[This is part 2 of the announcement]
Bugs Fixed, continued
* InnoDB: Write-ahead did not work as expected due to an
incorrectly initialized variable.
Thanks to Yuhui Wang for the contribution. (Bug
#29028838, Bug #93442)
* InnoDB: The base column information for a generated
column was not stored. (Bug #29021730)
* InnoDB: An implicit lock check on secondary indexes
needlessly compared columns using collation rules. (Bug
#29010725)
* InnoDB: Assertion code related to the innodb_flush_method
O_DIRECT_NO_FSYNC setting was no longer valid due to a
recent modification to that setting. Assertion code was
revised. (Bug #29007731)
References: See also: Bug #27309336.
* InnoDB: When starting the server with undo log encryption
enabled, the master key for newly created undo
tablespaces was generated without a server UUID. Undo
tablespaces should use the DefaultMasterKey if the server
UUID is not yet generated. (Bug #29006275)
* InnoDB: Data dictionary code did not check for a returned
data dictionary object, which could potentially cause the
server to exit due to a null pointer access. (Bug
#28977444, Bug #93362)
* InnoDB: An undo tablespace file was left behind by a
failed CREATE UNDO TABLESPACE operation. (Bug #28966457)
* InnoDB: A CREATE UNDO TABLESPACE statement failed on
Windows due to an invalid character in the file name. The
failure resulted in a hang condition due to a missing
OS_FILE_ON_ERROR_NO_EXIT attribute in the call that
creates the undo tablespace file. (Bug #28955676)
* InnoDB: Modifying the value of the
innodb_undo_log_encrypt variable was not a blocking
operation, which could lead to the modification being
reverted by a background thread after the operation
appeared to have been completed successfully. (Bug
#28952870)
* InnoDB: An invalid debug assertion was removed from the
temptable::Handler::primary_key_is_clustered function.
(Bug #28949332)
* InnoDB: An ALTER TABLE ... EXCHANGE PARTITION operation
did not properly update column table_id values in the
data dictionary. (Bug #28927005)
* InnoDB: Memory leaks discovered in the innochecksum
utility were removed. (Bug #28917614, Bug #93164)
* InnoDB: A DDL operation that followed a failed attempt to
create an index on a virtual column resulted in an
assertion failure. (Bug #28825718)
* InnoDB: A performance regression was observed for partial
update operations on compressed BLOBs less than or equal
to 128KB in size. (Bug #28784301)
* InnoDB: Running aggregated queries raised Valgrind
warnings. (Bug #28711717)
* InnoDB: A CHECK TABLE operation raised an assertion
failure. A pointer to a local call stack variable was not
set back to null before a function exit. (Bug #28525110)
* InnoDB: DDL log functions were modified to handle
ER_TOO_MANY_CONCURRENT_TRXS errors. (Bug #28523127, Bug
#92071)
* InnoDB: The purge thread failed to free LOB data pages.
(Bug #28510599)
* InnoDB: Some DDL log table transactions were not rolled
back prior to DDL log recovery. (Bug #28494969)
* InnoDB: A function invoked during SHOW CREATE TRIGGER
processing that retrieves the table name did not perform
the expected lowercase conversion. (Bug #28351038)
* InnoDB: The INFORMATION_SCHEMA.INNODB_FOREIGN TYPE column
reported incorrect values. (Bug #28315651, Bug #91577)
* InnoDB: A Linux AIO handler function failed to check if
completed I/O events succeeded. Thanks to Wei Zhao for
the contribution. (Bug #27850600, Bug #90402)
* InnoDB: An assertion failure was raised in a check that
determines if a transaction holds an implicit lock on a
secondary index. A transaction that does not change the
columns of a secondary index that includes virtual
columns could be incorrectly determined to hold an
implicit lock. (Bug #27491839)
* InnoDB: A function called by a CREATE TABLE thread
attempted to access a table object after it was freed by
a background thread.
Thanks to Yan Huang for the patch. (Bug #27373959, Bug
#89126)
* InnoDB: Two sessions concurrently executing an INSERT ...
ON DUPLICATE KEY UPDATE operation generated a deadlock.
During partial rollback of a tuple, another session could
update it. The fix for this bug reverts fixes for Bug
#11758237, Bug #17604730, and Bug #20040791. (Bug
#25966845)
* InnoDB: When the method used to access a joined table was
const, InnoDB attempted to unlock the matching row
multiple times. (Bug #20939184)
* InnoDB: The INDEX_LENGTH value in
INFORMATION_SCHEMA.TABLES was not updated when adding an
index. (Bug #19811005)
* Partitioning: Some partitioning DDL statements were
improperly rejected due to name validation checks which
used the wrong table identifier. (Bug #29317007)
* Partitioning: While rolling back ALTER TABLE ... COALESCE
PARTITION, the server sometimes attempted to lock and
close partitions which had been dropped as a result of
this statement. (Bug #28517446)
* Partitioning: An AUTO_INCREMENT key added to a
partitioned table by an ALTER TABLE statement using
ALGORITHM=INPLACE restarted on each partition. (Bug
#92241, Bug #28573894)
* Replication: Group Replication did not correctly handle
the situation where START GROUP REPLICATION was issued
when the port specified for the member's Group
Replication local address was currently busy. (Bug
#29347285)
* Replication: If the WAIT_FOR_EXECUTED_GTID_SET() function
was used with a timeout value including a fractional part
(for example, 1.5), an error in the casting logic meant
that the timeout was rounded down to the nearest whole
second, and to zero for values less than 1 second (for
example, 0.1). The casting logic has now been corrected
so that the timeout value is applied as originally
specified with no rounding. Thanks to Dirkjan Bussink for
the contribution. (Bug #29324564, Bug #94247)
* Replication: The consistency level AFTER for the system
variable group_replication_consistency did not include
the consistency guarantees provided by
BEFORE_ON_PRIMARY_FAILOVER. These consistency guarantees,
which were already implicitly present with the BEFORE and
BEFORE_AND_AFTER consistency levels, are now provided
with AFTER. (Bug #29315752)
* Replication: On Debian-based platforms (such as Ubuntu),
if the hostname resolved to 127.0.1.1 - which is the
default on these platforms - it was not possible to
create a cluster using the default settings. Now, in such
situations a proper validation of the instance is
performed before creating a cluster and adding instances
to it. (Bug #29246110)
* Replication: In a blocked group, if you set an invalid
value for group_replication_force_members and then issued
STOP GROUP_REPLICATION, the server could stop
unexpectedly. (Bug #29119961)
* Replication: MySQL Server's behavior when the
mysql.gtid_executed table cannot be accessed has been
refactored to provide appropriate error responses and
actions. The MySQL server now ensures that writes are
permitted to the mysql.gtid_executed table when the
server is in read only or super read only mode, so that
the binary log file can still be rotated in these modes.
If the mysql.gtid_executed table cannot be accessed for
writes, and the binary log file is rotated for any reason
other than reaching the maximum file size
(max_binlog_size), the current binary log file continues
to be used. An error message is returned to the client
that requested the rotation, and a warning is logged on
the server. If the mysql.gtid_executed table cannot be
accessed for writes and max_binlog_size is reached, the
server responds according to its binlog_error_action
setting. If IGNORE_ERROR is set, an error is logged on
the server and binary logging is halted, or if
ABORT_SERVER is set, the server shuts down. (Bug
#29111514)
* Replication: When issuing STOP GROUP_REPLICATION while
the member was trying to assess whether or not it had
lost majority, the server could stop unexpectedly. (Bug
#29053128)
* Replication: When a RESET MASTER TO statement is used to
specify the starting index number for binary log files,
the maximum number that you can specify has been reduced
from the maximum integer value to 2000000000. If the
maximum integer value was specified, the server was not
able to start up because no further binary log files
could be created. The server also previously experienced
a segmentation fault in that situation. (Bug #28980788,
Bug #28995220)
* Replication: On overloaded servers there was a
possibility that when a member joined the group, the
VIEW_CHANGE_LOG_EVENT event which marks that point was
not logged in the correct place. This could lead to
errors in the data transfer to the newly joining server
and data divergence. Now, the VIEW_CHANGE_LOG_EVENT event
is logged in the correct place in the binary log. In
addition, warnings are logged about the delay in logging
the event. (Bug #28971594)
* Replication: On a replication slave with GTIDs in use and
binary logging disabled, an assertion was raised in debug
mode when a DDL statement was filtered out by a table
filter. (Bug #28965972)
* Replication: Two issues with the deserialization of
statement based replication events in the binary log were
corrected. (Bug #28889181, Bug #29028491)
* Replication: If an applier thread was stopped while it
was in the process of opening a table, no error was set,
which could result in a segmentation fault or assertion
depending on the build type. Error handling is now
correctly activated in this situation. (Bug #28864557)
* Replication: When a host name was specified in the IP
address whitelist for Group Replication
(group_replication_ip_whitelist),IPv6 addresses were used
for name resolution and whitelist comparison when an IPv4
address was also available. An IPv4 address should always
be preferred for Group Replication connections. Now, if
the host name resolves to an IPv4 address, any IPv6
addresses are not considered for comparison to the
whitelist. (Bug #28841543)
* Replication: With GTIDs in use on the server, the master
info log on a replication slave was being synchronized
every time the master skipped a transaction using the
auto-skip function. The process ends with a dummy
heartbeat which is sent to the slave and caused a forced
flush to the log, and this could have a large cumulative
impact on the write load on the slave. The same issue
could occur in a circular replication topology with
events that originated from the same server and were
therefore ignored, which were also handled by the slave
with a forced flush to the log. The slave handling code
has now been changed to remove the forced flush for
heartbeat events and for ignored events received through
circular replication, so that the master info log is only
synchronized when appropriate (for example, when a CHANGE
MASTER statement is issued, or the binary log is
rotated). (Bug #28815555, Bug #85158)
* Replication: When an ALTER TABLE statement is used with a
DEFAULT clause to specify an expression default value for
a new column, and the expression default value refers to
a nondeterministic function, the statement is unsafe for
statement-based replication. Previously, such statements
were also evaluated in terms of GTID consistency, which
was not the appropriate check as the statements do not
impact GTID consistency. Now, these statements are
evaluated only for binary logging and are handled
depending on the binary logging format in use. When
binlog_format is set to STATEMENT, the statement is
logged but a warning message is written to the error log.
When binlog_format is set to MIXED or ROW, the statement
is not executed and an error message is written to the
error log. (Bug #28799939)
* Replication: In a replication group configured in
single-primary mode
(group_replication_single_primary_mode=ON, which is the
default), if severe network delays affected the group, it
was possible for the primary and the secondaries to reach
different decisions on a transaction, which could lead to
divergence in the gtid_executed sets on the members. The
issue has now been fixed. (Bug #28768550, Bug #28966455,
Bug #92690)
* Replication: Previously, relay logs could not be rotated
manually for the Group Replication
group_replication_applier channel using the FLUSH RELAY
LOGS statement. Due to this restriction, when encryption
was enabled for binary log files and relay log files
(binlog_encryption=ON), as available from MySQL 8.0.14,
the relay log file in use on that channel could not be
rotated immediately if encryption was disabled again. The
restriction had a similar impact on binary log master key
rotation, as available from MySQL 8.0.16. The restriction
has now been removed, and the FLUSH RELAY LOGS statement
and corresponding internal requests now operate on the
group_replication_applier channel as for any other
channel, with the exception that if the request is
received while a transaction is being applied, the
request is performed after the transaction ends. The
requester must wait while the transaction is completed
and the rotation takes place. This behavior prevents
transactions from being split, which is not permitted for
Group Replication. (Bug #28684376)
* Replication: When you use the
group_replication_force_members system variable to force
a new configuration for a group, the group communication
engine (XCom) now checks that you have not included any
group members that are currently unreachable. If any are
found, the reconfiguration is disallowed and an error is
returned. (Bug #28678845)
* Replication: GRANT statements that were written to the
binary log were logged incorrectly in some cases, which
could result in a GRANT statement that executed
successfully on the master causing an error on the
replication slave. (Bug #28643405, Bug #29155451, Bug
#93750)
* Replication: When a member joined a group on server
start, if the join process failed, for example because
the server was incompatible with the group, there was a
possibility that the offline member could still see
another member as being online. Now, in such a situation
the information shown in the
performance_schema.replication_group_members table is
restricted to the local member when it is OFFLINE. (Bug
#28533993)
* Replication: If a storage engine has the capability to
log in STATEMENT format but not in ROW format, when
binlog_format is set to STATEMENT, an unsafe SQL
statement should be logged and a warning message should
be written to the error log. However, such statements
were instead not executed and an error message was
written to the error log, which is the correct behavior
when binlog_format is set to MIXED or ROW. The issue has
now been corrected so that unsafe statements are logged
with a warning as expected when binlog_format is set to
STATEMENT. (Bug #28429993, Bug #73936)
* Replication: In the event of the recovery channel
failing, unprocessed relay logs were being erased. (Bug
#27940732)
* Replication: It is possible for a replication group
member to go offline briefly, then attempt to rejoin the
replication group again before the group has detected its
failure and been reconfigured to remove the member.
Previously, in this situation, the rejoining member could
participate in XCom's consensus protocol if it received
and processed messages intended for its pre-crash
incarnation. This could cause XCom to deliver different
values for the same consensus round, because the
rejoining member could make a different decision before
and after failure. To prevent this situation, a rejoining
member now ignores messages intended for its pre-crash
incarnation. (Bug #27383487)
* Replication: A replication group member could trigger a
local view after being expelled from the group due to a
loss of majority. This resulted in a message incorrectly
stating that the member had resumed regular operation
after the expulsion. Group Replication now checks before
delivering a local view that the member has not been
expelled. (Bug #27349236)
* Replication: If an invalid value was specified for the
group_replication_communication_debug_options system
variable, the Group Communication System set its
corresponding internal variable to GCS_DEBUG_NONE, and
the server returned the invalid value for a SHOW
VARIABLES query. The value of the system variable is now
checked during server initialization, and if an invalid
value was specified, an error message is logged and Group
Replication does not start automatically. (Bug #26729404)
* macOS: CMake 3.12.4 or higher (which forces
UseModernBuildSystem = NO) is now required on macOS if
building with Xcode rather than Makefiles. (Bug
#28893131)
* Microsoft Windows: Validity testing for the
named_pipe_full_access_group system variable did not
account for NULL values. (Bug #29256690)
* Microsoft Windows: When multiple instances of mysqld were
started with the --no-monitor option on the same host for
same user, the SHUTDOWN command shut down the wrong
server process. This fix creates a unique shutdown event
name for use with --no-monitor by appending the process
ID of the process. (Bug #28723675)
* JSON: The JSON path parser now propagates errors in the
same way as most other components of the MySQL server,
returning true on error and false on success. (Bug
#28851426)
* JSON: Removed an unneeded type lookup in
Json_wrapper::get_datetime(). (Bug #28851324)
* The authentication_ldap_simple plugin could enforce
authentication incorrectly. (Bug #29637712)
* RPM package obsoletes were updated to enable successful
upgrades from MariaDB to MySQL on EL8. (Bug #29413354)
* It was possible for the result of an outer join to
contain a non-NULL row where a NULL extended row was
expected. (Bug #29402481)
References: This issue is a regression of: Bug #27808758.
* SET PASSWORD FOR ... could not be executed as a prepared
statement. (Bug #29387041, Bug #94416)
* Builds under Visual Studio could fail while building
MySQL Router libraries. (Bug #29382197)
References: This issue is a regression of: Bug #29361890.
* Imported foreign keys did not work if defined before the
referenced table. (Bug #29379078, Bug #94400)
References: This issue is a regression of: Bug #28493257.
* When an error was raised while evaluating the ESCAPE
clause of a LIKE expression at resolve time, the error
status was not propagated to the caller. (Bug #29368521)
* Tablespace files for the innodb_table_stats_backup and
innodb_index_stats_backup metadata backup tables were not
removed after an in-place upgrade from MySQL 5.7 to MySQL
8.0. (Bug #29365552)
* While flattening a subquery, if a predicate which was
always false was present, the MySQL Optimizer did not
perform any kind of transformation, resulting in a
subquery that was not prepared, and which asserted later
when it was executed. To resolve this issue, when such a
predicate is present in a subquery, the subquery's query
expression is now unlinked from the query block. (Bug
#29356132)
* Upgrade from MySQL 8.0.11, 8.0.12, or 8.0.13 to MySQL
8.0.14 or 8.0.15 failed if an event, routine, or trigger
was defined with the ALLOW_INVALID_DATES SQL mode. The
SQL mode identifier in the data dictionary was changed in
MySQL 8.0.14, causing a migration failure. (Bug
#29350955)
* RPM builds ignored the WITH_SSL configuration setting.
(Bug #29347534)
* Length metadata for the TO_SECONDS() function was not
always calculated correctly. (Bug #29321387)
* Conditions using windowing functions removed due to being
always true or false were not always handled correctly.
(Bug #29320484)
* SET ROLE statements could leak memory. (Bug #29304583)
* Upgrading from MySQL 5.7 to MySQL 8.0 on Windows failed
with "Error 197 from SE while migrating tablespaces". The
error was due to an access share violation that occurred
when attempting to open a tablespace file. (Bug
#29292860)
References: This issue is a regression of: Bug #28642608.
* The COMPILATION_COMMENT_SERVER value could be incorrect
in RPM packages. (Bug #29284651)
* CREATE TABLE IF NOT EXISTS failed even if the table
already exists if the new table definition had no primary
key and the sql_require_primary_key system variable was
enabled. (Bug #29283055, Bug #94134)
* A delete from a partitioned table from which all
partitions had been pruned away was not always handled
correctly. (Bug #29280186)
* The CMake check for the GNU gold linker could fail with
Clang. (Bug #29278244)
* The data dictionary version was incremented for the MySQL
8.0.16 release. (Bug #29278241)
* An argument of the wrong type to a function used with
DISTINCT was not always handled correctly. (Bug
#29277571)
* Length metadata for the QUOTE() function was not always
calculated correctly. (Bug #29276074)
* When evaluating GREATEST() or LEAST(), MySQL checked for
correct signedness of the return value before checking it
for NULL. (Bug #29275835)
* When strict SQL mode was not in effect, the values of
some string functions returning NULL to indicate a result
greater than max_allowed_packet were handled
inconsistently behaviour, which could result in
incorrectly sorted output and possibly other
misbehaviors. (Bug #29272683)
References: See also: Bug #97301, Bug #29133127.
* Logic that checks if upgrade is supported to a particular
server version was inverted to check for server versions
from which upgrade is not supported. (Bug #29270297)
* Event creation could store an incorrect repetition
interval. (Bug #29269819, Bug #94085)
* A WHERE condition containing a view reference or an item
created by a transformation was not always handled
correctly. (Bug #29268867, Bug #29268698, Bug #28723669,
Bug #29244238)
* mysql_ssl_rsa_setup failed to compile using GCC 9. (Bug
#29245251)
* If CMake finds a libtirpc library that is too old to work
with MySQL, it tries to use Sun RPC from glibc instead.
(Bug #29240701)
* The server could fail to write slow queries to the slow
query log due to incorrect calculation of query execution
time. (Bug #29232684, Bug #93963)
* MySQL regular expression functions using positions
employed internal indexes based on 16-bit chunks, rather
than on codepoint positions. (Bug #29231490)
* On Windows, the MySQL MSI installer could fail to
correctly detect whether the Visual Studio 2015
Redistributable was installed. (Bug #29227209)
* SDI JSON files did not include the m_hidden field of
Index_impl objects. This made it hard to use the SDI JSON
to recreate the CREATE statement for the table because
InnoDB adds a number of hidden indexes. The SDI JSON now
includes the m_hidden field. This changes SDI format, so
the SDI version number was increased to the current
server version number. (Bug #29210646, Bug #93914)
* The position hint for the last row in a range frame was
updated to be one row past the actual last row in the
frame. (Bug #29201831)
* In a column definition, multiple constraint definitions
were not accepted when the first was a CHECK constraint.
(Bug #29191994)
* Error log information buffered during startup could be
buffered too long if the server was performing an
upgrade. (Bug #29189532)
* For nullable columns, if we find an expression that is
always true except when the column is NULL, the
expression is folded to column IS NOT NULL. When such an
expression was nested, this caused NULL rows to be
selected in error. To prevent this from happening, such
an expressions when nested is now instead folded to
IF(column IS NULL, NULL, TRUE). (Bug #29179604)
* PERIOD_ADD() did not handle values greater than 32 bits
in length for the period argument correctly on Windows
platforms. (Bug #29175262)
* When sql_auto_is_null is enabled, a WHERE clause of the
form WHERE auto_increment_col IS NULL is rewritten as
WHERE auto_increment_col = LAST_INSERT_ID(). This
transformation was only performed once per
auto-incremented value, which made it difficult to know
in advance whether the transformation would be performed.
Now the transformation is performed unconditionally
whenever sql_auto_is_null is enabled.
In addition, the value returned by LAST_INSERT_ID() is
now treated as unsigned, fixing a failure to match an
auto-incremented value outside the range of a signed
BIGINT. (Bug #29171668)
* The harness_plugin_eventlog declaration caused
compilation errors in some build environments. (Bug
#29160214)
* The logic for truncating or extending a decimal constant
to the desired number of fractional digits during
constant folding was lacking. Extra trailing zeros in the
fraction could trigger an attempt to widen the faction,
since the decision to widen was based on the number of
non-zero fractional digits, rather than on total number
of fractional digits, leading to an assert (in debug
builds) in the internal function widen_fraction(). This
issue is fixed by identifying where it is possible merely
to truncate excess trailing zeroes. In this case, it is
not necessary to adjust the comparison operator, and the
constant can be replaced by one having fewer trailing
zeroes instead. (Bug #29155439)
* On Windows, the internal function
get_mysql_time_from_str_no_warn() did not always perform
proper error checking. (Bug #29155126)
References: See also: Bug #29175262.
* Under certain conditions, RENAME TABLE statements that
renamed the same table multiple times could raise an
assertion or cause a server exit. (Bug #29140407)
* For debug builds, starting the server with
--event-scheduler=DISABLED could result in an assertion
being raised for certain events. (Bug #29140298, Bug
#93719)
* In debug builds, When strict SQL mode was not in effect,
the CONCAT() and CONCAT_WS() functions raised an
assertion if the result was longer than
max_allowed_packet. (Bug #29133127)
* An out-of-range fractional part could produce incorrect
timestamps in SET SESSION timestamp statements. (Bug
#29120569, Bug #93600)
* The mysql_service_component_sys_variable service could
access component system variables but not server or
plugin system variables. (Bug #29113463)
* ALTER TABLE ... CONVERT TO CHARACTER SET could produce a
memory-access error. (Bug #29058369, Bug #93603)
* The mysql.tablespaces.name column limit was 259 bytes,
which was less than required for permitted identifier
lengths. The column limit was raised to 268 bytes. (Bug
#29053560, Bug #93587)
* For debug builds, for spatial computations that raised an
assertion, partition-handling code ignored the error,
resulting in a server exit. (Bug #29047811)
* A zero length LOB that was stored externally caused an
assertion failure. (Bug #29047795)
* Handling of the COMPILE_DEFINITIONS and COMPILE_FLAGS
CMake options was adjusted to avoid cross-compiling
failures. (Bug #29041100)
* The server could try to read the first diagnostics area
message even when the max_error_count system variable was
set to zero, resulting in a memory-access error. (Bug
#29031684)
* When comparing DATE values with constant strings, MySQL
first tries to convert the string to a DATE and then to
perform the comparison. When the conversion failed, MySQL
executed the comparison treating the DATE as a string,
which could lead to unpredictable behavior. Now in such
cases, if the conversion of the string to a DATE fails,
the comparison fails with ER_WRONG_VALUE. (Bug #29025656)
* With the --users option, mysqlpump wrote CREATE USER and
GRANT statements to the output, but too late to apply to
the other objects created by the dump. Consequently,
restoring the dump file created the user accounts too
late to apply to other objects created by the file.
mysqlpump now writes user accounts to the dump file
before other objects. (Bug #29023216)
* For syntax errors at the position of the WITH keyword,
the parser error message identified the problem at the
incorrect location. (Bug #29022263)
* MySQL builds using recent versions of the International
Components for Unicode (ICU) now return
ER_REGEXP_ILLEGAL_ARGUMENT rather than the generic error
ER_REGEXP_ERROR for malformed regular expressions. (Bug
#29016798)
* On Windows, the LDAP library could not be found if the
file system was case sensitive. (Bug #29016220)
* Values selected from the TABLE_COMMENT column of the
INFORMATION_SCHEMA.TABLES table could be truncated. (Bug
#29014272, Bug #93451)
* Data layer memory leaks related to unfreed schemas were
discovered in ASAN and Valgrind builds. (Bug #29008688)
* A GROUP BY query with ROLLUP incorrectly raised
ER_WRONG_FIELD_WITH_GROUP when used with an ORDER BY
clause that contained an expression.
The same issue also led to incorrect ordering of the
result from a GROUP BY query with ROLLUP when ordering on
a function. (Bug #29006668, Bug #29054096)
* The parser accepted multiple COLLATE clauses in generated
column definitions. It now accepts a single COLLATE
clause. (Bug #28997518)
* INSERT ... ON DUPLICATE KEY UPDATE could perform
incorrect updates. (Bug #28995498, Bug #93410)
References: This issue is a regression of: Bug #26188578.
* A damaged mysql.user table could cause a server exit.
(Bug #28986737)
* Reading rows from the Performance Schema data_locks or
data_lock_waits table could return an unbounded number of
rows during a scan, resulting in a server exit due to
excessive memory allocation. This situation now produces
an error. (Bug #28977428, Bug #87748)
* Calculating the cost of materialization for a LATERAL
join on an empty table led to an assertion. Now in such
cases, the query cost is reported as zero. (Bug
#28976533)
* For debug builds, an invalid utf8 character in the
comment string for a column type in CREATE or ALTER DDL
statements raised an assertion. (Bug #28972424, Bug
#93321)
* The CMake check for tirpc headers now falls back to using
pkgconfig, to enable finding the headers on more Linux
platforms. This requires that pkgconfig be installed.
(Bug #28970313, Bug #93341, Bug #28997093)
* mysqld could undergo an unplanned shutdown when a
component in the WHERE clause of a subquery was
non-deterministic, in this particular case due to
referencing a user variable that was also set in the same
statement. This issue is fixed by ignoring predicates
having non-deterministic components.
In addition, since non-deterministic subqueries as a
general rule should be evaluated per outer row, these
should not be targets for materialization strategy. This
is prevented by not decorrelating the non-deterministic
predicates. When possible, the same strategy selection
for non-semijoin and non-deterministic subqueries is also
now enforced. (Bug #28970261)
* ALTER TABLE statements to change a table storage engine
could hang when user-level locks and explicit table locks
had been acquired earlier. (Bug #28966941)
* COUNT() with LIMIT ... OFFSET returned a different result
from that obtained with other aggregate functions used
with this clause. (Bug #28961843)
* Extracting the value of the MYSQL_HOME environment
variable could change the value of the variable stored
following MYSQL_HOME in the environment. (Bug #28960613)
* Opening and closing a connection multiple times in a
client program could cause the client to exit due to
improper character set reinitialization. (Bug #28956360,
Bug #93276)
* Item_subselect::walk_body() now walks FROM clauses. (Bug
#28955358)
* An issue was uncovered in query plan for a query with the
following two IN subqueries:
+ A subquery with a derived table containing an outer
reference to the topmost query
+ A subquery with a derived table, which was not
outer-correlated, was semi-joined, and was not
lateral.
For the second subquery: calculation of the map of
dependencies neglected to exclude irrelevant lateral
tables, and wrongly included dependencies of the derived
table in the first subquery. (Bug #28954838)
* An error generated while evaluating a constant expression
in certain GROUP BY queries was not checked for in a
timely fashion. (Bug #28949452)
* When comparing a DATE or DATETIME value with a string
MySQL first tries to convert the string to the same type
(DATE or DATETIME) and then to compare the two as values
of that type. If this conversion fails, MySQL raises an
Incorrect date value warning and then falls back to
comparing the values as strings, which is the expected
behavior. In some cases, even though the converted string
did not reflect a valid DATE or DATETIME value, no
warning was issued so and it was compared with the DATE
or DATETIME value as a value of that type. Now in such
cases, the converted string is always checked for
validity as a DATE or DATETIME before being compared,
unless the ALLOW_INVALID_DATES server SQL mode has been
set explicitly. (Bug #28940878)
* Long client host names could cause unexpected server
behavior. (Bug #28936359)
* Substitution of a large number of parameters in a single
prepared statement could lead to excessive and
unnecessary reallocation of memory. This is fixed by
having String::replace() use exponential buffer growth as
String::append() does. (Bug #28929977)
* The data dictionary cache could become out of sync with
data dictionary contents due to failure to check the
result of a transaction-related operation. (Bug
#28923782, Bug #93196)
* Determination of the number of online CPUs available to
the mysqld process is now more accurate. Thanks to Daniel
Black for the contribution. (Bug #28907677, Bug #93144)
* The INFORMATION_SCHEMA.TABLES TABLE_COMMENT column
reported the wrong error message due to an invalid view.
(Bug #28901919)
* After upgrading, the INFORMATION_SCHEMA.TABLES view
reported "View 'view_name' references invalid table(s) or
column(s) or function(s) or definer/invoker of view lack
rights to use them" for several sys schema views. The
upgrade process did not update the referencing view
state. (Bug #28901821)
* Creating histograms for large tables (millions of rows)
having many distinct column values took an excessive
amount of time.
The fix make this process marginally slower for data sets
with few distinct values and consumes slightly more
memory than previously, but for large data sets with many
distinct values it is significantly faster. (Bug
#28888936)
* Upgrade from MySQL 5.7 to MySQL 8.0 failed for instances
with a an InnoDB 4K page size due a key length error that
should be suppressed during upgrade. (Bug #28884503)
* An incorrectly configured keyring plugin could cause a
server exit. (Bug #28876033)
* Two issues were found in cases involving use of the
COUNT() function:
+ When the argument passed to COUNT() was a nullable
expression, it was possible for the function to
return an incorrect result.
+ A query using COUNT() together with an EXISTS
subquery returned an incorrect result.
(Bug #28857990, Bug #29240516)
* Commands for invoking ccache during MySQL builds were
generated incorrectly. (Bug #28841612, Bug #92927)
* Previously, for INSERT statements that listed insert
column names multiple times, only the first duplicated
name was reported. Now, all duplicated names are
reported. (Bug #28836669)
* When using ROLLUP, the GROUP BY may yield NULL in some
rows. Expressions referencing any of these rows could
lose the NULL, producing an incorrect result. (Bug
#28836345)
* After a failed attempt to open a table, attempts to
discover the table from the storage engine did not check
for an error return, which could cause the server to hang
if an error occurred. (Bug #28828450)
* The server could exit when trying to drop a user who had
been granted a particular role. (Bug #28817441)
* MySQL Router compilation failed if MySQL was configured
with -DWITH_MYSQLX=0. (Bug #28811356)
* A function called during data dictionary upgrade
temporarily modified the avoid_temporal_upgrade parameter
so that a check for old temporal types is always
performed. Because multiple threads could call the
function concurrently, a check was added to ensure that
the parameter value could be safely modified during data
dictionary upgrade. (Bug #28805429)
* An EXISTS subquery is converted to a semi-join operation.
The optimizer chooses a materialization lookup strategy
for this semi-join, but because the subquery is not
correlated with the outer query block, there were no keys
to use for the lookup, which caused the parent query to
fail. To solve this issue, we use two equal constant
items as keys, to ensure that the materialized query gets
the constant as a key (and so that the materialized table
consists of at most one row). (Bug #28805105)
* The Last_query_cost status variable now shows the cost of
queries which contain multiple blocks such as subqueries
or unions. (Previously, this variable was set only for
simple queries consisting of a single query block.) (Bug
#28786951)
* The message displayed by mysqld when a fatal signal
occurred has been simplified and made more informative.
It also does not display the calculation of memory usage,
which had become out of date and inaccurate. (Bug
#28773322, Bug #92731)
* The skip_name_resolve system variable could be persisted
using SET PERSIST_ONLY to enable it, but not to disable
it. (Bug #28749668)
* Removed obsolete and unnecessary condition pushdown
handling in iterators except where required by single
table update and delete queries. (Bug #28745859)
* IS NOT NULL predicates are added as part of early
filtering of NULL for ref access performed by the
Optimizer. For queries having a star-join topology,
redundant duplicates of these predicates were added,
which made evaluation of affected WHERE conditions less
efficient. (Bug #28727717)
* CMake could generate an incorrect order of system
includes when MySQL was configured to use the bundled
version of a system library. (Bug #28727631, Bug #92615)
* Some numeric operations involving double to long long
conversions could return different results on Windows and
Linux. (Bug #28706832)
* An attempt to access a null pointer could occur during
prepared statement execution. (Bug #28692136)
* FROM_UNIXTIME() returned an out-of-range value if passed
an argument that, when rounded up, exceeded the epoch
value. Now it returns NULL as for other out-of-range
values. (Bug #28671811, Bug #92501)
* A replicated DDL operation could result in a 'duplicate
entry on primary key' error during recovery on servers
started with the --slave-parallel-workers option. (Bug
#28670843)
* Empty host names in accounts could cause the server to
misbehave. (Bug #28653104)
* The MySQL 5.7 to MySQL 8.0 upgrade process did not check
for the existence of tablespace files, which could result
in an inconsistent data dictionary. (Bug #28642608)
* Type casting of ENUM behaved differently in subqueries
than not in subqueries. (Bug #28547906, Bug #92173)
* Statements that caused truncation of floating-point
values could be executed as prepared statements even when
the server SQL mode included STRICT_TRANS_TABLES. In
addition, if the binary logging mode was MIXED, such
statements were propagated to the slave, where they
caused errors. (Bug #28546855)
* Reducing the value of thread_cache_size at runtime did
not reduce the size of the thread cache. (Bug #28508923,
Bug #92024)
* Some error messages still referred to the PASSWORD()
function, which has been removed. (Bug #28498714)
* The Aborted_connects status variable was not incremented
for unsuccessful connection attempts, if connections were
managed by the thread_pool plugin. (Bug #28490126)
* mysqladmin shutdown did not wait for mysqld to shut down.
(Bug #28466137, Bug #91803)
References: This issue is a regression of: Bug #25364806.
* Specifying CURRENT_USER as the user in GRANT statements
could fail. (Bug #28454014)
* Repeated invocations of stored procedures which executed
queries undergoing short-circuit evaluation were not
always handled correctly. (Bug #28379655)
* Keyring migration should require only read access to the
source keyring, but failed unless the user had write
access. (Bug #28339014)
* If a user performing a keyring migration did not have
write access to the keyring file, the migration failed
but reported success in its final error log message. (Bug
#28330922)
* During FLUSH STATUS execution, the Performance Schema
unnecessarily aggregated session status to global status,
causing double counts for some status variables. (Bug
#28291258, Bug #91541)
* mysql_secure_installation no longer attempts to read a
password from the .mysql_secret file. This was was
created by mysql_install_db, a program that has been
removed. (Bug #28235716, Bug #91270)
* Some status variable values could temporarily increase
before returning to their original value. (Bug #27839644,
Bug #90351)
* Client programs did not exit if --ssl-fips-mode was given
but FIPS mode could not be set to the specified option
value. (Bug #27809371)
* Executing ALTER INSTANCE ROTATE INNODB MASTER KEY and
migrating keys from the keyring_file plugin to the
keyring_encrypted_file plugin could make encrypted tables
unusable. (Bug #27760952)
* When upgrading from MySQL Community to Commercial, the
root password prompt was shown even when the data
directory existed. (Bug #27741998)
* In the client/server protocol, malformed packets for
prepared statements could go undetected and cause an
assertion to be raised. (Bug #27627731)
* Changes were made in session connect and disconnect
handling for threads in order to speed up query
throughput when using the X Plugin. Previously, a session
(Srv_session) was attached to and detached from the
current thread with every command; now, this thread
switch takes place only when disconnecting the current
session, or when the current thread and the thread being
attached to belong to different plugins. (Bug #27463277)
* The PERSIST and PERSIST_ONLY keywords were reserved by
mistake. They are now nonreserved keywords. (Bug
#25220656)
* Installing and uninstalling a plugin concurrently with
client connection activity could cause a server exit.
(Bug #22980441)
* The LOAD_FILE() function could fail for files for which
stat() should be considered only advisory, such as files
under /proc. (Bug #18394503, Bug #72027)
* A query using an ORDER BY clause failed silently when the
table being queried had an implicit full-text index and
the sort buffer was of insufficient size to contain the
sorted keys. (Bug #93241, Bug #28940361)
* A UDF returning a string value now sets an explicit
return type. Depending on the arguments passed to the
function, this is one of VARCHAR, MEDIUMBLOB, or
LONGBLOB. (Bug #92890, Bug #28828169)
* Made a comparison in the internal method
Item_result::item_cmp_type() more efficient. Our thanks
to Daniel Black for the contribution. (Bug #92784, Bug
#28796107)
* A windowing function employed in an arithmetic expression
produced an incorrect result when the query containing it
used DISTINCT. (Bug #92503, Bug #28672483)
* Some queries involving complex joins leaked file handles.
(Bug #90902, Bug #28039829)
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe: http://lists.mysql.com/mysql
