trx_t::commit_in_memory(): Invoke mod_tables.clear().

trx_free_at_shutdown(): Invoke mod_tables.clear() for transactions
that are discarded on shutdown.

Everywhere else, assert mod_tables.empty() on freed transaction objects.

Let us calculate the hash table cell address while we are calculating
the latch address, to avoid repeated computations of the address.
The latch address can be derived from the cell address with a simple
bitmask operation.

Introduced by 85828b8f

This is running 2 git processes in parallel, which, if unlucky can cause
either of them to fail with "File already exists" error.

Given PASSWORD EXPIRE and PASSWORD EXPIRE [NEVER|INTERVAL x DAY] are
two different mechanisms, SHOW CREATE USER should display all the
information required to restore the state of an account which
includes both a manual expired state and an automatic policy.

The solution proposed here keeps a CREATE USER ... PASSWORD EXPIRE
statement and adds an aditional
ALTER USER .. PASSWORD EXPIRE [NEVER|INTERVAL x DAY] when necessary

This way a tool can restore almost the complete state of an account
as it was before a dump. The only information left still is the
value of the password_last_changed column from mysql.global_priv

- The commit 5fd3c747(MDEV-24709)
resets the recv_no_ibuf_operations in
recv_recovery_from_checkpoint_start(), but InnoDB fails to reset
the variable recv_no_log_write() during that time and that leads
to the assert failure.

This was caused by two different bugs:
1) Information_schema tables where not locked by lock_tables, but
   get_lock_data() was not filtering these out. This caused a crash when
   mysql_unlock_some_tables() tried to unlock tables early, including
   not locked information schema tables.

Fixed by not locking SYSTEM_TMP_TABLES

2) In some cases the optimizer will notice that we do not need to read
   the information_schema tables at all. In this case
   join_tab->read_record is not set, which caused a crash in
   get_schema_tables_result()

Fixed by ignoring const tables in get_schema_tables_result()

1. wait for the binlog thread to reach the certain state, don't use
   a debug_sync that's incorrectly placed to detect the state
2. no need to do a (non-deterministic) `show binlog events` to verify
   what is guaranteed by the directly preceding line

if it's the whole content of a test anyway.

add a new "debugger" to mtr, that runs the executable
under valgrind in gdb. valgrind pid is auto-detected,
but the delay (sleep) and vgdb path are hard-coded for now

"debugger" is anything that wraps execution of a target
binary (mysqld or mysqltest). Currently the list includes:
gdb, ddd, dbx, lldb, valgrind, strace, ktrace, rr,
devenv, windbg, vsjitdebugger.

for every debugger xxx, mtr will recognize four options:
--xxx, --boot-xxx, --manual-xxx, --client-xxx.
They all support an optional "=string" argument. String
being a semicolon-separated list of commands (e.g. for gdb)
or one (not semicolon-separated) command line of options
(e.g. for valgrind). Or both (e.g. --gdb='-quiet -nh;info files'

In embedded both --xxx and --client-xxx work.

Functionality changed/removed:
* --rr-args is gone
* --rr-dir is gone
* --manual-debug is gone
* --debugger={devenv|vc|windbg|vc_express|vsjitdebugger} is gone
* --strace-option is gone
* --stracer={strace|ktrace} is gone
* --valgrind only enables it for the server, not for everything
* --valgrind-all is gone
* --valgrind-mysqltest is gone
* --valgrind-mysqld is gone
* --valgrind-options is gone
* --valgrind-option is gone
* --valgrind-path is gone
* --callgrind is gone
* one cannot combine --valgrind --gdb anymore
* valgrind report doesn't add a fake test line to the output
* vc and vcexpress on windows are no longer supported

don't allocate Column_statistics_collected objects that won't
be used.

minor style fixes (StringBuffer<>, etc)

only collect persistent stats for columns explicitly listed
by the user in the  ANALYZE TABLE PERSISTENT FOR COLUMNS (...)
clause. The engine can extend table->read_set as much as
it wants, it should not affect the collected statistics.

Test case from the 3b94309a applies - it used to crash,
because ha_partition extended table->read_set after the loop that
initialized some objects based on bits in the read_set but before the
loop that used these objects based on bits in the read_set.

This reverts the commit 3b94309a but keeps the test

Because the fix is a hack that isn't supposed to do anything,
and relies on a side-effect of rnd_init inside ha_partition.

A different fix is coming up.

           on used subqueries

If a query was based on a table value constructor that contained subqueries
then EXPLAIN for such query did not contain any lines explaining the
execution plans of the subqueries.
This happened because
- no optimize() method was called for any subquery used by the table value
  constructor when EXPLAIN command for the query was processed;
- EXPLAIN node created for the table value constructor itself did not
  assume that some child nodes could be attached to it.

Approved by Oleksandr Byelkin <sanja@mariadb.com>

- This is caused by commit ad6171b9
(MDEV-22456). InnoDB reloads the evicted table again from dictionary.
In that case, AHI entries and current index object mismatches
happens. When index object mismatches then InnoDB should drop
the page hash AHI entries for the block. In
btr_search_drop_page_hash_index(), InnoDB should take exclusive
lock on the AHI latch if index is already freed to avoid the
freed memory access during buf_pool_resize()

Two new features for Galera
* Write a warning to error log if Galera replicates table with storage engine not supported by Galera (at the moment only InnoDB is supported
** Warning is pushed to client also
** MyISAM is allowed if wsrep_replicate_myisam=ON
* Write a warning to error log if Galera replicates table with no primary key
** Warning is pushed to client also
** MyISAM is allowed if wsrep_relicate_myisam=ON
* In both cases apply flood control if > 10 same warning is writen to error log
(requires log_warnings > 1), flood control will suppress warnings for 300 seconds

This can cause crashes when accessing already released memory

The issue was the Item_default created a internal field, pointing to
share->default_values, to be used with the DEFAULT() function.
This does not work for BLOB fields as these are freed at end of query.
Fixed by storing BLOB field data inside and area allocated by
Item_default_value,  like we do for nondeterministic default values.

We have innodb_use_native_aio=ON by default since the introduction of
that parameter in commit 2f9fb41b
(MySQL 5.5 and MariaDB 5.5).

However, to really benefit from the setting, the files should be
opened in O_DIRECT mode, to bypass the file system cache.
In this way, the reads and writes can be submitted with DMA, using
the InnoDB buffer pool directly, and no processor cycles need to be
used for copying data. The use of O_DIRECT benefits not only the
current libaio implementation, but also liburing.

os_file_set_nocache(): Test innodb_flush_method in the function,
not in the callers.