btr_cur_pessimistic_insert(): Relax a too strict debug assertion that
would fail when the function is invoked by btr_cur_pessimistic_update()
during innobase_add_instant_try(), that is, when updating the hidden
metadata record during a subsequent ADD COLUMN operation involves
splitting the leftmost clustered index leaf page.

This is a partial backport of 301bd62b
from 10.4.

Cause:
* row_start != 0 treated as it exists. Probably, possible row permutations had not been taken in mind.

Solution:
* Checking both row_start and row_end is correct, so versioned() function is used

MDEV-18875 Assertion `thd->transaction.stmt.ha_list == __null || trans == &thd->transaction.stmt' failed or bogus ER_DUP_ENTRY upon ALTER TABLE with versioning

Cause:
* when autocommit=0 (or transaction is issued by user),
 `ha_commit_trans` is called twice on ALTER TABLE, causing a duplicated
 insert into `transaction_registry` (ER_DUP_ENTRY).

Solution:
* ALTER TABLE makes an implicit commit by a second call. We actually
 need to make an insert only when it is a real commit. So is_real
 variable is additionally checked.

Instead of trying to detect MinProtocol=TLSv1.2 in all possible
sections of files in all possible locations where openssl is configured
to find its configuration in all possible distros (and these locations
are mutually different in Debian, Ubuntu, and RHEL, btw) - just make
mtr to instruct openssl not to read any config files at all.

This was broken since commit 193b5ed5

The function was only called by innobase_drop_tablespace(),
which was removed in commit 494e4b99
and added in commit 2e814d47.

The issue here is for degenerate joins we should execute the window
function but it is not getting executed in all the cases.

To get the window function values window function needs to be executed
always. This currently does not happen in few cases
where the join would return 0 or 1 row like
  1) IMPOSSIBLE WHERE
  2) MIN/MAX optimization
  3) EMPTY CONST TABLE

The fix is to make sure that window functions get executed
and the temporary table is setup for the execution of window functions

innodb.cmake: restrict -fno-builtin-memcmp for GCC versions older that 4.6
where optimization issue was solved.

Add proper wait condition when provider options are restored.

MDEV-21189 : Dropping partition with 'wsrep_OSU_method=RSU' and 'SESSION sql_log_bin = 0' cases the galera node to hang

Test cleanup. Best practice for using RSU, is to isolate the node
up-front, so this test did not reflect real world scenario

MDEV-21189 : Dropping partition with 'wsrep_OSU_method=RSU' and 'SESSION sql_log_bin = 0' cases the galera node to hang

Test cleanup. Best practice for using RSU, is to isolate the node
up-front, so this test did not reflect real world scenario

- post-push to fix the compilation issue

- Moved the recv_sys->heap memory condition inside recv_parse_log_recs().
So that, InnoDB can mark the status as STORE_NO earlier.

- InnoDB uses one third of buffer pool chunk size for reading the redo
log records. In that case, we can avoid the scenario where buffer ran
out of memory issue during recovery.

It's not safe to chenge the THD::thread_id.
So send the thread_id as an argument to the mysql_audit_external_lock.

in 10.1+ one should use

MY_CHECK_AND_SET_COMPILER_FLAG("-Wno-address-of-packed-member")

and it's already done in storage/tokudb/PerconaFT/CMakeLists.txt

A debug assertion that was added in
commit ed0793e0
turns out to be too strict. In the test innodb_gis.rtree_compress,4k
the function is sometimes being invoked by purge for a
spatial index root page that is not a leaf page (PAGE_LEVEL is 1).

If we manually poison memory region, ASAN requires that we manually unpoison
it too. We didn't do it and munmap() didn't do it too. That was the issue.

os_mem_free_large(): unpoison memory region. Just in case.

https://github.com/google/sanitizers/issues/182
https://github.com/llvm/llvm-project/blob/86acaa9457d3957cbe303e1e801f1e727f66ca89/compiler-rt/include/sanitizer/asan_interface.h#L21

Item_ref::val_(datetime|time)_packed() erroneously called
(*ref)->val_(datetime|time)_packed().

- Fixing to call (*ref)->val_(datetime|time)_packed_result().
- Backporting Item::val_(datetime|time)_packed_result() from 10.3.
- Fixing Item_field::get_date_result() to handle null_value in the same
  way how Item_field::get_date() does.

do not fallback to malloc(), always return properly aligned buffer

One may not call memcpy(dst, src=NULL, size), even if size==0.

Remove Query_tables_list::lock_tables_state - it is not used and it causes
errors like this:

sql_lex.h:1675:7: runtime error: load of value 2779096485, which is not a
valid value for type 'enum_lock_tables_state'

Fix wrong typecast

Problem was that wsrep_on was OFF.

Problem was that wsrep_on was OFF.

This is 10.4 version.

Fixed the condition for waking up/creating another thread.

If there is some work to do (if the request queue is not empty),
a thread should be woken or created.

The condition was incorrect since 18c9b345