MDEV-10180 main.events_restart fails on valgrind builder: void THD::inc_thread_count(): Assertion `!abort_loop' failed.

remove an overly strong assert

Replaced hard-coded cache line size.

Changes based on Daniel Black's work.

Bug #79636: CACHE_LINE_SIZE should be 128 on AArch64
Bug #79637: Hard-coded cache line size
Bug #79638: Reconcile CACHE_LINE_SIZE with CPU_LEVEL1_DCACHE_LINESIZE
Bug #79652: Suspicious padding in srv_conc_t

- changed CPU_LEVEL1_DCACHE_LINESIZE to default to 128 bytes on POWER
  and AArch64 architectures in cases when no value could be detected
  by CMake using getconf

- changed CACHE_LINE_SIZE definition in ut0counter.h to be an alias of
  CPU_LEVEL1_DCACHE_LINESIZE

- changed a number of hard-coded 64-byte cache line size values in the
  InnoDB code

- fixed insufficient padding for srv_conc members in srv0conc.cc

Ported to Mariadb by Daniel Black <daniel.black@au.ibm.com>
Added s390 cache size of 256 at same time.

that fixes many rpl tests failures

move thread_count handling into THD:
* increment thread_count in THD constructor
* decrement thread_count in THD destructor
* never modify thread_count directly!

because thread_count means just that: number of THDs
and shutdown code looks at it to know when to free
shared data structures that THD uses.

This fixes random crashes in ~THD on shutdown

it's always assert_<statement>,
not <do something>_if_<condition>

between net_end() and ha_close_connection()

this fixes lockups in main.locked_temporary-5955 and some
other innodb tests.

this also fixes tokudb.type_year failure
(in ./mtr tokudb.type_timestamp tokudb.type_year)

according to the standard, the gtrid/bqual pair must be
"globally unique", so XIDs are equal if their gtrids and
bquals are identical, formatID simply tells how
gtrid/bqual were generated. But if formatID == -1
(it means "null") gtrid/bqual don't have any value at all.

This fixes main.xa_sync failure in
./mtr main.quary_cache_debug main.xa_sync

- To ensure that mallocs are marked for the correct THD, even if it's
  allocated in another thread, I added the thread_id to the THD constructor
- Added st_my_thread_var to thr_lock_info_init() to avoid a call to my_thread_var
- Moved things from THD::THD() to THD::init()
- Moved some things to THD::cleanup()
- Added THD::free_connection() and THD::reset_for_reuse()
- Added THD to CONNECT::create_thd()
- Added THD::thread_dbug_id and st_my_thread_var->dbug_id. These are needed
  to ensure that we have a constant thread_id used for debugging with a THD,
  even if it changes thread_id (=connection_id)
- Set variables.pseudo_thread_id in constructor. Removed not needed sets.

MDEV-10168 - Remove  btr_search_n_succ and btr_search_n_hash_fail counters

for running obsolete versions of Windows

- Fix build failure when built without wsrep (Win)
- Update the expected warning in sys_vars.wsrep_start_position

As a side effect tc_remove_all_unused_tables() has to call
tc_wait_for_mdl_deadlock_detector() once per TABLE object now, while it called
it only once before. This should be acceptable since actual wait will still be
performed only once. It only adds redundant checks for all_tables_refs.

This is mostly needed to hide all references to free_tables, so that further
implementation of multi-instance list can be done completely inside
table_cache.cc

Allowing GROUP_CONCAT(... ORDER BY ..) in queries with ROLLUP.

The story of the related code:

1. The original patch from Wax
commit: 0b505fb4
date: Tue Mar 18 03:07:40 2003

opt_gorder_clause reused the regular order_clause,
which already had some protection against ROLLUP queries:

order_clause:
        ORDER_SYM BY
        {
          LEX *lex=Lex;
          if (lex->current_select->linkage != GLOBAL_OPTIONS_TYPE &&
              lex->current_select->select_lex()->olap !=
              UNSPECIFIED_OLAP_TYPE)
          {
            net_printf(lex->thd, ER_WRONG_USAGE,
                       "CUBE/ROLLUP",
                       "ORDER BY");
            YYABORT;
          }
        } order_list;

The assumption that ORDER BY in group_concat() had to have
the same ROLLUP restriction (with order_clause) was wrong.

Moreover, GROUP_CONCAT() in select_item_list was not affected
by this restriction, because WITH ROLLUP goes after
select_item_list and therefore sel->olap is always equal
to UNSPECIFIED_OLAP_TYPE during select_item_list.

GROUP BY was not affected for two reasons:
- it goes before WITH ROLLUP and sel->olap is still
  UNSPECIFIED_OLAP_TYPE
- Aggregate functions like AVG(), GROUP_CONCAT() in GROUP BY
  are not allowed

So only GROUP_CONCAT() in HAVING and ORDER BY clauses
were erroneously affected by this restriction.

2. Bug#27848 rollup in union part causes error with order of union
commit: 3f6073ae
Author: unknown <igor@olga.mysql.com>  2007-12-15 01:42:46

The condition in the ROLLUP protection code became more complex.
Note, opt_gconcat_order still reused the regular order_clause.

3. Bug#16347426 ASSERTION FAILED: (SELECT_INSERT &&
                 !TABLES->NEXT_NAME_RESOLUTION_TABLE) || !TAB
commit: 2d836633
author: Chaithra Gopalareddy <chaithra.gopalareddy@oracle.com>
date: 2013-04-14 06:00:49

opt_gorder_clause was refactored not to use order_clause and
to collect information directly to select->gorder_list.
The ROLLUP protection code was duplicated from order_clause
to the new version of opt_gorder_clause.

- Validate the specified wsrep_start_position value by also
checking the return status of wsrep->sst_received. This also
ensures that changes in wsrep_start_position is not allowed
when the node is not in JOINING state.
- Do not allow decrease in seqno within same UUID.
- The initial checkpoint in SEs should be [0...:-1].

Using /*empty*/, union_list, union_order_or_limit instead.
This is to get rid of lex->current_select->braces easier
(separately in union_list and in union_order_or_limit)

UNION, ROLLUP, GROUP_CONCAT, for better coverage.

MDEV-10102 Disallow CREATE VIEW .. PROCEDURE ANALYSE() syntactically

- Moving opt_union_order_or_limit inside union_opt,
  as it's not used in other places any more.
- Changing union_opt to have no type.
  Earlier (before all MDEV-8909 dependency tasks) it had the <num> type,
  and it's return value was used to generate errors.
  Now union_opt does not need a return value because the grammar
  disallows ORDER and LIMIT clauses in wrong context.

MDEV-10095 Fix derived tables to return a syntax error instead of "Illegal usage of UNION and LIMIT"

from select_derived_init.

(to simplify further MDEV-8909 changes)

1. Better semantic readability:
- Moving get_select_lex_derived inside select_derived_init
  and decomposing it into get_select_lex and
  $1->init_nested_join(lex->thd)
-  Moving DBUG_ASSERT($1 == Lex->current_select) inside
   select_derived_init

Now init_nested_join() and end_nested_join() reside inside
the same rule select_derived_init.
(It's very likely that they can be further removed,
 as there are no any joins in this rule).

3. Better grammar readability:
  Moving SELECT_SYM from select_derived_init to derived_query_specification.
  It's easier to read a rule when it starts from a terminal symbol.

1. Moving the "| get_select_lex_derived select_derived_init" part of
select_derived into a separate new rule derived_query_specification.
2. Using derived_query_specification directly in select_derived_union
rather than in select_derived.
3. Moving the sequence "opt_order_clause opt_limit_clause opt_select_lock_type"
from select_derived2 to select_derived_union,
after derived_query_specification.

Effectively, the parser now does not go through the sequence
"opt_order_clause opt_limit_clause ... opt_union_order_or_limit" any more.

This fixes the problem with double LIMIT clause and removes 2 shift/reduce
conflicts.