- It turns out, there are statements that will call lex_start(thd->lex) 
  after parsing has been finished. lex_start() will set lex->explain=NULL,
  which will lose the pointer to already allocated Explain_plan object.
- To get rid of this, switch to lazy creation of lex->explain.  Now, it is 
  created only when we get a part ot query plan.

- MYSQL_MULTI_DELETE_DONE probe compile failure
- show_explain_non_select.test

- When showing EXPLAIN output in the slow query log, 
  format it so that one could use grep or other tool
  to get the output.

- Save the query plan after the statement was executed so that its gets into
  the slow query log.

- Merge with 10.0-base

- Port grant_explain_non_select.{test,result} from mysql-5.6
- Per Sanja's hint, fix mysql_make_view() to take into 
  account that EXPLAIN now is not necessarily EXPLAIN SELECT.

- Fix a problem with EXPLAIN multi_table UPDATE: 
  = Do use multi_update object, because multi_update::prepare() does
    various setup, e.g. it disables index-only for the tables to be updated.
  = Protect multi_update::prepare() from being invoked multiple times. 
    If the query has subqueries, they may try to invoke it, for some reason.

- eliminate join_save_qpf() function.

The patch for bug mdev-5105 incorrectly counted conditions in nested joins.

        The emb_free_embedded_thd() has the thread-unsafe code
        so should be 'mutexed' also.

Update the SHOW EXPLAIN code to work with the 
new architecture (part#1):
Before, SHOW EXPLAIN operated on real query plan structures, 
which meant it had to check when SELECTs are created/deleted.
SELECTs would call apc_target->enable() when they got a query 
plan and disable() when their query plan was deleted.

Now, Explain data structure becomes available at once (and we
call apc_target->enable()) and then it stays until it is deleted
(when that happens, we call apc_target->disable()).

- Fix EXPLAIN INSERT DELAYED ... : do call end_delayed_insert().

        LOCK_thread_count locked when we do threads.append().

Objects of the classes Item_func_isnull and Item_func_isnotnull
must have the flag sargable set to TRUE.
Set the value of the flag sargable only in constructors of the 
classes inherited from Item_int_func.

- When INSERT catches an error, it should not attempt to send parts of 
  EXPLAIN output.

Make IO thread check for end of event group, so that upon
disconnect at the end of an event group it can report the last
read GTID as expected. Also inject a fake Rotate event at
reconnect when skipping part of an initial event group, to
give SQL thread the correct Read_Master_Log_Pos.

Reported by Pavel Ivanov.

Mark a few PCRE CMake variables as advanced,
so the are not presented in cmake-gui by default.

The bug caused a memory overwrite in the function update_ref_and_keys()
It happened due to a wrong value of SELECT_LEX::cond_count. This value
historically was calculated by the fix_fields method. Now the logic of
calling this method became too complicated and, as a result, this value
is calculated not always correctly.
The patch changes the way how and when  the values of SELECT_LEX::cond_count
and of SELECT_LEX::between_count are calculated. The new code does it just at
the beginning of update_ref_and_keys().
 

- Don't save UNION's EXPLAIN data if optimization failed with 
  an error. We could end up saving incomplete plan, which will 
  cause a crash when we attempt to print it.

        Additional patch for the 5.5.

Backport mysql-test/t/myisam_explain_non_select_all.test from mysql-5.6
- the .result file was modified because MariaDB choses different 
  query plans in a number cases. Also, we don't have some of the 
  "incorrect EXPLAIN output" bugs that they still have.

The .test file and includes were taken verbatim with one 
exception: two tests were disabled with --disable parsing:

1. @@sql_safe_updates is not enforced EXPLAINs of multitable 
   updates. In MariaDB, the execution itself will produce 
   ER_UPDATE_WITHOUT_KEY_IN_SAFE_MODE, but EXPLAIN won't.
2. Their case #71 hits some old bug in MyISAM (See their comments
   in explain_non_select.inc for details).

- Produce correct `key_len` when type=index.

        The Gis_point::init_from_wkt called the String::realloc(),
        and this call is quite slow in the DEBUG mode. Which makes
        loading the huge polygon hang forever.
        Fixed by using the String::realloc(size, inc_size) version instead
        as it's done for other spatial features.

- Produce correct #rows for ORDER BY ... LIMIT N queries that take advantage of 
  ordered index read to read only N rows.

- Generate correct contents of `Extra` column for UPDATEs/DELETEs that use quick selects
- UPDATEs with used_key_is_modified=true will show "Using buffer"

Do not pass PCRE_UCP flag for binary data.
This makes bytes 0x80..FF not to belong to 
generic character classes \d (digit) and \w (word character).

SELECT 0xFF RLIKE '\\w';
 -> 0

Note, this change does not affect non-binary data,
which is still examined with the PCRE_UCP flag by default.

- Update test results after last few csets
- Generate correct value for `possible_keys` column for single table UPDATE/DELETE.

- if EXPLAIN DELETE prints "Deleting all rows", it should show the 
  expected number of rows in the rows column.

- Add support for EXPLAIN INSERT.

- Better EXPLAIN-saving methods for quick selects