- GRANT created the account for nonexistent accounts even if the NO_AUTO_CREATE_USER SQL mode was enabled.
- ST_SymDifference() with multipolygon arguments could return incorrect results.
- A query with a NOT IN subquery that had COUNT(DISTINCT) could return incorrect results.
- When started using a very old data directory (from MySQL 5.0), the server could exit due to failure to properly read the old grant tables.
- A mulitple-table update involving generated columns that updated used a temporay table could cause a server exit or raise an assertion.
- Unit testing now uses Google Mock 1.7 rather than 1.6.
- For queries containing an expression of the form (x IN (subquery)) IN (subquery2), a combination of semi-join and subquery materialization strategies could cause a server exit.
- If a query contained an outer join such as LEFT JOIN (t1,t2,...) and a hint was used to disable join buffering on a right-side table but not on the others, a server exit occurred.
- For a cursor type of CURSOR_TYPE_READ_ONLY, retrieving the result set for the first execution of a prepared CALL staement could be missing the first result set row if the data was numeric; raise an assertion for debug builds if the data was string; cause loss of the server connection when calling mysql_stmt_fetch().
- Dangling blob pointers could remain when closing an InnoDB table, resulting in a subsequent read of invalid memory and a server exit.
- For some operations where sorting or grouping required a temporary table, the table could have zero columns and raise an assertion.
- For debug builds, merging a derived table into an outer query block could raise an assertion.
- Queries containing an expression of the form (x IS NULL) IN (subquery) could cause a server exit.
- During server SSL file autogeneration, ca.pem briefly had insecure file permissions.
- An assertion could be raised due to incorrect error handling if a SELECT ... FOR UPDATE subquery resulted in deadlock and caused a rollback.
- Selecting the result of an INSERT() function call to which input was passed as a hexidecimal string could expose more information than was passed to the function.
- Subqueries having COUNT() with GROUP BY could yield incorrect results.
- The updatable property of a view is set during view creation. If the underlying table was dropped and re-created as a nonupdatable one, the updatable property of the original view was not revised accordingly. This could cause a server exit for attempts to insert or replace into the view is made. (This problem was specific to views with multiple tables/views and did not occur with update statements.)
- The locking functions provided by the version_token plugin were renamed: vtoken_get_read_locks(), vtoken_get_write_locks(), and vtoken_release_locks() are now named version_tokens_lock_shared(), version_tokens_lock_exclusive(), and version_tokens_unlock(), respectively.
- These functions also failed to have any effect because they were implicitly unlocked at the end of the statement in which they were set.
- Servers linked against yaSSL and compiled with GCC 4.8.2 could fail to respond correctly to connection attempts until several seconds after startup.
- yaSSL was upgraded to version 2.3.7e.
- When upgrading an old data directory (MySQL 5.0 or 5.1), mysql_upgrade could fail to properly read the mysql.proc table.
- For tables with subpartitions, the server could exit due to incorrect error handling during partition pruning if the partition could be identified but not the subpartition.
- mysql_upgrade could fail to look for checked tables in the wrong database during the repair phase.
- DELETE could check privileges for the wrong database when table aliases were used.
- mysqldump used incorrect syntax for generated column definitions.
- Within a trigger, use of a cursor that accessed OLD or NEW values from a row could cause a server exit.
- If a file was specified using an --init-file option, mysqld --initialize produced errors for statements in the file such as GRANT that affect user accounts.
- Failure during execution of an XA PREPARE statement could result in an invalid XA transaction state. Subsequent attempts to start another XA transaction led to an ER_XAER_OUTSIDE error.
- The audit log plugin could audit accounts named in the audit_log_exclude_accounts system variable.
- If a generated foreign key index was renamed by the same ALTER TABLE statement that added a new foreign key with the same name, the server could exit.
- ALTER TABLE operations that dropped and added the same FULLTEXT index were not performed as in-place (fast) operations that avoid using a temporary copy of the table.
- When the number of days calculated by DATE_FORMAT() function was negative, the server could exit.
- ALTER TABLE operations to add or modify columns could create geometry columns containing invalid data due to missing valida