MWL#116: Efficient group commit: Fix assertion by moving init/deinit out of constructor in global static object.

MWL#116: Efficient group commit: Fix bug that binlog pos stored by XtraDB during commit was wrong when more than one commit in group.

Now the actual binlog position for each commit is stored in THD, and XtraDB
can fetch the correct value from within commit_ordered() or commit().


mysql-test/r/group_commit_binlog_pos.result:
  Test case for XtraDB binlog position.
mysql-test/t/group_commit_binlog_pos-master.opt:
  Test case for XtraDB binlog position.
mysql-test/t/group_commit_binlog_pos.test:
  Test case for XtraDB binlog position.
sql/log.cc:
  Save binlog position corresponding to commit in THD, and make accessible to storage engine.
sql/sql_parse.cc:
  Add generic crash point for use in test cases.
storage/xtradb/handler/ha_innodb.cc:
  Update to use new method of getting current binlog position that works with group commit.
storage/xtradb/handler/ha_innodb.h:
  Update to use new method of getting current binlog position that works with group commit.

Tweak the commit_ordered() semantics. Now it is only called for transactions
that go through 2-phase commit. This avoids forcing engines to make commits
visible before they are durable.

Also take LOCK_commit_ordered() around START TRANSACTION WITH CONSISTENT
SNAPSHOT, to get a truly consistent snapshot.

Preliminary commit for testing

Updated results for failing test cases (In all cases the estimated number of rows was different)

The condition over the outer tables now are extracted from
the on condition of any outer join. This condition is
saved in a special field of the JOIN_TAB structure for
the first inner table of the outer join. The condition
is checked before the first inner table is accessed. If 
it turns out to be false the table is not accessed at all
and a null complemented row is generated immediately.

the setting of optimizer switch for table elimination.

The implementation of the virtual method not_null_tables for the class
Item_outer_ref must always return 0.

Made sure that rr_quick is used to read the next record whenever
a quick select is used to retrieve the table records.

Fixed crashing bug when doing ALTER TABLE RENAME with transactional tables.

client/mysqltest.cc:
  Added errno to error message for system calls (delete, rename etc)
  Write error message for failures of system calls
mysql-test/include/cleanup_fake_relay_log.inc:
  Disable warnings for remove_file
mysql-test/include/diff_tables.inc:
  Disable warnings for remove_file
mysql-test/include/maria_empty_logs.inc:
  Disable warnings for remove_file
mysql-test/include/maria_make_snapshot.inc:
  Disable warnings for remove_file
mysql-test/include/maria_make_snapshot_for_feeding_recovery.inc:
  Disable warnings for remove_file
mysql-test/include/mysqlhotcopy.inc:
  Disable warnings for remove_file
mysql-test/include/ndb_backup.inc:
  Disable warnings for remove_file
mysql-test/include/ndb_backup_print.inc:
  Disable warnings for remove_file
mysql-test/r/alter_table_trans.result:
  Test of crashing ALTER TABLE RENAME bug
mysql-test/t/alter_table_trans.test:
  Test of crashing ALTER TABLE RENAME bug
mysql-test/t/mysqltest.test:
  Disable warnings for remove_file and move_file
mysys/my_copy.c:
  Fixed wrong error message
sql/sql_table.cc:
  Fixed crashing bug when doing ALTER TABLE RENAME with transactional tables.

libtool does not like it and fails the build.

- When reopening table during recovery, don't set file length from file sizes as file is not flushed.
New feature in Aria recovery:
- Create database directory if missing.



storage/maria/ma_recovery.c:
  Don't print Query: twice when running maria_read_log
  Create database directory if missing. (Needed when running maria_read_log on an empty database direcory during testing)
  More DBUG_print
  Fixed bug: When reopening table during recovery, don't set file length from file sizes as file is not flushed.

The problem could be demonstrated with an outer join of two single-row
tables where the values of the join attributes were null. Any query
with such a join could return a wrong result set if the where
condition of the query was not empty. For queries with empty
where conditions the result sets were correct.
This was the consequence of two bugs in the code:
 - Item_equal objects for on conditions of outer joins were
   not built if the processed query had no where condition
 - the check for null values in the code that evaluated constant 
   Item_equal objects was incorrect.
Fixed both above problems.
Added a test case for the bug and adjusted results for some other
test cases.

Fix for LP#634943 "marked as crashed", "zerofilling" and "wrong data in bitmap" when recovering Aria tables
This was an interaction of several bugs:
- Tables marked as opened was not properly unmarked on recovery if there was not changes since checkpoint
- zerofill of tables put wrong data in bitmap if directory for page was full
- Tables was thought as 'moved' during recovery if they had a create_lsn bigger than the lsn in the control file.


storage/maria/ha_maria.cc:
  If table is moved and crashed, threat it as crashed.
  (Not a related to this bug, but still good to have fixed)
storage/maria/ma_blockrec.c:
  Make enough_free_entries_on_page() global
storage/maria/ma_blockrec.h:
  Make enough_free_entries_on_page() global
storage/maria/ma_check.c:
  If directory is full, mark page full. Fixes bug in zerofill
storage/maria/ma_open.c:
  Don't marke files as MOVED during recovery if create_trid > trnman_max_trid, as this fails for tables created after checkpoint.
storage/maria/ma_recovery.c:
  Reset open_count in file that was open during crash and was part of checkpoint.
  Fixed wrong warning of 'open count' after recovery of files that was not touched since checkpoint.
storage/maria/maria_chk.c:
  Changed not documented option --log-dir to --logdir
  Document more of the options.
  Clean up output for --help
storage/maria/trnman.c:
  Added DBUG_PRINT

When not-exists optimization was applied to a table that
happened to be an inner table of two outer joins, one
embedded into another, then setting the match flag for
the embedding outer join on could be skipped. This caused
generation of extra null complemented rows.
Made sure that the match flags are set correctly in all cases
when not-exists optimization is used.  

Extend remove_function_from_trace.pl to work with many threads (patch from Sergei)


dbug/remove_function_from_trace.pl:
  Extend remove_function_from_trace.pl to work with many threads (patch from Sergei)
storage/maria/ma_bitmap.c:
  Added marker that table had changed since last checkpoint.
  This ensures that we will flush all bitmap pages from cache at checkpoint.
  This fixes bug LP#605798 "wrong data in bitmap" after recovery.
storage/maria/ma_check.c:
  Cleaned up error output
storage/maria/ma_recovery.c:
  Cleaned up error output
storage/maria/maria_def.h:
  Added changed_not_flushed

Don't give warning about block_insert if table is crashed.

scripts/mysql_install_db.sh:
  Added link to KB.
storage/maria/ma_recovery.c:
  Don't give warning about block_insert if table is crashed.

double compilation