List:Commits« Previous MessageNext Message »
From:konstantin Date:February 19 2008 11:43am
Subject:bk commit into 5.1 tree (kostja:1.2518) BUG#12713
View as plain text  
Below is the list of changes that have just been committed into a local
5.1 repository of kostja.  When kostja does a push these changes
will be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet@stripped, 2008-02-19 14:43:01+03:00, kostja@dipika.(none) +27 -0
  A fix and a test case for Bug#12713 "Error in a stored function called from 
  a SELECT doesn't cause ROLLBACK of statem".
  
  The idea of the fix is to ensure that we always commit the current
  statement at the end of dispatch_command(). In order to not issue
  redundant disc syncs, an optimization of the two-phase commit
  protocol is implemented to bypass the two phase commit if
  the transaction is read-only.

  mysql-test/include/commit.inc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +742 -0
    New BitKeeper file ``mysql-test/include/commit.inc''

  mysql-test/include/commit.inc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +0 -0

  mysql-test/r/commit_1innodb.result@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +888 -0
    New BitKeeper file ``mysql-test/r/commit_1innodb.result''

  mysql-test/r/commit_1innodb.result@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +0 -0

  mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result@stripped, 2008-02-19 14:42:57+03:00, kostja@dipika.(none) +1 -2
    Update test results.

  mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result@stripped, 2008-02-19 14:42:57+03:00, kostja@dipika.(none) +1 -3
    Update test results.

  mysql-test/suite/rpl_ndb/t/disabled.def@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +2 -0
    Disable the tests, for which this changeset reveals a bug:
    the injector thread does not always add 'statement commit' to the
    rows injected in circular replication set up.
    To be investigated separately.

  mysql-test/t/commit_1innodb.test@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +6 -0
    New BitKeeper file ``mysql-test/t/commit_1innodb.test''

  mysql-test/t/commit_1innodb.test@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +0 -0

  sql/ha_ndbcluster_binlog.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +1 -0
    Add close_thread_tables() to run_query: this ensures
    that all tables are closed and there is no pending statement transaction.

  sql/handler.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +553 -62
    Implement optimisation of read-only transactions.
    If a transaction consists only of DML statements that do not change
    data, we do not perform a two-phase commit for it 
    (run one phase commit only).

  sql/handler.h@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +113 -9
    Implement optimisation of read-only transactions.
    If a transaction consists only of DML statements that do not change
    data, we do not perform a two-phase commit for it 
    (run one phase commit only).

  sql/log.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +10 -0
    Mark the binlog transaction read-write whenever it's started.
    We never read from binlog, so it's safe and least intrusive to add
    this mark up here.

  sql/log_event.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +1 -1
    Update to the new layout of thd->transaction.

  sql/rpl_injector.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +20 -0
    Always commit statement transaction before committing the global one.

  sql/sp.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +17 -3
    Ad comments.

  sql/sp_head.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +1 -0
    Add comments.

  sql/sql_base.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +28 -30
    Commit transaction at the end of the statement. Always.

  sql/sql_class.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +2 -2
    Update thd_ha_data to return the right pointer in the new layout.
    
    Fix select_dumpvar::send_data to properly return operation status.
    A test case from commit.inc would lead to an assertion failure in the 
    diagnostics area (double assignment). Not test otherwise by the test suite.

  sql/sql_class.h@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +30 -2
    Implement a new layout of storage engine transaction info in which 
    it is easy to access all members related to the handlerton only
    based on ht->slot.

  sql/sql_cursor.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +3 -2
    Update to the new layout of thd->transaction.

  sql/sql_delete.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +17 -21
    Remove wrong and now redundant calls to ha_autocommit_or_rollback.
    The transaction is committed in one place, at the end of the statement.
    Remove calls to mysql_unlock_tables, since some engines count locks
    and commit statement transaction in unlock_tables(), which essentially
    equates mysql_unlock_tables to ha_autocommit_or_rollback.
    Previously it was necessary to unlock tables soon because we wanted
    to avoid sending of 'ok' packet to the client under locked tables.
    This is no longer necessary, since OK packet is also sent from one place
    at the end of transaction.

  sql/sql_do.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +11 -1
    Add DO always clears the error, we must rollback the current
    statement before this happens. Otherwise the statement will be committed,
    and not rolled back in the end.

  sql/sql_insert.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +15 -27
    Remove wrong and now redundant calls to ha_autocommit_or_rollback.
    The transaction is committed in one place, at the end of the statement.
    Remove calls to mysql_unlock_tables, since some engines count locks
    and commit statement transaction in unlock_tables(), which essentially
    equates mysql_unlock_tables to ha_autocommit_or_rollback.
    Previously it was necessary to unlock tables soon because we wanted
    to avoid sending of 'ok' packet to the client under locked tables.
    This is no longer necessary, since OK packet is also sent from one place
    at the end of transaction.

  sql/sql_load.cc@stripped, 2008-02-19 14:42:58+03:00, kostja@dipika.(none) +0 -10
    Remove wrong and now redundant calls to ha_autocommit_or_rollback.
    The transaction is committed in one place, at the end of the statement.
    Remove calls to mysql_unlock_tables, since some engines count locks
    and commit statement transaction in unlock_tables(), which essentially
    equates mysql_unlock_tables to ha_autocommit_or_rollback.
    Previously it was necessary to unlock tables soon because we wanted
    to avoid sending of 'ok' packet to the client under locked tables.
    This is no longer necessary, since OK packet is also sent from one place
    at the end of transaction.

  sql/sql_parse.cc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +10 -17
    Implement optimisation of read-only transactions: bypass 2-phase
    commit for them.
    Always commit statement transaction before commiting the global one.
    Fix an unrelated crash in check_table_access, when called from 
    information_schema.

  sql/sql_partition.cc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +21 -17
    Partitions commit at the end of a DDL operation.
    Make sure that send_ok() is done only if the commit has succeeded.

  sql/sql_table.cc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +12 -9
    Use ha_autocommit_or_rollback and end_active_trans everywhere.
    Add end_trans to mysql_admin_table, so that it leaves no pending
    transaction.

  sql/sql_udf.cc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +8 -1
    Remvove a redundant call to close_thread_tables()

  sql/sql_update.cc@stripped, 2008-02-19 14:42:59+03:00, kostja@dipika.(none) +1 -28
    Remove wrong and now redundant calls to ha_autocommit_or_rollback.
    The transaction is committed in one place, at the end of the statement.
    Remove calls to mysql_unlock_tables, since some engines count locks
    and commit statement transaction in unlock_tables(), which essentially
    equates mysql_unlock_tables to ha_autocommit_or_rollback.
    Previously it was necessary to unlock tables soon because we wanted
    to avoid sending of 'ok' packet to the client under locked tables.
    This is no longer necessary, since OK packet is also sent from one place
    at the end of transaction.

diff -Nrup a/mysql-test/include/commit.inc b/mysql-test/include/commit.inc
--- /dev/null	Wed Dec 31 16:00:00 196900
+++ b/mysql-test/include/commit.inc	2008-02-19 14:42:59 +03:00
@@ -0,0 +1,742 @@
+## Bug#12713 (Error in a stored function called from a SELECT doesn't cause
+##    ROLLBACK of statem)
+
+##
+## Pre-Requisites :
+## - $engine_type should be set
+##
+
+set sql_mode=no_engine_substitution;
+eval set storage_engine = $engine_type;
+set autocommit=1;
+
+--disable_warnings
+drop table if exists t1;
+drop table if exists t2;
+drop table if exists t3;
+drop function if exists f2;
+drop procedure if exists bug12713_call;
+drop procedure if exists bug12713_dump_spvars;
+drop procedure if exists dummy;
+--enable_warnings
+
+create table t1 (a int);
+create table t2 (a int unique);
+create table t3 (a int);
+
+# a workaround for Bug#32633: Can not create any routine if
+# SQL_MODE=no_engine_substitution
+
+set sql_mode=default;
+
+insert into t1 (a) values (1), (2);
+insert into t3 (a) values (1), (2);
+
+delimiter |;
+
+## Cause a failure every time
+create function f2(x int) returns int
+begin
+  insert into t2 (a) values (x);
+  insert into t2 (a) values (x);
+  return x;
+end|
+
+delimiter ;|
+
+set autocommit=0;
+
+flush status;
+##============================================================================
+## Design notes
+##
+## In each case, statement rollback is expected.
+## for transactional engines, the rollback should be properly executed
+## for non transactional engines, the rollback may cause warnings.
+##
+## The test pattern is as follows
+## - insert 1000+N
+## - statement with a side effect, that fails to insert N twice
+## - a statement rollback is expected (expecting 1 row 1000+N only) in t2
+## - a rollback is performed
+## - expecting a clean table t2.
+##============================================================================
+
+insert into t2 (a) values (1001);
+--error ER_DUP_ENTRY
+insert into t1 (a) values (f2(1));
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1002);
+--error ER_DUP_ENTRY
+insert into t3 (a) select f2(2) from t1;
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1003);
+--error ER_DUP_ENTRY
+update t1 set a= a + f2(3);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1004);
+--error ER_DUP_ENTRY
+update t1, t3 set t1.a = 0, t3.a = 0 where (f2(4) = 4) and (t1.a = t3.a);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1005);
+--error ER_DUP_ENTRY
+delete from t1 where (a = f2(5));
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1006);
+--error ER_DUP_ENTRY
+delete from t1, t3 using t1, t3 where (f2(6) = 6) ;
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1007);
+--error ER_DUP_ENTRY
+replace t1 values (f2(7));
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1008);
+--error ER_DUP_ENTRY
+replace into t3 (a) select f2(8) from t1;
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1009);
+--error ER_DUP_ENTRY
+select f2(9) from t1 ;
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1010);
+--error ER_DUP_ENTRY
+show databases where (f2(10) = 10);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1011);
+--error ER_DUP_ENTRY
+show tables where (f2(11) = 11);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1012);
+--error ER_DUP_ENTRY
+show triggers where (f2(12) = 12);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1013);
+--error ER_DUP_ENTRY
+show table status where (f2(13) = 13);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1014);
+--error ER_DUP_ENTRY
+show open tables where (f2(14) = 14);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1015);
+--error ER_DUP_ENTRY
+show columns in mysql.proc where (f2(15) = 15);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1016);
+--error ER_DUP_ENTRY
+show status where (f2(16) = 16);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1017);
+--error ER_DUP_ENTRY
+show variables where (f2(17) = 17);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1018);
+--error ER_DUP_ENTRY
+show charset where (f2(18) = 18);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1019);
+--error ER_DUP_ENTRY
+show collation where (f2(19) = 19);
+select * from t2;
+rollback;
+select * from t2;
+
+--echo # We need at least one procedure to make sure the WHERE clause is
+--echo # evaluated
+create procedure dummy() begin end;
+insert into t2 (a) values (1020);
+--error ER_DUP_ENTRY
+show procedure status where (f2(20) = 20);
+select * from t2;
+rollback;
+select * from t2;
+drop procedure dummy;
+
+insert into t2 (a) values (1021);
+--error ER_DUP_ENTRY
+show function status where (f2(21) = 21);
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1022);
+prepare stmt from "insert into t1 (a) values (f2(22))";
+--error ER_DUP_ENTRY
+execute stmt;
+select * from t2;
+rollback;
+select * from t2;
+
+insert into t2 (a) values (1023);
+do (f2(23));
+select * from t2;
+rollback;
+select * from t2;
+
+## Please note :
+## This will insert a record 1024 in t1 (statement commit)
+## This will insert a record 24 in t1 (statement commit)
+## then will rollback the second insert only (24) (statement rollback)
+## then will rollback the complete transaction (transaction rollback)
+
+delimiter |;
+
+create procedure bug12713_call ()
+begin
+  insert into t2 (a) values (24);
+  insert into t2 (a) values (24);
+end|
+
+delimiter ;|
+
+insert into t2 (a) values (1024);
+--error ER_DUP_ENTRY
+call bug12713_call();
+select * from t2;
+rollback;
+select * from t2;
+
+--echo =======================================================================
+--echo Testing select_to_file
+--echo =======================================================================
+
+insert into t2 (a) values (1025);
+
+--replace_result $MYSQLTEST_VARDIR ..
+--error ER_DUP_ENTRY
+eval select f2(25) into outfile "$MYSQLTEST_VARDIR/tmp/dml.out" from t1;
+select * from t2;
+rollback;
+select * from t2;
+--remove_file $MYSQLTEST_VARDIR/tmp/dml.out
+
+insert into t2 (a) values (1026);
+--replace_result $MYSQLTEST_VARDIR ..
+--error ER_DUP_ENTRY
+eval load data infile "../std_data_ln/words.dat" into table t1 (a) set a:=f2(26);
+
+select * from t2;
+rollback;
+select * from t2;
+
+--echo =======================================================================
+--echo Testing select_dumpvar
+--echo =======================================================================
+
+insert into t2 (a) values (1027);
+--error ER_DUP_ENTRY
+select f2(27) into @foo;
+select * from t2;
+rollback;
+select * from t2;
+
+--echo =======================================================================
+--echo Testing Select_fetch_into_spvars 
+--echo =======================================================================
+
+delimiter |;
+
+create procedure bug12713_dump_spvars ()
+begin
+  declare foo int;
+
+  declare continue handler for sqlexception
+  begin
+    select "Exception trapped";
+  end;
+
+  select f2(28) into foo;
+  select * from t2;
+end|
+
+delimiter ;|
+
+insert into t2 (a) values (1028);
+call bug12713_dump_spvars ();
+rollback;
+select * from t2;
+
+--echo =======================================================================
+--echo Cleanup
+--echo =======================================================================
+
+set autocommit=default;
+
+drop table t1;
+drop table t2;
+drop table t3;
+drop function f2;
+drop procedure bug12713_call;
+drop procedure bug12713_dump_spvars;
+--echo #
+--echo # Bug#12713 Error in a stored function called from a SELECT doesn't
+--echo # cause ROLLBACK of statem
+--echo #
+--echo # Verify that two-phase commit is not issued for read-only
+--echo # transactions.
+--echo #
+--echo # Verify that two-phase commit is issued for read-write transactions,
+--echo # even if the change is done inside a stored function called from
+--echo # SELECT or SHOW statement.
+--echo #
+set autocommit=0;
+--disable_warnings
+drop table if exists t1;
+drop table if exists t2;
+drop function if exists f1;
+drop procedure if exists p_verify_status_increment;
+--enable_warnings
+
+set sql_mode=no_engine_substitution;
+create table t1 (a int unique); 
+create table t2 (a int) engine=myisam;
+set sql_mode=default;
+--echo #
+--echo # An auxiliary procedure to track Handler_prepare and Handler_commit
+--echo # statistics.
+--echo #
+delimiter |;
+create procedure
+p_verify_status_increment(commit_inc_mixed int, prepare_inc_mixed int,
+                          commit_inc_row int, prepare_inc_row int)
+begin
+  declare commit_inc int;
+  declare prepare_inc int;
+  declare old_commit_count int default ifnull(@commit_count, 0);
+  declare old_prepare_count int default ifnull(@prepare_count, 0);
+  declare c_res int;
+# Use a cursor to have just one access to I_S instead of 2, it is very slow
+# and amounts for over 90% of test CPU time
+  declare c cursor for
+     select variable_value
+     from information_schema.session_status
+     where variable_name='Handler_commit' or variable_name='Handler_prepare'
+     order by variable_name;
+
+  if @@global.binlog_format = 'ROW' then
+    set commit_inc= commit_inc_row;
+    set prepare_inc= prepare_inc_row;
+  else
+    set commit_inc= commit_inc_mixed;
+    set prepare_inc= prepare_inc_mixed;
+  end if;
+
+  open c;
+  fetch c into c_res;
+  set @commit_count=c_res;
+  fetch c into c_res;
+  set @prepare_count=c_res;
+  close c;
+
+  if old_commit_count + commit_inc <> @commit_count then
+    select concat("Expected commit increment: ", commit_inc,
+        " actual: ", @commit_count - old_commit_count)
+    as 'ERROR';
+  elseif old_prepare_count + prepare_inc <> @prepare_count then
+    select concat("Expected prepare increment: ", prepare_inc,
+        " actual: ", @prepare_count - old_prepare_count)
+    as 'ERROR';
+  else
+    select '' as 'SUCCESS';
+  end if;
+end|
+delimiter ;|
+--echo # Reset Handler_commit and Handler_prepare counters
+flush status;
+--echo #
+--echo # 1. Read-only statement: SELECT
+--echo #
+select * from t1;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 2. Read-write statement: INSERT, insert 1 row. 
+--echo #
+insert into t1 (a) values (1);
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 3. Read-write statement: UPDATE, update 1 row. 
+--echo #
+update t1 set a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 4. Read-write statement: UPDATE, update 0 rows, 1 row matches WHERE 
+--echo #
+--echo # Note the wrong Handler_prepare/Handler_commit count is due to
+--echo # Bug#29157 "UPDATE, changed rows incorrect" and
+--echo # Bug#Bug #33846   UPDATE word:Wrong 'Changed rows' if InnoDB, unique
+--echo # key and no rows qualify WHERE
+--echo #
+update t1 set a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 5. Read-write statement: UPDATE, update 0 rows, 0 rows match WHERE 
+--echo #
+--echo # In mixed replication mode, there is a read-only transaction
+--echo # in InnoDB and also the statement is written to the binary log.
+--echo # So we have two commits but no 2pc, since the first engine's
+--echo # transaction is read-only.
+--echo # In the row level replication mode, we only have the read-only
+--echo # transaction in InnoDB and nothing is written to the binary log.
+--echo #
+update t1 set a=3 where a=1;
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 6. Read-write statement: DELETE, delete 0 rows. 
+--echo #
+delete from t1 where a=1;
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 7. Read-write statement: DELETE, delete 1 row. 
+--echo #
+delete from t1 where a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 8. Read-write statement: unqualified DELETE
+--echo #
+--echo # In statement or mixed replication mode, we call
+--echo # handler::ha_delete_all_rows() and write statement text
+--echo # to the binary log. This results in two read-write transactions.
+--echo # In row level replication mode, we do not call
+--echo # handler::ha_delete_all_rows(), but delete rows one by one.
+--echo # Since there are no rows, nothing is written to the binary log.
+--echo # Thus we have just one read-only transaction in InnoDB.
+delete from t1;
+call p_verify_status_increment(2, 2, 1, 0);
+commit;
+call p_verify_status_increment(2, 2, 1, 0);
+
+--echo # 9. Read-write statement: REPLACE, change 1 row. 
+--echo #
+replace t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 10. Read-write statement: REPLACE, change 0 rows. 
+--echo #
+replace t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 11. Read-write statement: IODKU, change 1 row. 
+--echo #
+insert t1 set a=1 on duplicate key update a=a+1;
+call p_verify_status_increment(2, 2, 2, 2);
+select * from t1;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 12. Read-write statement: IODKU, change 0 rows. 
+--echo #
+insert t1 set a=2 on duplicate key update a=2;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 13. Read-write statement: INSERT IGNORE, change 0 rows. 
+--echo #
+insert ignore t1 set a=2;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 14. Read-write statement: INSERT IGNORE, change 1 row. 
+--echo #
+insert ignore t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+--echo # 15. Read-write statement: UPDATE IGNORE, change 0 rows. 
+--echo #
+update ignore t1 set a=2 where a=1;
+call p_verify_status_increment(2, 2, 1, 0);
+commit;
+call p_verify_status_increment(2, 2, 1, 0);
+--echo #
+--echo # Create a stored function that modifies a
+--echo # non-transactional table. Demonstrate that changes in
+--echo # non-transactional tables do not affect the two phase commit
+--echo # algorithm.
+--echo #
+delimiter |;
+create function f1() returns int
+begin
+  insert t2 set a=2;
+  return 2;
+end|
+delimiter ;|
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo # 16. A function changes non-trans-table.
+--echo #
+select f1();
+call p_verify_status_increment(0, 0, 0, 0);
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo # 17. Read-only statement, a function changes non-trans-table.
+--echo #
+select f1() from t1;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 18. Read-write statement: UPDATE, change 0 (transactional) rows. 
+--echo #
+select count(*) from t2;
+update t1 set a=2 where a=f1()+10;
+select count(*) from t2;
+call p_verify_status_increment(2, 0, 2, 0);
+commit;
+call p_verify_status_increment(2, 0, 2, 0);
+--echo #
+--echo # Replace the non-transactional table with a temporary
+--echo # transactional table. Demonstrate that a change to a temporary
+--echo # transactional table does not provoke 2-phase commit, although
+--echo # does trigger a commit and a binlog write (in statement mode).
+--echo #
+drop table t2;
+set sql_mode=no_engine_substitution;
+create temporary table t2 (a int);
+call p_verify_status_increment(0, 0, 0, 0);
+set sql_mode=default;
+--echo # 19. A function changes temp-trans-table.
+--echo #
+select f1();
+--echo # Two commits because a binary log record is written
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 20. Read-only statement, a function changes non-trans-table.
+--echo #
+select f1() from t1;
+--echo # Two commits because a binary log record is written
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 21. Read-write statement: UPDATE, change 0 (transactional) rows. 
+--echo #
+update t1 set a=2 where a=f1()+10;
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 22. DDL: ALTER TEMPORARY TABLE, should not cause a 2pc
+--echo #
+alter table t2 add column b int default 5;
+--echo # A commit is done internally by ALTER. 
+call p_verify_status_increment(2, 0, 2, 0);
+commit;
+--echo # There is nothing left to commit
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo # 23. DDL: RENAME TEMPORARY TABLE, does not start a transaction
+--echo
+--echo # No test because of Bug#8729 "rename table fails on temporary table"
+
+--echo # 24. DDL: TRUNCATE TEMPORARY TABLE, does not start a transaction
+--echo
+truncate table t2;
+call p_verify_status_increment(2, 0, 2, 0);
+commit;
+--echo # There is nothing left to commit
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo # 25. Read-write statement: unqualified DELETE 
+--echo
+delete from t2;
+call p_verify_status_increment(2, 0, 1, 0);
+commit;
+--echo # There is nothing left to commit
+call p_verify_status_increment(2, 0, 1, 0);
+
+--echo # 25. DDL: DROP TEMPORARY TABLE, does not start a transaction
+--echo #
+drop temporary table t2;
+call p_verify_status_increment(0, 0, 0, 0);
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo # 26. Verify that SET AUTOCOMMIT issues an implicit commit
+--echo #
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+set autocommit=1;
+call p_verify_status_increment(2, 2, 2, 2);
+rollback;
+select a from t1 where a=3;
+call p_verify_status_increment(1, 0, 1, 0);
+delete from t1 where a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+set autocommit=0;
+call p_verify_status_increment(0, 0, 0, 0);
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+--echo # Sic: not actually changing the value of autocommit
+set autocommit=0;
+call p_verify_status_increment(0, 0, 0, 0);
+rollback;
+select a from t1 where a=3;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 27. Savepoint management
+--echo #
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+savepoint a;
+call p_verify_status_increment(0, 0, 0, 0);
+insert t1 set a=4;
+--echo # Sic: a bug. Binlog did not register itself this time.
+call p_verify_status_increment(1, 0, 1, 0);
+release savepoint a;
+rollback;
+call p_verify_status_increment(0, 0, 0, 0);
+select a from t1 where a=3;
+call p_verify_status_increment(1, 0, 1, 0);
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+
+--echo # 28. Read-write statement: DO
+--echo #
+create table t2 (a int);
+call p_verify_status_increment(0, 0, 0, 0);
+do (select f1() from t1 where a=2);
+call p_verify_status_increment(2, 2, 2, 2);
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+
+--echo # 29. Read-write statement: MULTI-DELETE
+--echo # 
+delete t1, t2 from t1 join t2 on (t1.a=t2.a) where t1.a=2;
+commit;
+call p_verify_status_increment(4, 4, 4, 4);
+
+--echo # 30. Read-write statement: INSERT-SELECT, MULTI-UPDATE, REPLACE-SELECT
+--echo # 
+insert into t2 select a from t1;
+commit;
+replace into t2 select a from t1;
+commit;
+call p_verify_status_increment(8, 8, 8, 8);
+#
+# Multi-update is one of the few remaining statements that still
+# locks the tables at prepare step (and hence starts the transaction.
+# Disable the PS protocol, since in this protocol we get a different
+# number of commmits (there is an extra commit after prepare
+#
+--disable_ps_protocol
+update t1, t2 set t1.a=4, t2.a=8 where t1.a=t2.a and t1.a=1;
+--enable_ps_protocol
+commit;
+call p_verify_status_increment(4, 4, 4, 4);
+
+--echo # 31. DDL: various DDL with transactional tables
+--echo #
+--echo # Sic: no table is created.
+create table if not exists t2 (a int) select 6 union select 7;
+--echo # Sic: first commits the statement, and then the transaction.
+call p_verify_status_increment(4, 4, 4, 4);
+create table t3 select a from t2;
+call p_verify_status_increment(4, 4, 4, 4);
+alter table t3 add column (b int);
+call p_verify_status_increment(2, 0, 2, 0);
+alter table t3 rename t4;
+call p_verify_status_increment(1, 0, 1, 0);
+rename table t4 to t3;
+call p_verify_status_increment(1, 0, 1, 0);
+truncate table t3;
+call p_verify_status_increment(2, 2, 2, 2);
+create view v1 as select * from t2;
+call p_verify_status_increment(1, 0, 1, 0);
+check table t1;
+call p_verify_status_increment(3, 0, 3, 0);
+--echo # Sic: after this bug is fixed, CHECK leaves no pending transaction
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+check table t1, t2, t3;
+call p_verify_status_increment(6, 0, 6, 0);
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+drop view v1;
+call p_verify_status_increment(0, 0, 0, 0);
+
+--echo #
+--echo # Cleanup
+--echo #
+drop table t1;
+drop procedure p_verify_status_increment;
+drop function f1;
diff -Nrup a/mysql-test/r/commit_1innodb.result b/mysql-test/r/commit_1innodb.result
--- /dev/null	Wed Dec 31 16:00:00 196900
+++ b/mysql-test/r/commit_1innodb.result	2008-02-19 14:42:59 +03:00
@@ -0,0 +1,888 @@
+set sql_mode=no_engine_substitution;
+set storage_engine = InnoDB;
+set autocommit=1;
+drop table if exists t1;
+drop table if exists t2;
+drop table if exists t3;
+drop function if exists f2;
+drop procedure if exists bug12713_call;
+drop procedure if exists bug12713_dump_spvars;
+drop procedure if exists dummy;
+create table t1 (a int);
+create table t2 (a int unique);
+create table t3 (a int);
+set sql_mode=default;
+insert into t1 (a) values (1), (2);
+insert into t3 (a) values (1), (2);
+create function f2(x int) returns int
+begin
+insert into t2 (a) values (x);
+insert into t2 (a) values (x);
+return x;
+end|
+set autocommit=0;
+flush status;
+insert into t2 (a) values (1001);
+insert into t1 (a) values (f2(1));
+ERROR 23000: Duplicate entry '1' for key 'a'
+select * from t2;
+a
+1001
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1002);
+insert into t3 (a) select f2(2) from t1;
+ERROR 23000: Duplicate entry '2' for key 'a'
+select * from t2;
+a
+1002
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1003);
+update t1 set a= a + f2(3);
+ERROR 23000: Duplicate entry '3' for key 'a'
+select * from t2;
+a
+1003
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1004);
+update t1, t3 set t1.a = 0, t3.a = 0 where (f2(4) = 4) and (t1.a = t3.a);
+ERROR 23000: Duplicate entry '4' for key 'a'
+select * from t2;
+a
+1004
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1005);
+delete from t1 where (a = f2(5));
+ERROR 23000: Duplicate entry '5' for key 'a'
+select * from t2;
+a
+1005
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1006);
+delete from t1, t3 using t1, t3 where (f2(6) = 6) ;
+ERROR 23000: Duplicate entry '6' for key 'a'
+select * from t2;
+a
+1006
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1007);
+replace t1 values (f2(7));
+ERROR 23000: Duplicate entry '7' for key 'a'
+select * from t2;
+a
+1007
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1008);
+replace into t3 (a) select f2(8) from t1;
+ERROR 23000: Duplicate entry '8' for key 'a'
+select * from t2;
+a
+1008
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1009);
+select f2(9) from t1 ;
+ERROR 23000: Duplicate entry '9' for key 'a'
+select * from t2;
+a
+1009
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1010);
+show databases where (f2(10) = 10);
+ERROR 23000: Duplicate entry '10' for key 'a'
+select * from t2;
+a
+1010
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1011);
+show tables where (f2(11) = 11);
+ERROR 23000: Duplicate entry '11' for key 'a'
+select * from t2;
+a
+1011
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1012);
+show triggers where (f2(12) = 12);
+ERROR 23000: Duplicate entry '12' for key 'a'
+select * from t2;
+a
+1012
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1013);
+show table status where (f2(13) = 13);
+ERROR 23000: Duplicate entry '13' for key 'a'
+select * from t2;
+a
+1013
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1014);
+show open tables where (f2(14) = 14);
+ERROR 23000: Duplicate entry '14' for key 'a'
+select * from t2;
+a
+1014
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1015);
+show columns in mysql.proc where (f2(15) = 15);
+ERROR 23000: Duplicate entry '15' for key 'a'
+select * from t2;
+a
+1015
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1016);
+show status where (f2(16) = 16);
+ERROR 23000: Duplicate entry '16' for key 'a'
+select * from t2;
+a
+1016
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1017);
+show variables where (f2(17) = 17);
+ERROR 23000: Duplicate entry '17' for key 'a'
+select * from t2;
+a
+1017
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1018);
+show charset where (f2(18) = 18);
+ERROR 23000: Duplicate entry '18' for key 'a'
+select * from t2;
+a
+1018
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1019);
+show collation where (f2(19) = 19);
+ERROR 23000: Duplicate entry '19' for key 'a'
+select * from t2;
+a
+1019
+rollback;
+select * from t2;
+a
+# We need at least one procedure to make sure the WHERE clause is
+# evaluated
+create procedure dummy() begin end;
+insert into t2 (a) values (1020);
+show procedure status where (f2(20) = 20);
+ERROR 23000: Duplicate entry '20' for key 'a'
+select * from t2;
+a
+1020
+rollback;
+select * from t2;
+a
+drop procedure dummy;
+insert into t2 (a) values (1021);
+show function status where (f2(21) = 21);
+ERROR 23000: Duplicate entry '21' for key 'a'
+select * from t2;
+a
+1021
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1022);
+prepare stmt from "insert into t1 (a) values (f2(22))";
+execute stmt;
+ERROR 23000: Duplicate entry '22' for key 'a'
+select * from t2;
+a
+1022
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1023);
+do (f2(23));
+Warnings:
+Error	1062	Duplicate entry '23' for key 'a'
+select * from t2;
+a
+1023
+rollback;
+select * from t2;
+a
+create procedure bug12713_call ()
+begin
+insert into t2 (a) values (24);
+insert into t2 (a) values (24);
+end|
+insert into t2 (a) values (1024);
+call bug12713_call();
+ERROR 23000: Duplicate entry '24' for key 'a'
+select * from t2;
+a
+24
+1024
+rollback;
+select * from t2;
+a
+=======================================================================
+Testing select_to_file
+=======================================================================
+insert into t2 (a) values (1025);
+select f2(25) into outfile "../tmp/dml.out" from t1;
+ERROR 23000: Duplicate entry '25' for key 'a'
+select * from t2;
+a
+1025
+rollback;
+select * from t2;
+a
+insert into t2 (a) values (1026);
+load data infile "../std_data_ln/words.dat" into table t1 (a) set a:=f2(26);
+ERROR 23000: Duplicate entry '26' for key 'a'
+select * from t2;
+a
+1026
+rollback;
+select * from t2;
+a
+=======================================================================
+Testing select_dumpvar
+=======================================================================
+insert into t2 (a) values (1027);
+select f2(27) into @foo;
+ERROR 23000: Duplicate entry '27' for key 'a'
+select * from t2;
+a
+1027
+rollback;
+select * from t2;
+a
+=======================================================================
+Testing Select_fetch_into_spvars 
+=======================================================================
+create procedure bug12713_dump_spvars ()
+begin
+declare foo int;
+declare continue handler for sqlexception
+begin
+select "Exception trapped";
+end;
+select f2(28) into foo;
+select * from t2;
+end|
+insert into t2 (a) values (1028);
+call bug12713_dump_spvars ();
+Exception trapped
+Exception trapped
+a
+1028
+rollback;
+select * from t2;
+a
+=======================================================================
+Cleanup
+=======================================================================
+set autocommit=default;
+drop table t1;
+drop table t2;
+drop table t3;
+drop function f2;
+drop procedure bug12713_call;
+drop procedure bug12713_dump_spvars;
+#
+# Bug#12713 Error in a stored function called from a SELECT doesn't
+# cause ROLLBACK of statem
+#
+# Verify that two-phase commit is not issued for read-only
+# transactions.
+#
+# Verify that two-phase commit is issued for read-write transactions,
+# even if the change is done inside a stored function called from
+# SELECT or SHOW statement.
+#
+set autocommit=0;
+drop table if exists t1;
+drop table if exists t2;
+drop function if exists f1;
+drop procedure if exists p_verify_status_increment;
+set sql_mode=no_engine_substitution;
+create table t1 (a int unique);
+create table t2 (a int) engine=myisam;
+set sql_mode=default;
+#
+# An auxiliary procedure to track Handler_prepare and Handler_commit
+# statistics.
+#
+create procedure
+p_verify_status_increment(commit_inc_mixed int, prepare_inc_mixed int,
+commit_inc_row int, prepare_inc_row int)
+begin
+declare commit_inc int;
+declare prepare_inc int;
+declare old_commit_count int default ifnull(@commit_count, 0);
+declare old_prepare_count int default ifnull(@prepare_count, 0);
+declare c_res int;
+# Use a cursor to have just one access to I_S instead of 2, it is very slow
+# and amounts for over 90% of test CPU time
+declare c cursor for
+select variable_value
+from information_schema.session_status
+where variable_name='Handler_commit' or variable_name='Handler_prepare'
+     order by variable_name;
+if @@global.binlog_format = 'ROW' then
+set commit_inc= commit_inc_row;
+set prepare_inc= prepare_inc_row;
+else
+set commit_inc= commit_inc_mixed;
+set prepare_inc= prepare_inc_mixed;
+end if;
+open c;
+fetch c into c_res;
+set @commit_count=c_res;
+fetch c into c_res;
+set @prepare_count=c_res;
+close c;
+if old_commit_count + commit_inc <> @commit_count then
+select concat("Expected commit increment: ", commit_inc,
+" actual: ", @commit_count - old_commit_count)
+as 'ERROR';
+elseif old_prepare_count + prepare_inc <> @prepare_count then
+select concat("Expected prepare increment: ", prepare_inc,
+" actual: ", @prepare_count - old_prepare_count)
+as 'ERROR';
+else
+select '' as 'SUCCESS';
+end if;
+end|
+# Reset Handler_commit and Handler_prepare counters
+flush status;
+#
+# 1. Read-only statement: SELECT
+#
+select * from t1;
+a
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 2. Read-write statement: INSERT, insert 1 row. 
+#
+insert into t1 (a) values (1);
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 3. Read-write statement: UPDATE, update 1 row. 
+#
+update t1 set a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 4. Read-write statement: UPDATE, update 0 rows, 1 row matches WHERE 
+#
+# Note the wrong Handler_prepare/Handler_commit count is due to
+# Bug#29157 "UPDATE, changed rows incorrect" and
+# Bug#Bug #33846   UPDATE word:Wrong 'Changed rows' if InnoDB, unique
+# key and no rows qualify WHERE
+#
+update t1 set a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 5. Read-write statement: UPDATE, update 0 rows, 0 rows match WHERE 
+#
+# In mixed replication mode, there is a read-only transaction
+# in InnoDB and also the statement is written to the binary log.
+# So we have two commits but no 2pc, since the first engine's
+# transaction is read-only.
+# In the row level replication mode, we only have the read-only
+# transaction in InnoDB and nothing is written to the binary log.
+#
+update t1 set a=3 where a=1;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 6. Read-write statement: DELETE, delete 0 rows. 
+#
+delete from t1 where a=1;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 7. Read-write statement: DELETE, delete 1 row. 
+#
+delete from t1 where a=2;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 8. Read-write statement: unqualified DELETE
+#
+# In statement or mixed replication mode, we call
+# handler::ha_delete_all_rows() and write statement text
+# to the binary log. This results in two read-write transactions.
+# In row level replication mode, we do not call
+# handler::ha_delete_all_rows(), but delete rows one by one.
+# Since there are no rows, nothing is written to the binary log.
+# Thus we have just one read-only transaction in InnoDB.
+delete from t1;
+call p_verify_status_increment(2, 2, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 1, 0);
+SUCCESS
+
+# 9. Read-write statement: REPLACE, change 1 row. 
+#
+replace t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 10. Read-write statement: REPLACE, change 0 rows. 
+#
+replace t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 11. Read-write statement: IODKU, change 1 row. 
+#
+insert t1 set a=1 on duplicate key update a=a+1;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+select * from t1;
+a
+2
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 12. Read-write statement: IODKU, change 0 rows. 
+#
+insert t1 set a=2 on duplicate key update a=2;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 13. Read-write statement: INSERT IGNORE, change 0 rows. 
+#
+insert ignore t1 set a=2;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 14. Read-write statement: INSERT IGNORE, change 1 row. 
+#
+insert ignore t1 set a=1;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 15. Read-write statement: UPDATE IGNORE, change 0 rows. 
+#
+update ignore t1 set a=2 where a=1;
+call p_verify_status_increment(2, 2, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 1, 0);
+SUCCESS
+
+#
+# Create a stored function that modifies a
+# non-transactional table. Demonstrate that changes in
+# non-transactional tables do not affect the two phase commit
+# algorithm.
+#
+create function f1() returns int
+begin
+insert t2 set a=2;
+return 2;
+end|
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+# 16. A function changes non-trans-table.
+#
+select f1();
+f1()
+2
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+# 17. Read-only statement, a function changes non-trans-table.
+#
+select f1() from t1;
+f1()
+2
+2
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 18. Read-write statement: UPDATE, change 0 (transactional) rows. 
+#
+select count(*) from t2;
+count(*)
+3
+update t1 set a=2 where a=f1()+10;
+select count(*) from t2;
+count(*)
+5
+call p_verify_status_increment(2, 0, 2, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 2, 0);
+SUCCESS
+
+#
+# Replace the non-transactional table with a temporary
+# transactional table. Demonstrate that a change to a temporary
+# transactional table does not provoke 2-phase commit, although
+# does trigger a commit and a binlog write (in statement mode).
+#
+drop table t2;
+set sql_mode=no_engine_substitution;
+create temporary table t2 (a int);
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+set sql_mode=default;
+# 19. A function changes temp-trans-table.
+#
+select f1();
+f1()
+2
+# Two commits because a binary log record is written
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 20. Read-only statement, a function changes non-trans-table.
+#
+select f1() from t1;
+f1()
+2
+2
+# Two commits because a binary log record is written
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 21. Read-write statement: UPDATE, change 0 (transactional) rows. 
+#
+update t1 set a=2 where a=f1()+10;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 22. DDL: ALTER TEMPORARY TABLE, should not cause a 2pc
+#
+alter table t2 add column b int default 5;
+# A commit is done internally by ALTER. 
+call p_verify_status_increment(2, 0, 2, 0);
+SUCCESS
+
+commit;
+# There is nothing left to commit
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+# 23. DDL: RENAME TEMPORARY TABLE, does not start a transaction
+
+# No test because of Bug#8729 "rename table fails on temporary table"
+# 24. DDL: TRUNCATE TEMPORARY TABLE, does not start a transaction
+
+truncate table t2;
+call p_verify_status_increment(2, 0, 2, 0);
+SUCCESS
+
+commit;
+# There is nothing left to commit
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+# 25. Read-write statement: unqualified DELETE 
+
+delete from t2;
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+commit;
+# There is nothing left to commit
+call p_verify_status_increment(2, 0, 1, 0);
+SUCCESS
+
+# 25. DDL: DROP TEMPORARY TABLE, does not start a transaction
+#
+drop temporary table t2;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+# 26. Verify that SET AUTOCOMMIT issues an implicit commit
+#
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+set autocommit=1;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+rollback;
+select a from t1 where a=3;
+a
+3
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+delete from t1 where a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+set autocommit=0;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# Sic: not actually changing the value of autocommit
+set autocommit=0;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+rollback;
+select a from t1 where a=3;
+a
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 27. Savepoint management
+#
+insert t1 set a=3;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+savepoint a;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+insert t1 set a=4;
+# Sic: a bug. Binlog did not register itself this time.
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+release savepoint a;
+rollback;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+select a from t1 where a=3;
+a
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+# 28. Read-write statement: DO
+#
+create table t2 (a int);
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+do (select f1() from t1 where a=2);
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+commit;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+# 29. Read-write statement: MULTI-DELETE
+# 
+delete t1, t2 from t1 join t2 on (t1.a=t2.a) where t1.a=2;
+commit;
+call p_verify_status_increment(4, 4, 4, 4);
+SUCCESS
+
+# 30. Read-write statement: INSERT-SELECT, MULTI-UPDATE, REPLACE-SELECT
+# 
+insert into t2 select a from t1;
+commit;
+replace into t2 select a from t1;
+commit;
+call p_verify_status_increment(8, 8, 8, 8);
+SUCCESS
+
+update t1, t2 set t1.a=4, t2.a=8 where t1.a=t2.a and t1.a=1;
+commit;
+call p_verify_status_increment(4, 4, 4, 4);
+SUCCESS
+
+# 31. DDL: various DDL with transactional tables
+#
+# Sic: no table is created.
+create table if not exists t2 (a int) select 6 union select 7;
+Warnings:
+Note	1050	Table 't2' already exists
+# Sic: first commits the statement, and then the transaction.
+call p_verify_status_increment(4, 4, 4, 4);
+SUCCESS
+
+create table t3 select a from t2;
+call p_verify_status_increment(4, 4, 4, 4);
+SUCCESS
+
+alter table t3 add column (b int);
+call p_verify_status_increment(2, 0, 2, 0);
+SUCCESS
+
+alter table t3 rename t4;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+rename table t4 to t3;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+truncate table t3;
+call p_verify_status_increment(2, 2, 2, 2);
+SUCCESS
+
+create view v1 as select * from t2;
+call p_verify_status_increment(1, 0, 1, 0);
+SUCCESS
+
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+call p_verify_status_increment(3, 0, 3, 0);
+SUCCESS
+
+# Sic: after this bug is fixed, CHECK leaves no pending transaction
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+check table t1, t2, t3;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+test.t2	check	status	OK
+test.t3	check	status	OK
+call p_verify_status_increment(6, 0, 6, 0);
+SUCCESS
+
+commit;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+drop view v1;
+call p_verify_status_increment(0, 0, 0, 0);
+SUCCESS
+
+#
+# Cleanup
+#
+drop table t1;
+drop procedure p_verify_status_increment;
+drop function f1;
diff -Nrup a/mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result b/mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result
--- a/mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result	2007-10-14 00:12:45 +04:00
+++ b/mysql-test/suite/binlog/r/binlog_row_mix_innodb_myisam.result	2008-02-19 14:42:57 +03:00
@@ -373,7 +373,7 @@ master-bin.000001	#	Write_rows	#	#	table
 master-bin.000001	#	Query	#	#	use `test`; BEGIN
 master-bin.000001	#	Table_map	#	#	table_id: # (test.t2)
 master-bin.000001	#	Write_rows	#	#	table_id: # flags: STMT_END_F
-master-bin.000001	#	Xid	#	#	COMMIT /* XID */
+master-bin.000001	#	Query	#	#	use `test`; COMMIT
 master-bin.000001	#	Query	#	#	use `test`; DROP TABLE t2
 master-bin.000001	#	Table_map	#	#	table_id: # (test.t1)
 master-bin.000001	#	Write_rows	#	#	table_id: # flags: STMT_END_F
@@ -384,7 +384,6 @@ master-bin.000001	#	Write_rows	#	#	table
 master-bin.000001	#	Table_map	#	#	table_id: # (test.t1)
 master-bin.000001	#	Write_rows	#	#	table_id: # flags: STMT_END_F
 master-bin.000001	#	Query	#	#	use `test`; TRUNCATE table t2
-master-bin.000001	#	Xid	#	#	COMMIT /* XID */
 master-bin.000001	#	Table_map	#	#	table_id: # (test.t1)
 master-bin.000001	#	Write_rows	#	#	table_id: # flags: STMT_END_F
 master-bin.000001	#	Query	#	#	use `test`; DROP TABLE `t1` /* generated by server */
diff -Nrup a/mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result b/mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result
--- a/mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result	2007-10-14 00:12:46 +04:00
+++ b/mysql-test/suite/binlog/r/binlog_stm_mix_innodb_myisam.result	2008-02-19 14:42:57 +03:00
@@ -244,7 +244,6 @@ master-bin.000001	#	Query	#	#	use `test`
 master-bin.000001	#	Query	#	#	use `test`; drop table t1,t2
 master-bin.000001	#	Query	#	#	use `test`; create temporary table ti (a int) engine=innodb
 master-bin.000001	#	Query	#	#	use `test`; insert into ti values(1)
-master-bin.000001	#	Xid	#	#	COMMIT /* XID */
 master-bin.000001	#	Query	#	#	use `test`; create temporary table t1 (a int) engine=myisam
 master-bin.000001	#	Query	#	#	use `test`; insert t1 values (1)
 master-bin.000001	#	Query	#	#	use `test`; create table t0 (n int)
@@ -349,11 +348,10 @@ master-bin.000001	#	Query	#	#	use `test`
 master-bin.000001	#	Query	#	#	use `test`; INSERT INTO t1 values (8,8)
 master-bin.000001	#	Query	#	#	use `test`; INSERT INTO t1 values (9,9)
 master-bin.000001	#	Query	#	#	use `test`; TRUNCATE table t2
-master-bin.000001	#	Xid	#	#	COMMIT /* XID */
 master-bin.000001	#	Query	#	#	use `test`; INSERT INTO t1 values (10,10)
 master-bin.000001	#	Query	#	#	use `test`; BEGIN
 master-bin.000001	#	Query	#	#	use `test`; INSERT INTO t2 values (100,100)
-master-bin.000001	#	Xid	#	#	COMMIT /* XID */
+master-bin.000001	#	Query	#	#	use `test`; COMMIT
 master-bin.000001	#	Query	#	#	use `test`; DROP TABLE t1,t2
 reset master;
 create table t1 (a int) engine=innodb;
diff -Nrup a/mysql-test/suite/rpl_ndb/t/disabled.def b/mysql-test/suite/rpl_ndb/t/disabled.def
--- a/mysql-test/suite/rpl_ndb/t/disabled.def	2007-12-11 13:51:00 +03:00
+++ b/mysql-test/suite/rpl_ndb/t/disabled.def	2008-02-19 14:42:58 +03:00
@@ -21,3 +21,5 @@ rpl_ndb_mix_innodb       : Bug #32720   
 # the below testcase have been reworked to avoid the bug, test contains comment, keep bug open
 
 #rpl_ndb_dd_advance	 : Bug#25913 rpl_ndb_dd_advance fails randomly
+rpl_ndb_circular         : Bug#33849 COMMIT event missing in cluster circular replication.
+rpl_ndb_circular_simplex : Bug#33849 COMMIT event missing in cluster circular replication.
diff -Nrup a/mysql-test/t/commit_1innodb.test b/mysql-test/t/commit_1innodb.test
--- /dev/null	Wed Dec 31 16:00:00 196900
+++ b/mysql-test/t/commit_1innodb.test	2008-02-19 14:42:59 +03:00
@@ -0,0 +1,6 @@
+-- source include/have_log_bin.inc
+-- source include/have_innodb.inc
+
+let $engine_type = InnoDB;
+
+-- source include/commit.inc
diff -Nrup a/sql/ha_ndbcluster_binlog.cc b/sql/ha_ndbcluster_binlog.cc
--- a/sql/ha_ndbcluster_binlog.cc	2007-12-15 22:23:55 +03:00
+++ b/sql/ha_ndbcluster_binlog.cc	2008-02-19 14:42:58 +03:00
@@ -283,6 +283,7 @@ static void run_query(THD *thd, char *bu
                       thd_ndb->m_error_code,
                       (int) thd->is_error(), thd->is_slave_error);
   }
+  close_thread_tables(thd);
   /*
     XXX: this code is broken. mysql_parse()/mysql_reset_thd_for_next_command()
     can not be called from within a statement, and
diff -Nrup a/sql/handler.cc b/sql/handler.cc
--- a/sql/handler.cc	2008-01-25 20:37:27 +03:00
+++ b/sql/handler.cc	2008-02-19 14:42:58 +03:00
@@ -576,6 +576,295 @@ void ha_close_connection(THD* thd)
  ======================= TRANSACTIONS ===================================*/
 
 /**
+  Transaction handling in the server
+  ==================================
+
+  In each client connection, MySQL maintains two transactional
+  states:
+  - a statement transaction,
+  - a standard, also called normal transaction.
+
+  Historical note
+  ---------------
+  "Statement transaction" is a non-standard term that comes
+  from the times when MySQL supported BerkeleyDB storage engine.
+
+  First of all, it should be said that in BerkeleyDB auto-commit
+  mode auto-commits operations that are atomic to the storage
+  engine itself, such as a write of a record, and are too
+  high-granular to be atomic from the application perspective
+  (MySQL). One SQL statement could involve many BerkeleyDB
+  auto-committed operations and thus BerkeleyDB auto-commit was of
+  little use to MySQL.
+
+  Secondly, instead of SQL standard savepoints, BerkeleyDB
+  provided the concept of "nested transactions". In a nutshell,
+  transactions could be arbitrarily nested, but when the parent
+  transaction was committed or aborted, all its child (nested)
+  transactions were handled committed or aborted as well.
+  Commit of a nested transaction, in turn, made its changes
+  visible, but not durable: it destroyed the nested transaction,
+  all its changes would become available to the parent and
+  currently active nested transactions of this parent.
+
+  So the mechanism of nested transactions was employed to
+  provide "all or nothing" guarantee of SQL statements
+  required by the standard.
+  A nested transaction would be created at start of each SQL
+  statement, and destroyed (committed or aborted) at statement
+  end. Such nested transaction was internally referred to as
+  a "statement transaction" and gave birth to the term.
+
+  <Historical note ends>
+
+  Since then a statement transaction is started for each statement
+  that accesses transactional tables or uses the binary log.  If
+  the statement succeeds, the statement transaction is committed.
+  If the statement fails, the transaction is rolled back. Commits
+  of statement transactions are not durable -- each such
+  transaction is nested in the normal transaction, and if the
+  normal transaction is rolled back, the effects of all enclosed
+  statement transactions are undone as well.  Technically,
+  a statement transaction can be viewed as a savepoint which is
+  maintained automatically in order to make effects of one
+  statement atomic.
+
+  The normal transaction is started by the user and is ended
+  usually upon a user request as well. The normal transaction
+  encloses transactions of all statements issued between
+  its beginning and its end.
+  In autocommit mode, the normal transaction is equivalent
+  to the statement transaction.
+
+  Since MySQL supports PSEA (pluggable storage engine
+  architecture), more than one transactional engine can be
+  active at a time. Hence transactions, from the server
+  point of view, are always distributed. In particular,
+  transactional state is maintained independently for each
+  engine. In order to commit a transaction the two phase
+  commit protocol is employed.
+
+  Not all statements are executed in context of a transaction.
+  Administrative and status information statements do not modify
+  engine data, and thus do not start a statement transaction and
+  also have no effect on the normal transaction. Examples of such
+  statements are SHOW STATUS and RESET SLAVE.
+
+  Similarly DDL statements are not transactional,
+  and therefore a transaction is [almost] never started for a DDL
+  statement. The difference between a DDL statement and a purely
+  administrative statement though is that a DDL statement always
+  commits the current transaction before proceeding, if there is
+  any.
+
+  At last, SQL statements that work with non-transactional
+  engines also have no effect on the transaction state of the
+  connection. Even though they are written to the binary log,
+  and the binary log is, overall, transactional, the writes
+  are done in "write-through" mode, directly to the binlog
+  file, followed with a OS cache sync, in other words,
+  bypassing the binlog undo log (translog).
+  They do not commit the current normal transaction.
+  A failure of a statement that uses non-transactional tables
+  would cause a rollback of the statement transaction, but
+  in case there no non-transactional tables are used,
+  no statement transaction is started.
+
+  Data layout
+  -----------
+
+  The server stores its transaction-related data in
+  thd->transaction. This structure has two members of type
+  THD_TRANS. These members correspond to the statement and
+  normal transactions respectively:
+
+  - thd->transaction.stmt contains a list of engines
+  that are participating in the given statement
+  - thd->transaction.all contains a list of engines that
+  have participated in any of the statement transactions started
+  within the context of the normal transaction.
+  Each element of the list contains a pointer to the storage
+  engine, engine-specific transactional data, and engine-specific
+  transaction flags.
+
+  In autocommit mode thd->transaction.all is empty.
+  Instead, data of thd->transaction.stmt is
+  used to commit/rollback the normal transaction.
+
+  The list of registered engines has a few important properties:
+  - no engine is registered in the list twice
+  - engines are present in the list a reverse temporal order --
+  new participants are always added to the beginning of the list.
+
+  Transaction life cycle
+  ----------------------
+
+  When a new connection is established, thd->transaction
+  members are initialized to an empty state.
+  If a statement uses any tables, all affected engines
+  are registered in the statement engine list. In
+  non-autocommit mode, the same engines are registered in
+  the normal transaction list.
+  At the end of the statement, the server issues a commit
+  or a roll back for all engines in the statement list.
+  At this point transaction flags of an engine, if any, are
+  propagated from the statement list to the list of the normal
+  transaction.
+  When commit/rollback is finished, the statement list is
+  cleared. It will be filled in again by the next statement,
+  and emptied again at the next statement's end.
+
+  The normal transaction is committed in a similar way
+  (by going over all engines in thd->transaction.all list)
+  but at different times:
+  - upon COMMIT SQL statement is issued by the user
+  - implicitly, by the server, at the beginning of a DDL statement
+  or SET AUTOCOMMIT={0|1} statement.
+
+  The normal transaction can be rolled back as well:
+  - if the user has requested so, by issuing ROLLBACK SQL
+  statement
+  - if one of the storage engines requested a rollback
+  by setting thd->transaction_rollback_request. This may
+  happen in case, e.g., when the transaction in the engine was
+  chosen a victim of the internal deadlock resolution algorithm
+  and rolled back internally. When such a situation happens, there
+  is little the server can do and the only option is to rollback
+  transactions in all other participating engines.  In this case
+  the rollback is accompanied by an error sent to the user.
+
+  As follows from the use cases above, the normal transaction
+  is never committed when there is an outstanding statement
+  transaction. In most cases there is no conflict, since
+  commits of the normal transaction are issued by a stand-alone
+  administrative or DDL statement, thus no outstanding statement
+  transaction of the previous statement exists. Besides,
+  all statements that manipulate with the normal transaction
+  are prohibited in stored functions and triggers, therefore
+  no conflicting situation can occur in a sub-statement either.
+  The remaining rare cases when the server explicitly has
+  to commit the statement transaction prior to committing the normal
+  one cover error-handling scenarios (see for example
+  SQLCOM_LOCK_TABLES).
+
+  When committing a statement or a normal transaction, the server
+  either uses the two-phase commit protocol, or issues a commit
+  in each engine independently. The two-phase commit protocol
+  is used only if:
+  - all participating engines support two-phase commit (provide
+    handlerton::prepare PSEA API call) and
+  - transactions in at least two engines modify data (i.e. are
+  not read-only).
+
+  Note that the two phase commit is used for
+  statement transactions, even though they are not durable anyway.
+  This is done to ensure logical consistency of data in a multiple-
+  engine transaction.
+  For example, imagine that some day MySQL supports unique
+  constraint checks deferred till the end of statement. In such
+  case a commit in one of the engines may yield ER_DUP_KEY,
+  and MySQL should be able to gracefully abort statement
+  transactions of other participants.
+
+  After the normal transaction has been committed,
+  thd->transaction.all list is cleared.
+
+  When a connection is closed, the current normal transaction, if
+  any, is rolled back.
+
+  Roles and responsibilities
+  --------------------------
+
+  The server has no way to know that an engine participates in
+  the statement and a transaction has been started
+  in it unless the engine says so. Thus, in order to be
+  a part of a transaction, the engine must "register" itself.
+  This is done by invoking trans_register_ha() server call.
+  Normally the engine registers itself whenever handler::external_lock()
+  is called. trans_register_ha() can be invoked many times: if
+  an engine is already registered, the call does nothing.
+  In case autocommit is not set, the engine must register itself
+  twice -- both in the statement list and in the normal transaction
+  list.
+  In which list to register is a parameter of trans_register_ha().
+
+  Note, that although the registration interface in itself is
+  fairly clear, the current usage practice often leads to undesired
+  effects. E.g. since a call to trans_register_ha() in most engines
+  is embedded into implementation of handler::external_lock(), some
+  DDL statements start a transaction (at least from the server
+  point of view) even though they are not expected to. E.g.
+  CREATE TABLE does not start a transaction, since
+  handler::external_lock() is never called during CREATE TABLE. But
+  CREATE TABLE ... SELECT does, since handler::external_lock() is
+  called for the table that is being selected from. This has no
+  practical effects currently, but must be kept in mind
+  nevertheless.
+
+  Once an engine is registered, the server will do the rest
+  of the work.
+
+  During statement execution, whenever any of data-modifying
+  PSEA API methods is used, e.g. handler::write_row() or
+  handler::update_row(), the read-write flag is raised in the
+  statement transaction for the involved engine.
+  Currently All PSEA calls are "traced", and the data can not be
+  changed in a way other than issuing a PSEA call. Important:
+  unless this invariant is preserved the server will not know that
+  a transaction in a given engine is read-write and will not
+  involve the two-phase commit protocol!
+
+  At the end of a statement, server call
+  ha_autocommit_or_rollback() is invoked. This call in turn
+  invokes handlerton::prepare() for every involved engine.
+  Prepare is followed by a call to handlerton::commit_one_phase()
+  If a one-phase commit will suffice, handlerton::prepare() is not
+  invoked and the server only calls handlerton::commit_one_phase().
+  At statement commit, the statement-related read-write engine
+  flag is propagated to the corresponding flag in the normal
+  transaction.  When the commit is complete, the list of registered
+  engines is cleared.
+
+  Rollback is handled in a similar fashion.
+
+  Additional notes on DDL and the normal transaction.
+  ---------------------------------------------------
+
+  DDLs and operations with non-transactional engines
+  do not "register" in thd->transaction lists, and thus do not
+  modify the transaction state. Besides, each DDL in
+  MySQL is prefixed with an implicit normal transaction commit
+  (a call to end_active_trans()), and thus leaves nothing
+  to modify.
+  However, as it has been pointed out with CREATE TABLE .. SELECT,
+  some DDL statements can start a *new* transaction.
+
+  Behaviour of the server in this case is currently badly
+  defined.
+  DDL statements use a form of "semantic" logging
+  to maintain atomicity: if CREATE TABLE .. SELECT failed,
+  the newly created table is deleted.
+  In addition, some DDL statements issue interim transaction
+  commits: e.g. ALTER TABLE issues a commit after data is copied
+  from the original table to the internal temporary table. Other
+  statements, e.g. CREATE TABLE ... SELECT do not always commit
+  after itself.
+  And finally there is a group of DDL statements such as
+  RENAME/DROP TABLE that doesn't start a new transaction
+  and doesn't commit.
+
+  This diversity makes it hard to say what will happen if
+  by chance a stored function is invoked during a DDL --
+  whether any modifications it makes will be committed or not
+  is not clear. Fortunately, SQL grammar of few DDLs allows
+  invocation of a stored function.
+
+  A consistent behaviour is perhaps to always commit the normal
+  transaction after all DDLs, just like the statement transaction
+  is always committed at the end of all statements.
+*/
+
+/**
   Register a storage engine for a transaction.
 
   Every storage engine MUST call this function when it starts
@@ -592,7 +881,7 @@ void ha_close_connection(THD* thd)
 void trans_register_ha(THD *thd, bool all, handlerton *ht_arg)
 {
   THD_TRANS *trans;
-  handlerton **ht;
+  Ha_trx_info *ha_info;
   DBUG_ENTER("trans_register_ha");
   DBUG_PRINT("enter",("%s", all ? "all" : "stmt"));
 
@@ -604,12 +893,13 @@ void trans_register_ha(THD *thd, bool al
   else
     trans= &thd->transaction.stmt;
 
-  for (ht=trans->ht; *ht; ht++)
-    if (*ht == ht_arg)
-      DBUG_VOID_RETURN;  /* already registered, return */
+  ha_info= thd->ha_data[ht_arg->slot].ha_info + static_cast<unsigned>(all);
+
+  if (ha_info->is_started())
+    DBUG_VOID_RETURN; /* already registered, return */
+
+  ha_info->register_ha(trans, ht_arg);
 
-  trans->ht[trans->nht++]=ht_arg;
-  DBUG_ASSERT(*ht == ht_arg);
   trans->no_2pc|=(ht_arg->prepare==0);
   if (thd->transaction.xid_state.xid.is_null())
     thd->transaction.xid_state.xid.set(thd->query_id);
@@ -626,18 +916,19 @@ int ha_prepare(THD *thd)
 {
   int error=0, all=1;
   THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
-  handlerton **ht=trans->ht;
+  Ha_trx_info *ha_info= trans->ha_list;
   DBUG_ENTER("ha_prepare");
 #ifdef USING_TRANSACTIONS
-  if (trans->nht)
+  if (ha_info)
   {
-    for (; *ht; ht++)
+    for (; ha_info; ha_info= ha_info->next())
     {
       int err;
+      handlerton *ht= ha_info->ht();
       status_var_increment(thd->status_var.ha_prepare_count);
-      if ((*ht)->prepare)
+      if (ht->prepare)
       {
-        if ((err= (*(*ht)->prepare)(*ht, thd, all)))
+        if ((err= ht->prepare(ht, thd, all)))
         {
           my_error(ER_ERROR_DURING_COMMIT, MYF(0), err);
           ha_rollback_trans(thd, all);
@@ -649,7 +940,7 @@ int ha_prepare(THD *thd)
       {
         push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                             ER_ILLEGAL_HA, ER(ER_ILLEGAL_HA),
-                            ha_resolve_storage_engine_name(*ht));
+                            ha_resolve_storage_engine_name(ht));
       }
     }
   }
@@ -658,6 +949,62 @@ int ha_prepare(THD *thd)
 }
 
 /**
+  Check if we can skip the two-phase commit.
+
+  A helper function to evaluate if two-phase commit is mandatory.
+  As a side effect, propagates the read-only/read-write flags
+  of the statement transaction to its enclosing normal transaction.
+
+  @retval TRUE   we must run a two-phase commit. Returned
+                 if we have at least two engines with read-write changes.
+  @retval FALSE  Don't need two-phase commit. Even if we have two
+                 transactional engines, we can run two independent
+                 commits if changes in one of the engines are read-only.
+*/
+
+static
+bool
+ha_check_and_coalesce_trx_read_only(THD *thd, Ha_trx_info *ha_list,
+                                    bool all)
+{
+  /* The number of storage engines that have actual changes. */
+  unsigned rw_ha_count= 0;
+  Ha_trx_info *ha_info;
+
+  for (ha_info= ha_list; ha_info; ha_info= ha_info->next())
+  {
+    if (ha_info->is_trx_read_write())
+      ++rw_ha_count;
+
+    if (! all)
+    {
+      Ha_trx_info *ha_info_all= &thd->ha_data[ha_info->ht()->slot].ha_info[1];
+      DBUG_ASSERT(ha_info != ha_info_all);
+      /*
+        Merge read-only/read-write information about statement
+        transaction to its enclosing normal transaction. Do this
+        only if in a real transaction -- that is, if we know
+        that ha_info_all is registered in thd->transaction.all.
+        Since otherwise we only clutter the normal transaction flags.
+      */
+      if (ha_info_all->is_started()) /* FALSE if autocommit. */
+        ha_info_all->coalesce_trx_with(ha_info);
+    }
+    else if (rw_ha_count > 1)
+    {
+      /*
+        It is a normal transaction, so we don't need to merge read/write
+        information up, and the need for two-phase commit has been
+        already established. Break the loop prematurely.
+      */
+      break;
+    }
+  }
+  return rw_ha_count > 1;
+}
+
+
+/**
   @retval
     0   ok
   @retval
@@ -674,12 +1021,25 @@ int ha_prepare(THD *thd)
 int ha_commit_trans(THD *thd, bool all)
 {
   int error= 0, cookie= 0;
+  /*
+    'all' means that this is either an explicit commit issued by
+    user, or an implicit commit issued by a DDL.
+  */
   THD_TRANS *trans= all ? &thd->transaction.all : &thd->transaction.stmt;
-  bool is_real_trans= all || thd->transaction.all.nht == 0;
-  handlerton **ht= trans->ht;
+  bool is_real_trans= all || thd->transaction.all.ha_list == 0;
+  Ha_trx_info *ha_info= trans->ha_list;
   my_xid xid= thd->transaction.xid_state.xid.get_my_xid();
   DBUG_ENTER("ha_commit_trans");
 
+  /*
+    We must not commit the normal transaction if a statement
+    transaction is pending. Otherwise statement transaction
+    flags will not get propagated to its normal transaction's
+    counterpart.
+  */
+  DBUG_ASSERT(thd->transaction.stmt.ha_list == NULL ||
+              trans == &thd->transaction.stmt);
+
   if (thd->in_sub_stmt)
   {
     /*
@@ -701,8 +1061,10 @@ int ha_commit_trans(THD *thd, bool all)
     DBUG_RETURN(2);
   }
 #ifdef USING_TRANSACTIONS
-  if (trans->nht)
+  if (ha_info)
   {
+    bool must_2pc;
+
     if (is_real_trans && wait_if_global_read_lock(thd, 0, 0))
     {
       ha_rollback_trans(thd, all);
@@ -727,12 +1089,26 @@ int ha_commit_trans(THD *thd, bool all)
     if (is_real_trans)                          /* not a statement commit */
       thd->stmt_map.close_transient_cursors();
 
-    if (!trans->no_2pc && trans->nht > 1)
+    must_2pc= ha_check_and_coalesce_trx_read_only(thd, ha_info, all);
+
+    if (!trans->no_2pc && must_2pc)
     {
-      for (; *ht && !error; ht++)
+      for (; ha_info && !error; ha_info= ha_info->next())
       {
         int err;
-        if ((err= (*(*ht)->prepare)(*ht, thd, all)))
+        handlerton *ht= ha_info->ht();
+        /*
+          Do not call two-phase commit if this particular
+          transaction is read-only. This allows for simpler
+          implementation in engines that are always read-only.
+        */
+        if (! ha_info->is_trx_read_write())
+          continue;
+        /*
+          Sic: we know that prepare() is not NULL since otherwise
+          trans->no_2pc would have been set.
+        */
+        if ((err= ht->prepare(ht, thd, all)))
         {
           my_error(ER_ERROR_DURING_COMMIT, MYF(0), err);
           error= 1;
@@ -770,24 +1146,26 @@ int ha_commit_one_phase(THD *thd, bool a
 {
   int error=0;
   THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
-  bool is_real_trans=all || thd->transaction.all.nht == 0;
-  handlerton **ht=trans->ht;
+  bool is_real_trans=all || thd->transaction.all.ha_list == 0;
+  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;
   DBUG_ENTER("ha_commit_one_phase");
 #ifdef USING_TRANSACTIONS
-  if (trans->nht)
+  if (ha_info)
   {
-    for (ht=trans->ht; *ht; ht++)
+    for (; ha_info; ha_info= ha_info_next)
     {
       int err;
-      if ((err= (*(*ht)->commit)(*ht, thd, all)))
+      handlerton *ht= ha_info->ht();
+      if ((err= ht->commit(ht, thd, all)))
       {
         my_error(ER_ERROR_DURING_COMMIT, MYF(0), err);
         error=1;
       }
       status_var_increment(thd->status_var.ha_commit_count);
-      *ht= 0;
+      ha_info_next= ha_info->next();
+      ha_info->reset(); /* keep it conveniently zero-filled */
     }
-    trans->nht=0;
+    trans->ha_list= 0;
     trans->no_2pc=0;
     if (is_real_trans)
       thd->transaction.xid_state.xid.null();
@@ -810,8 +1188,17 @@ int ha_rollback_trans(THD *thd, bool all
 {
   int error=0;
   THD_TRANS *trans=all ? &thd->transaction.all : &thd->transaction.stmt;
-  bool is_real_trans=all || thd->transaction.all.nht == 0;
+  Ha_trx_info *ha_info= trans->ha_list, *ha_info_next;
+  bool is_real_trans=all || thd->transaction.all.ha_list == 0;
   DBUG_ENTER("ha_rollback_trans");
+
+  /*
+    We must not rollback the normal transaction if a statement
+    transaction is pending.
+  */
+  DBUG_ASSERT(thd->transaction.stmt.ha_list == NULL ||
+              trans == &thd->transaction.stmt);
+
   if (thd->in_sub_stmt)
   {
     /*
@@ -826,24 +1213,26 @@ int ha_rollback_trans(THD *thd, bool all
     DBUG_RETURN(1);
   }
 #ifdef USING_TRANSACTIONS
-  if (trans->nht)
+  if (ha_info)
   {
     /* Close all cursors that can not survive ROLLBACK */
     if (is_real_trans)                          /* not a statement commit */
       thd->stmt_map.close_transient_cursors();
 
-    for (handlerton **ht=trans->ht; *ht; ht++)
+    for (; ha_info; ha_info= ha_info_next)
     {
       int err;
-      if ((err= (*(*ht)->rollback)(*ht, thd, all)))
+      handlerton *ht= ha_info->ht();
+      if ((err= ht->rollback(ht, thd, all)))
       { // cannot happen
         my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
         error=1;
       }
       status_var_increment(thd->status_var.ha_rollback_count);
-      *ht= 0;
+      ha_info_next= ha_info->next();
+      ha_info->reset(); /* keep it conveniently zero-filled */
     }
-    trans->nht=0;
+    trans->ha_list= 0;
     trans->no_2pc=0;
     if (is_real_trans)
       thd->transaction.xid_state.xid.null();
@@ -889,17 +1278,19 @@ int ha_autocommit_or_rollback(THD *thd, 
 {
   DBUG_ENTER("ha_autocommit_or_rollback");
 #ifdef USING_TRANSACTIONS
-  if (thd->transaction.stmt.nht)
+  if (thd->transaction.stmt.ha_list)
   {
     if (!error)
     {
-      if (ha_commit_stmt(thd))
+      if (ha_commit_trans(thd, 0))
 	error=1;
     }
-    else if (thd->transaction_rollback_request && !thd->in_sub_stmt)
-      (void) ha_rollback(thd);
-    else
-      (void) ha_rollback_stmt(thd);
+    else 
+    {
+      (void) ha_rollback_trans(thd, 0);
+      if (thd->transaction_rollback_request && !thd->in_sub_stmt)
+        (void) ha_rollback(thd);
+    }
 
     thd->variables.tx_isolation=thd->session_tx_isolation;
   }
@@ -1246,43 +1637,49 @@ int ha_rollback_to_savepoint(THD *thd, S
   int error=0;
   THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
                                         &thd->transaction.all);
-  handlerton **ht=trans->ht, **end_ht;
+  Ha_trx_info *ha_info, *ha_info_next;
+
   DBUG_ENTER("ha_rollback_to_savepoint");
 
-  trans->nht=sv->nht;
   trans->no_2pc=0;
-  end_ht=ht+sv->nht;
   /*
     rolling back to savepoint in all storage engines that were part of the
     transaction when the savepoint was set
   */
-  for (; ht < end_ht; ht++)
+  for (ha_info= sv->ha_list; ha_info; ha_info= ha_info->next())
   {
     int err;
-    DBUG_ASSERT((*ht)->savepoint_set != 0);
-    if ((err= (*(*ht)->savepoint_rollback)(*ht, thd, (uchar *)(sv+1)+(*ht)->savepoint_offset)))
+    handlerton *ht= ha_info->ht();
+    DBUG_ASSERT(ht);
+    DBUG_ASSERT(ht->savepoint_set != 0);
+    if ((err= ht->savepoint_rollback(ht, thd,
+                                     (uchar *)(sv+1)+ht->savepoint_offset)))
     { // cannot happen
       my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
       error=1;
     }
     status_var_increment(thd->status_var.ha_savepoint_rollback_count);
-    trans->no_2pc|=(*ht)->prepare == 0;
+    trans->no_2pc|= ht->prepare == 0;
   }
   /*
     rolling back the transaction in all storage engines that were not part of
     the transaction when the savepoint was set
   */
-  for (; *ht ; ht++)
+  for (ha_info= trans->ha_list; ha_info != sv->ha_list;
+       ha_info= ha_info_next)
   {
     int err;
-    if ((err= (*(*ht)->rollback)(*ht, thd, !thd->in_sub_stmt)))
+    handlerton *ht= ha_info->ht();
+    if ((err= ht->rollback(ht, thd, !thd->in_sub_stmt)))
     { // cannot happen
       my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err);
       error=1;
     }
     status_var_increment(thd->status_var.ha_rollback_count);
-    *ht=0; // keep it conveniently zero-filled
+    ha_info_next= ha_info->next();
+    ha_info->reset(); /* keep it conveniently zero-filled */
   }
+  trans->ha_list= sv->ha_list;
   DBUG_RETURN(error);
 }
 
@@ -1297,26 +1694,32 @@ int ha_savepoint(THD *thd, SAVEPOINT *sv
   int error=0;
   THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
                                         &thd->transaction.all);
-  handlerton **ht=trans->ht;
+  Ha_trx_info *ha_info= trans->ha_list;
   DBUG_ENTER("ha_savepoint");
 #ifdef USING_TRANSACTIONS
-  for (; *ht; ht++)
+  for (; ha_info; ha_info= ha_info->next())
   {
     int err;
-    if (! (*ht)->savepoint_set)
+    handlerton *ht= ha_info->ht();
+    DBUG_ASSERT(ht);
+    if (! ht->savepoint_set)
     {
       my_error(ER_CHECK_NOT_IMPLEMENTED, MYF(0), "SAVEPOINT");
       error=1;
       break;
     }
-    if ((err= (*(*ht)->savepoint_set)(*ht, thd, (uchar *)(sv+1)+(*ht)->savepoint_offset)))
+    if ((err= ht->savepoint_set(ht, thd, (uchar *)(sv+1)+ht->savepoint_offset)))
     { // cannot happen
       my_error(ER_GET_ERRNO, MYF(0), err);
       error=1;
     }
     status_var_increment(thd->status_var.ha_savepoint_count);
   }
-  sv->nht=trans->nht;
+  /*
+    Remember the list of registered storage engines. All new
+    engines are prepended to the beginning of the list.
+  */
+  sv->ha_list= trans->ha_list;
 #endif /* USING_TRANSACTIONS */
   DBUG_RETURN(error);
 }
@@ -1324,20 +1727,19 @@ int ha_savepoint(THD *thd, SAVEPOINT *sv
 int ha_release_savepoint(THD *thd, SAVEPOINT *sv)
 {
   int error=0;
-  THD_TRANS *trans= (thd->in_sub_stmt ? &thd->transaction.stmt :
-                                        &thd->transaction.all);
-  handlerton **ht=trans->ht, **end_ht;
+  Ha_trx_info *ha_info= sv->ha_list;
   DBUG_ENTER("ha_release_savepoint");
 
-  end_ht=ht+sv->nht;
-  for (; ht < end_ht; ht++)
+  for (; ha_info; ha_info= ha_info->next())
   {
     int err;
-    if (!(*ht)->savepoint_release)
+    handlerton *ht= ha_info->ht();
+    /* Savepoint life time is enclosed into transaction life time. */
+    DBUG_ASSERT(ht);
+    if (!ht->savepoint_release)
       continue;
-    if ((err= (*(*ht)->savepoint_release)(*ht, thd, 
-                                          (uchar *)(sv+1)+
-                                          (*ht)->savepoint_offset)))
+    if ((err= ht->savepoint_release(ht, thd,
+                                    (uchar *)(sv+1) + ht->savepoint_offset)))
     { // cannot happen
       my_error(ER_GET_ERRNO, MYF(0), err);
       error=1;
@@ -2506,6 +2908,36 @@ int handler::ha_check(THD *thd, HA_CHECK
   return update_frm_version(table);
 }
 
+/**
+  A helper function to mark a transaction read-write,
+  if it is started.
+*/
+
+inline
+void
+handler::mark_trx_read_write()
+{
+  Ha_trx_info *ha_info= &ha_thd()->ha_data[ht->slot].ha_info[0];
+  /*
+    When a storage engine method is called, the transaction must
+    have been started, unless it's a DDL call, for which the
+    storage engine starts the transaction internally, and commits
+    it internally, without registering in the ha_list.
+    Unfortunately here we can't know know for sure if the engine
+    has registered the transaction or not, so we must check.
+  */
+  if (ha_info->is_started())
+  {
+    DBUG_ASSERT(has_transactions());
+    /*
+      table_share can be NULL in ha_delete_table(). See implementation
+      of standalone function ha_delete_table() in sql_base.cc.
+    */
+    if (table_share == NULL || table_share->tmp_table == NO_TMP_TABLE)
+      ha_info->set_trx_read_write();
+  }
+}
+
 
 /**
   Repair table: public interface.
@@ -2516,6 +2948,9 @@ int handler::ha_check(THD *thd, HA_CHECK
 int handler::ha_repair(THD* thd, HA_CHECK_OPT* check_opt)
 {
   int result;
+
+  mark_trx_read_write();
+
   if ((result= repair(thd, check_opt)))
     return result;
   return update_frm_version(table);
@@ -2532,6 +2967,8 @@ int
 handler::ha_bulk_update_row(const uchar *old_data, uchar *new_data,
                             uint *dup_key_found)
 {
+  mark_trx_read_write();
+
   return bulk_update_row(old_data, new_data, dup_key_found);
 }
 
@@ -2545,6 +2982,8 @@ handler::ha_bulk_update_row(const uchar 
 int
 handler::ha_delete_all_rows()
 {
+  mark_trx_read_write();
+
   return delete_all_rows();
 }
 
@@ -2558,6 +2997,8 @@ handler::ha_delete_all_rows()
 int
 handler::ha_reset_auto_increment(ulonglong value)
 {
+  mark_trx_read_write();
+
   return reset_auto_increment(value);
 }
 
@@ -2571,6 +3012,8 @@ handler::ha_reset_auto_increment(ulonglo
 int
 handler::ha_backup(THD* thd, HA_CHECK_OPT* check_opt)
 {
+  mark_trx_read_write();
+
   return backup(thd, check_opt);
 }
 
@@ -2584,6 +3027,8 @@ handler::ha_backup(THD* thd, HA_CHECK_OP
 int
 handler::ha_restore(THD* thd, HA_CHECK_OPT* check_opt)
 {
+  mark_trx_read_write();
+
   return restore(thd, check_opt);
 }
 
@@ -2597,6 +3042,8 @@ handler::ha_restore(THD* thd, HA_CHECK_O
 int
 handler::ha_optimize(THD* thd, HA_CHECK_OPT* check_opt)
 {
+  mark_trx_read_write();
+
   return optimize(thd, check_opt);
 }
 
@@ -2610,6 +3057,8 @@ handler::ha_optimize(THD* thd, HA_CHECK_
 int
 handler::ha_analyze(THD* thd, HA_CHECK_OPT* check_opt)
 {
+  mark_trx_read_write();
+
   return analyze(thd, check_opt);
 }
 
@@ -2623,6 +3072,8 @@ handler::ha_analyze(THD* thd, HA_CHECK_O
 bool
 handler::ha_check_and_repair(THD *thd)
 {
+  mark_trx_read_write();
+
   return check_and_repair(thd);
 }
 
@@ -2636,6 +3087,8 @@ handler::ha_check_and_repair(THD *thd)
 int
 handler::ha_disable_indexes(uint mode)
 {
+  mark_trx_read_write();
+
   return disable_indexes(mode);
 }
 
@@ -2649,6 +3102,8 @@ handler::ha_disable_indexes(uint mode)
 int
 handler::ha_enable_indexes(uint mode)
 {
+  mark_trx_read_write();
+
   return enable_indexes(mode);
 }
 
@@ -2662,6 +3117,8 @@ handler::ha_enable_indexes(uint mode)
 int
 handler::ha_discard_or_import_tablespace(my_bool discard)
 {
+  mark_trx_read_write();
+
   return discard_or_import_tablespace(discard);
 }
 
@@ -2677,6 +3134,8 @@ handler::ha_discard_or_import_tablespace
 void
 handler::ha_prepare_for_alter()
 {
+  mark_trx_read_write();
+
   prepare_for_alter();
 }
 
@@ -2690,6 +3149,8 @@ handler::ha_prepare_for_alter()
 int
 handler::ha_rename_table(const char *from, const char *to)
 {
+  mark_trx_read_write();
+
   return rename_table(from, to);
 }
 
@@ -2703,6 +3164,8 @@ handler::ha_rename_table(const char *fro
 int
 handler::ha_delete_table(const char *name)
 {
+  mark_trx_read_write();
+
   return delete_table(name);
 }
 
@@ -2716,6 +3179,8 @@ handler::ha_delete_table(const char *nam
 void
 handler::ha_drop_table(const char *name)
 {
+  mark_trx_read_write();
+
   return drop_table(name);
 }
 
@@ -2729,6 +3194,8 @@ handler::ha_drop_table(const char *name)
 int
 handler::ha_create(const char *name, TABLE *form, HA_CREATE_INFO *info)
 {
+  mark_trx_read_write();
+
   return create(name, form, info);
 }
 
@@ -2743,6 +3210,8 @@ int
 handler::ha_create_handler_files(const char *name, const char *old_name,
                         int action_flag, HA_CREATE_INFO *info)
 {
+  mark_trx_read_write();
+
   return create_handler_files(name, old_name, action_flag, info);
 }
 
@@ -2761,6 +3230,8 @@ handler::ha_change_partitions(HA_CREATE_
                      const uchar *pack_frm_data,
                      size_t pack_frm_len)
 {
+  mark_trx_read_write();
+
   return change_partitions(create_info, path, copied, deleted,
                            pack_frm_data, pack_frm_len);
 }
@@ -2775,6 +3246,8 @@ handler::ha_change_partitions(HA_CREATE_
 int
 handler::ha_drop_partitions(const char *path)
 {
+  mark_trx_read_write();
+
   return drop_partitions(path);
 }
 
@@ -2788,6 +3261,8 @@ handler::ha_drop_partitions(const char *
 int
 handler::ha_rename_partitions(const char *path)
 {
+  mark_trx_read_write();
+
   return rename_partitions(path);
 }
 
@@ -2801,6 +3276,8 @@ handler::ha_rename_partitions(const char
 int
 handler::ha_optimize_partitions(THD *thd)
 {
+  mark_trx_read_write();
+
   return optimize_partitions(thd);
 }
 
@@ -2814,6 +3291,8 @@ handler::ha_optimize_partitions(THD *thd
 int
 handler::ha_analyze_partitions(THD *thd)
 {
+  mark_trx_read_write();
+
   return analyze_partitions(thd);
 }
 
@@ -2827,6 +3306,8 @@ handler::ha_analyze_partitions(THD *thd)
 int
 handler::ha_check_partitions(THD *thd)
 {
+  mark_trx_read_write();
+
   return check_partitions(thd);
 }
 
@@ -2840,6 +3321,8 @@ handler::ha_check_partitions(THD *thd)
 int
 handler::ha_repair_partitions(THD *thd)
 {
+  mark_trx_read_write();
+
   return repair_partitions(thd);
 }
 
@@ -2866,7 +3349,7 @@ int ha_enable_transaction(THD *thd, bool
       is an optimization hint that storage engine is free to ignore.
       So, let's commit an open transaction (if any) now.
     */
-    if (!(error= ha_commit_stmt(thd)))
+    if (!(error= ha_commit_trans(thd, 0)))
       error= end_trans(thd, COMMIT);
   }
   DBUG_RETURN(error);
@@ -4042,6 +4525,9 @@ int handler::ha_write_row(uchar *buf)
 {
   int error;
   DBUG_ENTER("handler::ha_write_row");
+
+  mark_trx_read_write();
+
   if (unlikely(error= write_row(buf)))
     DBUG_RETURN(error);
   if (unlikely(error= binlog_log_row<Write_rows_log_event>(table, 0, buf)))
@@ -4060,6 +4546,8 @@ int handler::ha_update_row(const uchar *
    */
   DBUG_ASSERT(new_data == table->record[0]);
 
+  mark_trx_read_write();
+
   if (unlikely(error= update_row(old_data, new_data)))
     return error;
   if (unlikely(error= binlog_log_row<Update_rows_log_event>(table, old_data, new_data)))
@@ -4070,6 +4558,9 @@ int handler::ha_update_row(const uchar *
 int handler::ha_delete_row(const uchar *buf)
 {
   int error;
+
+  mark_trx_read_write();
+
   if (unlikely(error= delete_row(buf)))
     return error;
   if (unlikely(error= binlog_log_row<Delete_rows_log_event>(table, buf, 0)))
diff -Nrup a/sql/handler.h b/sql/handler.h
--- a/sql/handler.h	2007-12-20 21:16:51 +03:00
+++ b/sql/handler.h	2008-02-19 14:42:58 +03:00
@@ -721,14 +721,14 @@ struct handlerton
 #define HTON_SUPPORT_LOG_TABLES      (1 << 7) //Engine supports log tables
 #define HTON_NO_PARTITION            (1 << 8) //You can not partition these tables
 
-typedef struct st_thd_trans
+class Ha_trx_info;
+
+struct THD_TRANS
 {
-  /* number of entries in the ht[] */
-  uint        nht;
   /* true is not all entries in the ht[] support 2pc */
   bool        no_2pc;
-  /* storage engines that registered themselves for this transaction */
-  handlerton *ht[MAX_HA];
+  /* storage engines that registered in this transaction */
+  Ha_trx_info *ha_list;
   /* 
     The purpose of this flag is to keep track of non-transactional
     tables that were modified in scope of:
@@ -758,7 +758,106 @@ typedef struct st_thd_trans
     saved value.
   */
   bool modified_non_trans_table;
-} THD_TRANS;
+
+  void reset() { no_2pc= FALSE; modified_non_trans_table= FALSE; }
+};
+
+
+/**
+  Either statement transaction or normal transaction - related
+  thread-specific storage engine data.
+
+  If a storage engine participates in a statement/transaction,
+  an instance of this class is present in
+  thd->transaction.{stmt|all}.ha_list. The addition to
+  {stmt|all}.ha_list is made by trans_register_ha().
+
+  When it's time to commit or rollback, each element of ha_list
+  is used to access storage engine's prepare()/commit()/rollback()
+  methods, and also to evaluate if a full two phase commit is
+  necessary.
+
+  @sa General description of transaction handling in handler.cc.
+*/
+
+class Ha_trx_info
+{
+public:
+  /** Register this storage engine in the given transaction context. */
+  void register_ha(THD_TRANS *trans, handlerton *ht_arg)
+  {
+    DBUG_ASSERT(m_flags == 0);
+    DBUG_ASSERT(m_ht == NULL);
+    DBUG_ASSERT(m_next == NULL);
+
+    m_ht= ht_arg;
+    m_flags= (int) TRX_READ_ONLY; /* Assume read-only at start. */
+
+    m_next= trans->ha_list;
+    trans->ha_list= this;
+  }
+
+  /** Clear, prepare for reuse. */
+  void reset()
+  {
+    m_next= NULL;
+    m_ht= NULL;
+    m_flags= 0;
+  }
+
+  Ha_trx_info() { reset(); }
+
+  void set_trx_read_write()
+  {
+    DBUG_ASSERT(is_started());
+    m_flags|= (int) TRX_READ_WRITE;
+  }
+  bool is_trx_read_write() const
+  {
+    DBUG_ASSERT(is_started());
+    return m_flags & (int) TRX_READ_WRITE;
+  }
+  bool is_started() const { return m_ht != NULL; }
+  /** Mark this transaction read-write if the argument is read-write. */
+  void coalesce_trx_with(const Ha_trx_info *stmt_trx)
+  {
+    /*
+      Must be called only after the transaction has been started.
+      Can be called many times, e.g. when we have many
+      read-write statements in a transaction.
+    */
+    DBUG_ASSERT(is_started());
+    if (stmt_trx->is_trx_read_write())
+      set_trx_read_write();
+  }
+  Ha_trx_info *next() const
+  {
+    DBUG_ASSERT(is_started());
+    return m_next;
+  }
+  handlerton *ht() const
+  {
+    DBUG_ASSERT(is_started());
+    return m_ht;
+  }
+private:
+  enum { TRX_READ_ONLY= 0, TRX_READ_WRITE= 1 };
+  /** Auxiliary, used for ha_list management */
+  Ha_trx_info *m_next;
+  /**
+    Although a given Ha_trx_info instance is currently always used
+    for the same storage engine, 'ht' is not-NULL only when the
+    corresponding storage is a part of a transaction.
+  */
+  handlerton *m_ht;
+  /**
+    Transaction flags related to this engine.
+    Not-null only if this instance is a part of transaction.
+    May assume a combination of enum values above.
+  */
+  uchar       m_flags;
+};
+
 
 enum enum_tx_isolation { ISO_READ_UNCOMMITTED, ISO_READ_COMMITTED,
 			 ISO_REPEATABLE_READ, ISO_SERIALIZABLE};
@@ -1640,8 +1739,15 @@ protected:
     provide useful functionality.
   */
   virtual int rename_table(const char *from, const char *to);
+  /**
+    Delete a table in the engine. Called for base as well as temporary
+    tables.
+  */
   virtual int delete_table(const char *name);
 private:
+  /* Private helpers */
+  inline void mark_trx_read_write();
+private:
   /*
     Low-level primitives for storage engines.  These should be
     overridden by the storage engine class. To call these methods, use
@@ -1821,9 +1927,7 @@ extern TYPELIB tx_isolation_typelib;
 extern TYPELIB myisam_stats_method_typelib;
 extern ulong total_ha, total_ha_2pc;
 
-	/* Wrapper functions */
-#define ha_commit_stmt(thd) (ha_commit_trans((thd), FALSE))
-#define ha_rollback_stmt(thd) (ha_rollback_trans((thd), FALSE))
+       /* Wrapper functions */
 #define ha_commit(thd) (ha_commit_trans((thd), TRUE))
 #define ha_rollback(thd) (ha_rollback_trans((thd), TRUE))
 
diff -Nrup a/sql/log.cc b/sql/log.cc
--- a/sql/log.cc	2008-01-23 01:45:42 +03:00
+++ b/sql/log.cc	2008-02-19 14:42:58 +03:00
@@ -3332,6 +3332,16 @@ THD::binlog_start_trans_and_stmt()
     if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
       trans_register_ha(this, TRUE, binlog_hton);
     trans_register_ha(this, FALSE, binlog_hton);
+    /*
+      Mark statement transaction as read/write. We never start
+      a binary log transaction and keep it read-only,
+      therefore it's best to mark the transaction read/write just
+      at the same time we start it.
+      Not necessary to mark the normal transaction read/write
+      since the statement-level flag will be propagated automatically
+      inside ha_commit_trans.
+    */
+    ha_data[binlog_hton->slot].ha_info[0].set_trx_read_write();
   }
   DBUG_VOID_RETURN;
 }
diff -Nrup a/sql/log_event.cc b/sql/log_event.cc
--- a/sql/log_event.cc	2007-12-20 23:24:03 +03:00
+++ b/sql/log_event.cc	2008-02-19 14:42:58 +03:00
@@ -2898,7 +2898,7 @@ int Format_description_log_event::do_app
     original place when it comes to us; we'll know this by checking
     log_pos ("artificial" events have log_pos == 0).
   */
-  if (!artificial_event && created && thd->transaction.all.nht)
+  if (!artificial_event && created && thd->transaction.all.ha_list)
   {
     /* This is not an error (XA is safe), just an information */
     rli->report(INFORMATION_LEVEL, 0,
diff -Nrup a/sql/rpl_injector.cc b/sql/rpl_injector.cc
--- a/sql/rpl_injector.cc	2007-04-12 18:13:46 +04:00
+++ b/sql/rpl_injector.cc	2008-02-19 14:42:58 +03:00
@@ -62,6 +62,26 @@ int injector::transaction::commit()
 {
    DBUG_ENTER("injector::transaction::commit()");
    m_thd->binlog_flush_pending_rows_event(true);
+   /*
+     Cluster replication does not preserve statement or
+     transaction boundaries of the master.  Instead, a new
+     transaction on replication slave is started when a new GCI
+     (global checkpoint identifier) is issued, and is committed
+     when the last event of the check point has been received and
+     processed. This ensures consistency of each cluster in
+     cluster replication, and there is no requirement for stronger
+     consistency: MySQL replication is asynchronous with other
+     engines as well.
+
+     A practical consequence of that is that row level replication
+     stream passed through the injector thread never contains
+     COMMIT events.
+     Here we should preserve the server invariant that there is no
+     outstanding statement transaction when the normal transaction
+     is committed by committing the statement transaction
+     explicitly.
+   */
+   ha_autocommit_or_rollback(m_thd, 0);
    end_trans(m_thd, COMMIT);
    DBUG_RETURN(0);
 }
diff -Nrup a/sql/sp.cc b/sql/sp.cc
--- a/sql/sp.cc	2008-01-23 01:45:42 +03:00
+++ b/sql/sp.cc	2008-02-19 14:42:58 +03:00
@@ -665,8 +665,16 @@ sp_returns_type(THD *thd, String &result
               (TYPE_ENUM_PROCEDURE or TYPE_ENUM_FUNCTION).
   @param sp   Stored routine object to store.
 
-  @return Error code. SP_OK is returned on success. Other SP_ constants are
-  used to indicate about errors.
+  @note Opens and closes the thread tables. Therefore assumes
+  that there are no locked tables in this thread at the time of
+  invocation.
+  Unlike some other DDL statements, *does* close the tables
+  in the end, since the call to this function is normally
+  followed by an implicit grant (sp_grant_privileges())
+  and this subsequent call opens and closes mysql.procs_priv.
+
+  @return Error code. SP_OK is returned on success. Other
+  SP_ constants are used to indicate about errors.
 */
 
 int
@@ -1223,7 +1231,13 @@ done:
 }
 
 
-/* Drop all routines in database 'db' */
+/**
+  Drop all routines in database 'db'
+
+  @note Close the thread tables, the calling code might want to
+  delete from other system tables afterwards.
+*/
+
 int
 sp_drop_db_routines(THD *thd, char *db)
 {
diff -Nrup a/sql/sp_head.cc b/sql/sp_head.cc
--- a/sql/sp_head.cc	2008-02-01 13:50:17 +03:00
+++ b/sql/sp_head.cc	2008-02-19 14:42:58 +03:00
@@ -2700,6 +2700,7 @@ sp_lex_keeper::reset_lex_and_exec_core(T
   m_lex->unit.cleanup();
 
   thd_proc_info(thd, "closing tables");
+  /* Here we also commit or rollback the current statement. */
   close_thread_tables(thd);
   thd_proc_info(thd, 0);
 
diff -Nrup a/sql/sql_base.cc b/sql/sql_base.cc
--- a/sql/sql_base.cc	2008-02-01 13:50:17 +03:00
+++ b/sql/sql_base.cc	2008-02-19 14:42:58 +03:00
@@ -1324,29 +1324,45 @@ void close_thread_tables(THD *thd)
     Mark all temporary tables used by this statement as free for reuse.
   */
   mark_temp_tables_as_free_for_reuse(thd);
-
-  if (thd->locked_tables || prelocked_mode)
+  /*
+    Let us commit transaction for statement. Since in 5.0 we only have
+    one statement transaction and don't allow several nested statement
+    transactions this call will do nothing if we are inside of stored
+    function or trigger (i.e. statement transaction is already active and
+    does not belong to statement for which we do close_thread_tables()).
+    TODO: This should be fixed in later releases.
+   */
+  if (!(thd->state_flags & Open_tables_state::BACKUPS_AVAIL))
   {
+    thd->main_da.can_overwrite_status= TRUE;
+    ha_autocommit_or_rollback(thd, thd->is_error());
+    thd->main_da.can_overwrite_status= FALSE;
+
     /*
-      Let us commit transaction for statement. Since in 5.0 we only have
-      one statement transaction and don't allow several nested statement
-      transactions this call will do nothing if we are inside of stored
-      function or trigger (i.e. statement transaction is already active and
-      does not belong to statement for which we do close_thread_tables()).
-      TODO: This should be fixed in later releases.
+      Reset transaction state, but only if we're not inside a
+      sub-statement of a prelocked statement.
     */
-    ha_commit_stmt(thd);
+    if (! prelocked_mode || thd->lex->requires_prelocking())
+      thd->transaction.stmt.reset();
+  }
+
+  if (thd->locked_tables || prelocked_mode)
+  {
 
     /* Ensure we are calling ha_reset() for all used tables */
     mark_used_tables_as_free_for_reuse(thd, thd->open_tables);
 
-    /* We are under simple LOCK TABLES so should not do anything else. */
+    /*
+      We are under simple LOCK TABLES or we're inside a sub-statement
+      of a prelocked statement, so should not do anything else.
+    */
     if (!prelocked_mode || !thd->lex->requires_prelocking())
       DBUG_VOID_RETURN;
 
     /*
-      We are in prelocked mode, so we have to leave it now with doing
-      implicit UNLOCK TABLES if need.
+      We are in the top-level statement of a prelocked statement,
+      so we have to leave the prelocked mode now with doing implicit
+      UNLOCK TABLES if needed.
     */
     DBUG_PRINT("info",("thd->prelocked_mode= NON_PRELOCKED"));
     thd->prelocked_mode= NON_PRELOCKED;
@@ -1375,19 +1391,6 @@ void close_thread_tables(THD *thd)
     thd->lock=0;
   }
   /*
-    assume handlers auto-commit (if some doesn't - transaction handling
-    in MySQL should be redesigned to support it; it's a big change,
-    and it's not worth it - better to commit explicitly only writing
-    transactions, read-only ones should better take care of themselves.
-    saves some work in 2pc too)
-    see also sql_parse.cc - dispatch_command()
-  */
-  if (!(thd->state_flags & Open_tables_state::BACKUPS_AVAIL))
-    bzero(&thd->transaction.stmt, sizeof(thd->transaction.stmt));
-  if (!thd->active_transaction())
-    thd->transaction.xid_state.xid.null();
-
-  /*
     Note that we need to hold LOCK_open while changing the
     open_tables list. Another thread may work on it.
     (See: remove_table_from_cache(), mysql_wait_completed_table())
@@ -5059,10 +5062,7 @@ int decide_logging_format(THD *thd, TABL
     DBUG_PRINT("info", ("error: %d", error));
 
     if (error)
-    {
-      ha_rollback_stmt(thd);
       return -1;
-    }
 
     /*
       We switch to row-based format if we are in mixed mode and one of
@@ -5216,7 +5216,6 @@ int lock_tables(THD *thd, TABLE_LIST *ta
           table->table->query_id= thd->query_id;
           if (check_lock_and_start_stmt(thd, table->table, table->lock_type))
           {
-            ha_rollback_stmt(thd);
             mysql_unlock_tables(thd, thd->locked_tables);
             thd->locked_tables= 0;
             thd->options&= ~(OPTION_TABLE_LOCK);
@@ -5251,7 +5250,6 @@ int lock_tables(THD *thd, TABLE_LIST *ta
       if (!table->placeholder() &&
 	  check_lock_and_start_stmt(thd, table->table, table->lock_type))
       {
-	ha_rollback_stmt(thd);
 	DBUG_RETURN(-1);
       }
     }
diff -Nrup a/sql/sql_class.cc b/sql/sql_class.cc
--- a/sql/sql_class.cc	2008-01-23 01:45:42 +03:00
+++ b/sql/sql_class.cc	2008-02-19 14:42:58 +03:00
@@ -264,7 +264,7 @@ const char *set_thd_proc_info(THD *thd, 
 extern "C"
 void **thd_ha_data(const THD *thd, const struct handlerton *hton)
 {
-  return (void **) thd->ha_data + hton->slot;
+  return (void **) &thd->ha_data[hton->slot].ha_ptr;
 }
 
 extern "C"
@@ -2513,7 +2513,7 @@ bool select_dumpvar::send_data(List<Item
       suv->update();
     }
   }
-  DBUG_RETURN(0);
+  DBUG_RETURN(thd->is_error());
 }
 
 bool select_dumpvar::send_eof()
diff -Nrup a/sql/sql_class.h b/sql/sql_class.h
--- a/sql/sql_class.h	2008-01-25 20:14:59 +03:00
+++ b/sql/sql_class.h	2008-02-19 14:42:58 +03:00
@@ -686,7 +686,8 @@ private:
 struct st_savepoint {
   struct st_savepoint *prev;
   char                *name;
-  uint                 length, nht;
+  uint                 length;
+  Ha_trx_info         *ha_list;
 };
 
 enum xa_states {XA_NOTR=0, XA_ACTIVE, XA_IDLE, XA_PREPARED};
@@ -1093,6 +1094,33 @@ private:
 
 
 /**
+  Storage engine specific thread local data.
+*/
+
+struct Ha_data
+{
+  /**
+    Storage engine specific thread local data.
+    Lifetime: one user connection.
+  */
+  void *ha_ptr;
+  /**
+    0: Life time: one statement within a transaction. If @@autocommit is
+    on, also represents the entire transaction.
+    @sa trans_register_ha()
+
+    1: Life time: one transaction within a connection.
+    If the storage engine does not participate in a transaction,
+    this should not be used.
+    @sa trans_register_ha()
+  */
+  Ha_trx_info ha_info[2];
+
+  Ha_data() :ha_ptr(NULL) {}
+};
+
+
+/**
   @class THD
   For each client connection we create a separate thread with THD serving as
   a thread/connection descriptor
@@ -1231,7 +1259,7 @@ public:
   uint in_sub_stmt;
 
   /* container for handler's private per-connection data */
-  void *ha_data[MAX_HA];
+  Ha_data ha_data[MAX_HA];
 
 #ifndef MYSQL_CLIENT
   int binlog_setup_trx_data();
diff -Nrup a/sql/sql_cursor.cc b/sql/sql_cursor.cc
--- a/sql/sql_cursor.cc	2007-10-16 23:37:25 +04:00
+++ b/sql/sql_cursor.cc	2008-02-19 14:42:58 +03:00
@@ -322,9 +322,10 @@ Sensitive_cursor::post_open(THD *thd)
 
   close_at_commit= FALSE; /* reset in case we're reusing the cursor */
   info= &ht_info[0];
-  for (handlerton **pht= thd->transaction.stmt.ht; *pht; pht++)
+  for (Ha_trx_info *ha_trx_info= thd->transaction.stmt.ha_list;
+       ha_trx_info; ha_trx_info= ha_trx_info->next())
   {
-    handlerton *ht= *pht;
+    handlerton *ht= ha_trx_info->ht();
     close_at_commit|= test(ht->flags & HTON_CLOSE_CURSORS_AT_COMMIT);
     if (ht->create_cursor_read_view)
     {
diff -Nrup a/sql/sql_delete.cc b/sql/sql_delete.cc
--- a/sql/sql_delete.cc	2008-01-25 20:14:59 +03:00
+++ b/sql/sql_delete.cc	2008-02-19 14:42:58 +03:00
@@ -24,6 +24,14 @@
 #include "sp_head.h"
 #include "sql_trigger.h"
 
+/**
+  Implement DELETE SQL word.
+
+  @note Like implementations of other DDL/DML in MySQL, this function
+  relies on the caller to close the thread tables. This is done in the
+  end of dispatch_command().
+*/
+
 bool mysql_delete(THD *thd, TABLE_LIST *table_list, COND *conds,
                   SQL_LIST *order, ha_rows limit, ulonglong options,
                   bool reset_auto_increment)
@@ -380,17 +388,6 @@ cleanup:
   }
   DBUG_ASSERT(transactional_table || !deleted || thd->transaction.stmt.modified_non_trans_table);
   free_underlaid_joins(thd, select_lex);
-  if (transactional_table)
-  {
-    if (ha_autocommit_or_rollback(thd,error >= 0))
-      error=1;
-  }
-
-  if (thd->lock)
-  {
-    mysql_unlock_tables(thd, thd->lock);
-    thd->lock=0;
-  }
   if (error < 0 || (thd->lex->ignore && !thd->is_fatal_error))
   {
     thd->row_count_func= deleted;
@@ -751,11 +748,9 @@ void multi_delete::abort()
     The same if all tables are transactional, regardless of where we are.
     In all other cases do attempt deletes ...
   */
-  if ((table_being_deleted == delete_tables &&
-       table_being_deleted->table->file->has_transactions()) ||
-      !normal_tables)
-    ha_rollback_stmt(thd);
-  else if (do_delete)
+  if (do_delete && normal_tables &&
+      (table_being_deleted != delete_tables ||
+       !table_being_deleted->table->file->has_transactions()))
   {
     /*
       We have to execute the recorded do_deletes() and write info into the
@@ -921,11 +916,6 @@ bool multi_delete::send_eof()
   if (local_error != 0)
     error_handled= TRUE; // to force early leave from ::send_error()
 
-  /* Commit or rollback the current SQL statement */
-  if (transactional_tables)
-    if (ha_autocommit_or_rollback(thd,local_error > 0))
-      local_error=1;
-
   if (!local_error)
   {
     thd->row_count_func= deleted;
@@ -1055,6 +1045,12 @@ trunc_by_del:
   error= mysql_delete(thd, table_list, (COND*) 0, (SQL_LIST*) 0,
                       HA_POS_ERROR, LL(0), TRUE);
   ha_enable_transaction(thd, TRUE);
+  /*
+    Safety, in case the engine ignored ha_enable_transaction(FALSE)
+    above. Also clears thd->transaction.*.
+  */
+  error= ha_autocommit_or_rollback(thd, error);
+  ha_commit(thd);
   thd->options= save_options;
   thd->current_stmt_binlog_row_based= save_binlog_row_based;
   DBUG_RETURN(error);
diff -Nrup a/sql/sql_do.cc b/sql/sql_do.cc
--- a/sql/sql_do.cc	2006-12-31 03:06:36 +03:00
+++ b/sql/sql_do.cc	2008-02-19 14:42:58 +03:00
@@ -28,7 +28,17 @@ bool mysql_do(THD *thd, List<Item> &valu
   while ((value = li++))
     value->val_int();
   free_underlaid_joins(thd, &thd->lex->select_lex);
-  thd->clear_error(); // DO always is OK
+
+  if (thd->is_error())
+  {
+    /*
+      Rollback the effect of the statement, since next instruction
+      will clear the error and the rollback in the end of
+      dispatch_command() won't work.
+    */
+    ha_autocommit_or_rollback(thd, thd->is_error());
+    thd->clear_error(); // DO always is OK
+  }
   send_ok(thd);
   DBUG_RETURN(FALSE);
 }
diff -Nrup a/sql/sql_insert.cc b/sql/sql_insert.cc
--- a/sql/sql_insert.cc	2008-01-25 20:14:59 +03:00
+++ b/sql/sql_insert.cc	2008-02-19 14:42:58 +03:00
@@ -541,6 +541,10 @@ bool open_and_lock_for_insert_delayed(TH
 
 /**
   INSERT statement implementation
+
+  @note Like implementations of other DDL/DML in MySQL, this function
+  relies on the caller to close the thread tables. This is done in the
+  end of dispatch_command().
 */
 
 bool mysql_insert(THD *thd,TABLE_LIST *table_list,
@@ -893,12 +897,9 @@ bool mysql_insert(THD *thd,TABLE_LIST *t
     }
     DBUG_ASSERT(transactional_table || !changed || 
                 thd->transaction.stmt.modified_non_trans_table);
-    if (transactional_table)
-      error=ha_autocommit_or_rollback(thd,error);
-    
+
     if (thd->lock)
     {
-      mysql_unlock_tables(thd, thd->lock);
       /*
         Invalidate the table in the query cache if something changed
         after unlocking when changes become fisible.
@@ -909,7 +910,6 @@ bool mysql_insert(THD *thd,TABLE_LIST *t
       {
         query_cache_invalidate3(thd, table_list, 1);
       }
-      thd->lock=0;
     }
   }
   thd_proc_info(thd, "end");
@@ -2445,7 +2445,7 @@ err:
     first call to ha_*_row() instead. Remove code that are used to
     cover for the case outlined above.
    */
-  ha_rollback_stmt(thd);
+  ha_autocommit_or_rollback(thd, 1);
 
 #ifndef __WIN__
 end:
@@ -3139,18 +3139,6 @@ bool select_insert::send_eof()
                       thd->query, thd->query_length,
                       trans_table, FALSE, killed_status);
   }
-  /*
-    We will call ha_autocommit_or_rollback() also for
-    non-transactional tables under row-based replication: there might
-    be events in the binary logs transaction, and we need to write
-    them to the binary log.
-   */
-  if (trans_table || thd->current_stmt_binlog_row_based)
-  {
-    int error2= ha_autocommit_or_rollback(thd, error);
-    if (error2 && !error)
-      error= error2;
-  }
   table->file->ha_release_auto_increment();
 
   if (error)
@@ -3228,7 +3216,6 @@ void select_insert::abort() {
     table->file->ha_release_auto_increment();
   }
 
-  ha_rollback_stmt(thd);
   DBUG_VOID_RETURN;
 }
 
@@ -3667,7 +3654,10 @@ bool select_create::send_eof()
       nevertheless.
     */
     if (!table->s->tmp_table)
-      ha_commit(thd);               // Can fail, but we proceed anyway
+    {
+      ha_autocommit_or_rollback(thd, 0);
+      end_active_trans(thd);
+    }
 
     table->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY);
     table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE);
@@ -3691,12 +3681,9 @@ void select_create::abort()
    by removing the table, even for non-transactional tables.
   */
   tmp_disable_binlog(thd);
-  select_insert::abort();
-  reenable_binlog(thd);
-
   /*
-    We roll back the statement, including truncating the transaction
-    cache of the binary log, if the statement failed.
+    In select_insert::abort() we roll back the statement, including
+    truncating the transaction cache of the binary log.
 
     We roll back the statement prior to deleting the table and prior
     to releasing the lock on the table, since there might be potential
@@ -3707,8 +3694,9 @@ void select_create::abort()
     of the table succeeded or not, since we need to reset the binary
     log state.
   */
-  if (thd->current_stmt_binlog_row_based)
-    ha_rollback_stmt(thd);
+  select_insert::abort();
+  reenable_binlog(thd);
+
 
   if (m_plock)
   {
diff -Nrup a/sql/sql_load.cc b/sql/sql_load.cc
--- a/sql/sql_load.cc	2007-10-30 14:49:36 +03:00
+++ b/sql/sql_load.cc	2008-02-19 14:42:58 +03:00
@@ -470,9 +470,6 @@ bool mysql_load(THD *thd,sql_exchange *e
       }
     }
 #endif /*!EMBEDDED_LIBRARY*/
-    if (transactional_table)
-      ha_autocommit_or_rollback(thd,error);
-
     error= -1;				// Error on read
     goto err;
   }
@@ -510,8 +507,6 @@ bool mysql_load(THD *thd,sql_exchange *e
     }
   }
 #endif /*!EMBEDDED_LIBRARY*/
-  if (transactional_table)
-    error=ha_autocommit_or_rollback(thd,error);
 
   /* ok to client sent only after binlog write and engine commit */
   send_ok(thd, info.copied + info.deleted, 0L, name);
@@ -519,11 +514,6 @@ err:
   DBUG_ASSERT(transactional_table || !(info.copied || info.deleted) ||
               thd->transaction.stmt.modified_non_trans_table);
   table->file->ha_release_auto_increment();
-  if (thd->lock)
-  {
-    mysql_unlock_tables(thd, thd->lock);
-    thd->lock=0;
-  }
   table->auto_increment_field_not_null= FALSE;
   thd->abort_on_warning= 0;
   DBUG_RETURN(error);
diff -Nrup a/sql/sql_parse.cc b/sql/sql_parse.cc
--- a/sql/sql_parse.cc	2008-02-01 13:55:37 +03:00
+++ b/sql/sql_parse.cc	2008-02-19 14:42:59 +03:00
@@ -1465,21 +1465,13 @@ bool dispatch_command(enum enum_server_c
     break;
   }
 
-  thd_proc_info(thd, "closing tables");
-  /* Free tables */
-  close_thread_tables(thd);
+  /* If commit fails, we should be able to reset the OK status. */
+  thd->main_da.can_overwrite_status= TRUE;
+  ha_autocommit_or_rollback(thd, thd->is_error());
+  thd->main_da.can_overwrite_status= FALSE;
+
+  thd->transaction.stmt.reset();
 
-  /*
-    assume handlers auto-commit (if some doesn't - transaction handling
-    in MySQL should be redesigned to support it; it's a big change,
-    and it's not worth it - better to commit explicitly only writing
-    transactions, read-only ones should better take care of themselves.
-    saves some work in 2pc too)
-    see also sql_base.cc - close_thread_tables()
-  */
-  bzero(&thd->transaction.stmt, sizeof(thd->transaction.stmt));
-  if (!thd->active_transaction())
-    thd->transaction.xid_state.xid.null();
 
   /* report error issued during command execution */
   if (thd->killed_errno())
@@ -1496,6 +1488,10 @@ bool dispatch_command(enum enum_server_c
   net_end_statement(thd);
   query_cache_end_of_result(thd);
 
+  thd->proc_info= "closing tables";
+  /* Free tables */
+  close_thread_tables(thd);
+
   log_slow_statement(thd);
 
   thd_proc_info(thd, "cleaning up");
@@ -3011,10 +3007,8 @@ end_with_restore_list:
           /* INSERT ... SELECT should invalidate only the very first table */
           TABLE_LIST *save_table= first_table->next_local;
           first_table->next_local= 0;
-          mysql_unlock_tables(thd, thd->lock);
           query_cache_invalidate3(thd, first_table, 1);
           first_table->next_local= save_table;
-          thd->lock=0;
         }
         delete sel_result;
       }
@@ -3985,7 +3979,6 @@ end_with_restore_list:
           push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
                        ER_PROC_AUTO_GRANT_FAIL,
                        ER(ER_PROC_AUTO_GRANT_FAIL));
-        close_thread_tables(thd);
       }
 #endif
     break;
diff -Nrup a/sql/sql_partition.cc b/sql/sql_partition.cc
--- a/sql/sql_partition.cc	2007-12-20 21:16:52 +03:00
+++ b/sql/sql_partition.cc	2008-02-19 14:42:59 +03:00
@@ -3968,31 +3968,35 @@ static int fast_end_partition(THD *thd, 
                               bool written_bin_log)
 {
   int error;
+  char tmp_name[80];
   DBUG_ENTER("fast_end_partition");
 
   thd->proc_info="end";
+
   if (!is_empty)
     query_cache_invalidate3(thd, table_list, 0);
-  error= ha_commit_stmt(thd);
-  if (ha_commit(thd))
+
+  error= ha_autocommit_or_rollback(thd, 0);
+  if (end_active_trans(thd))
     error= 1;
-  if (!error || is_empty)
+
+  if (error)
   {
-    char tmp_name[80];
-    if ((!is_empty) && (!written_bin_log) &&
-        (!thd->lex->no_write_to_binlog))
-      write_bin_log(thd, FALSE, thd->query, thd->query_length);
-    close_thread_tables(thd);
-    my_snprintf(tmp_name, sizeof(tmp_name), ER(ER_INSERT_INFO),
-                (ulong) (copied + deleted),
-                (ulong) deleted,
-                (ulong) 0);
-    send_ok(thd, (ha_rows) (copied+deleted),0L,tmp_name);
-    DBUG_RETURN(FALSE);
+    /* If error during commit, no need to rollback, it's done. */
+    table->file->print_error(error, MYF(0));
+    DBUG_RETURN(TRUE);
   }
-  table->file->print_error(error, MYF(0));
-  close_thread_tables(thd);
-  DBUG_RETURN(TRUE);
+
+  if ((!is_empty) && (!written_bin_log) &&
+      (!thd->lex->no_write_to_binlog))
+    write_bin_log(thd, FALSE, thd->query, thd->query_length);
+
+  my_snprintf(tmp_name, sizeof(tmp_name), ER(ER_INSERT_INFO),
+              (ulong) (copied + deleted),
+              (ulong) deleted,
+              (ulong) 0);
+  send_ok(thd, (ha_rows) (copied+deleted),0L, tmp_name);
+  DBUG_RETURN(FALSE);
 }
 
 
diff -Nrup a/sql/sql_table.cc b/sql/sql_table.cc
--- a/sql/sql_table.cc	2008-02-01 13:50:18 +03:00
+++ b/sql/sql_table.cc	2008-02-19 14:42:59 +03:00
@@ -4131,6 +4131,7 @@ static bool mysql_admin_table(THD* thd, 
       switch ((*prepare_func)(thd, table, check_opt)) {
       case  1:           // error, message written to net
         ha_autocommit_or_rollback(thd, 1);
+        end_trans(thd, ROLLBACK);
         close_thread_tables(thd);
         DBUG_PRINT("admin", ("simple error, admin next table"));
         continue;
@@ -4189,6 +4190,7 @@ static bool mysql_admin_table(THD* thd, 
                           table_name);
       protocol->store(buff, length, system_charset_info);
       ha_autocommit_or_rollback(thd, 0);
+      end_trans(thd, COMMIT);
       close_thread_tables(thd);
       lex->reset_query_tables_list(FALSE);
       table->table=0;				// For query cache
@@ -4461,6 +4463,7 @@ send_result_message:
       }
     }
     ha_autocommit_or_rollback(thd, 0);
+    end_trans(thd, COMMIT);
     close_thread_tables(thd);
     table->table=0;				// For query cache
     if (protocol->write())
@@ -4470,8 +4473,9 @@ send_result_message:
   send_eof(thd);
   DBUG_RETURN(FALSE);
 
- err:
+err:
   ha_autocommit_or_rollback(thd, 1);
+  end_trans(thd, ROLLBACK);
   close_thread_tables(thd);			// Shouldn't be needed
   if (table)
     table->table=0;
@@ -4994,8 +4998,8 @@ mysql_discard_or_import_tablespace(THD *
   query_cache_invalidate3(thd, table_list, 0);
 
   /* The ALTER TABLE is always in its own transaction */
-  error = ha_commit_stmt(thd);
-  if (ha_commit(thd))
+  error = ha_autocommit_or_rollback(thd, 0);
+  if (end_active_trans(thd))
     error=1;
   if (error)
     goto err;
@@ -5003,7 +5007,6 @@ mysql_discard_or_import_tablespace(THD *
 
 err:
   ha_autocommit_or_rollback(thd, error);
-  close_thread_tables(thd);
   thd->tablespace_op=FALSE;
   
   if (error == 0)
@@ -6526,8 +6529,8 @@ view_err:
     VOID(pthread_mutex_unlock(&LOCK_open));
     alter_table_manage_keys(table, table->file->indexes_are_disabled(),
                             alter_info->keys_onoff);
-    error= ha_commit_stmt(thd);
-    if (ha_commit(thd))
+    error= ha_autocommit_or_rollback(thd, 0);
+    if (end_active_trans(thd))
       error= 1;
   }
   thd->count_cuted_fields= CHECK_FIELD_IGNORE;
@@ -6615,7 +6618,7 @@ view_err:
 
     /* Need to commit before a table is unlocked (NDB requirement). */
     DBUG_PRINT("info", ("Committing before unlocking table"));
-    if (ha_commit_stmt(thd) || ha_commit(thd))
+    if (ha_autocommit_or_rollback(thd, 0) || end_active_trans(thd))
       goto err1;
     committed= 1;
   }
@@ -7116,9 +7119,9 @@ copy_data_between_tables(TABLE *from,TAB
     Ensure that the new table is saved properly to disk so that we
     can do a rename
   */
-  if (ha_commit_stmt(thd))
+  if (ha_autocommit_or_rollback(thd, 0))
     error=1;
-  if (ha_commit(thd))
+  if (end_active_trans(thd))
     error=1;
 
  err:
diff -Nrup a/sql/sql_udf.cc b/sql/sql_udf.cc
--- a/sql/sql_udf.cc	2007-11-14 16:28:20 +03:00
+++ b/sql/sql_udf.cc	2008-02-19 14:42:59 +03:00
@@ -382,6 +382,14 @@ static udf_func *add_udf(LEX_STRING *nam
 }
 
 
+/**
+  Create a user defined function. 
+
+  @note Like implementations of other DDL/DML in MySQL, this function
+  relies on the caller to close the thread tables. This is done in the
+  end of dispatch_command().
+*/
+
 int mysql_create_function(THD *thd,udf_func *udf)
 {
   int error;
@@ -489,7 +497,6 @@ int mysql_create_function(THD *thd,udf_f
     table->field[3]->store((longlong) u_d->type, TRUE);
   error = table->file->ha_write_row(table->record[0]);
 
-  close_thread_tables(thd);
   if (error)
   {
     my_error(ER_ERROR_ON_WRITE, MYF(0), "mysql.func", error);
diff -Nrup a/sql/sql_update.cc b/sql/sql_update.cc
--- a/sql/sql_update.cc	2008-01-25 20:15:00 +03:00
+++ b/sql/sql_update.cc	2008-02-19 14:42:59 +03:00
@@ -803,17 +803,6 @@ int mysql_update(THD *thd,
   }
   DBUG_ASSERT(transactional_table || !updated || thd->transaction.stmt.modified_non_trans_table);
   free_underlaid_joins(thd, select_lex);
-  if (transactional_table)
-  {
-    if (ha_autocommit_or_rollback(thd, error >= 0))
-      error=1;
-  }
-
-  if (thd->lock)
-  {
-    mysql_unlock_tables(thd, thd->lock);
-    thd->lock=0;
-  }
 
   /* If LAST_INSERT_ID(X) was used, report X */
   id= thd->arg_of_last_insert_id_function ?
@@ -1716,13 +1705,8 @@ void multi_update::abort()
     If not attempt to do remaining updates.
   */
 
-  if (trans_safe)
+  if (! trans_safe)
   {
-    DBUG_ASSERT(transactional_tables);
-    (void) ha_autocommit_or_rollback(thd, 1);
-  }
-  else
-  { 
     DBUG_ASSERT(thd->transaction.stmt.modified_non_trans_table);
     if (do_update && table_count > 1)
     {
@@ -1754,11 +1738,6 @@ void multi_update::abort()
     thd->transaction.all.modified_non_trans_table= TRUE;
   }
   DBUG_ASSERT(trans_safe || !updated || thd->transaction.stmt.modified_non_trans_table);
-  
-  if (transactional_tables)
-  {
-    (void) ha_autocommit_or_rollback(thd, 1);
-  }
 }
 
 
@@ -1995,12 +1974,6 @@ bool multi_update::send_eof()
   }
   if (local_error != 0)
     error_handled= TRUE; // to force early leave from ::send_error()
-
-  if (transactional_tables)
-  {
-    if (ha_autocommit_or_rollback(thd, local_error != 0))
-      local_error=1;
-  }
 
   if (local_error > 0) // if the above log write did not fail ...
   {
Thread
bk commit into 5.1 tree (kostja:1.2518) BUG#12713konstantin19 Feb