Bug#43580: Issue with Innodb on multi-table update

                        
Certain multi-updates gave different results on InnoDB from
to MyISAM, due to on-the-fly updates being used on the former and
the update order matters.
Fixed by turning off on-the-fly updates when update order 
dependencies are present.


mysql-test/r/innodb_mysql.result:
  Bug#43580: Test result.
mysql-test/suite/rpl/r/rpl_slave_skip.result:
  Bug#43580: Changed test result. The InnoDB result is now what it would have been on MyISAM.
mysql-test/t/innodb_mysql.test:
  Bug#43580: Test case.
sql/sql_base.cc:
  Bug#43580: Added a word of caution about using tmp_set here.
sql/sql_update.cc:
  Bug#43580: Fix.
  Calls to TABLE::mark_columns_needed_for_update() are moved
  from mysql_multi_update_prepare() and right before the decison
  to do on-the-fly updates to the place where we do (or don't do)
  on-the-fly updates.

1 files changed, 7 insertions, 0 deletions

diff --git a/sql/sql_base.cc b/sql/sql_base.cc
index 0dc29f7e3c2..d4d813d0fbd 100644
--- a/sql/sql_base.cc
+++ b/sql/sql_base.cc
@@ -5585,6 +5585,13 @@ static void update_field_dependencies(THD *thd, Field *field, TABLE *table)
       other_bitmap=   table->read_set;
     }
 
+    /* 
+       The test-and-set mechanism in the bitmap is not reliable during
+       multi-UPDATE statements under MARK_COLUMNS_READ mode
+       (thd->mark_used_columns == MARK_COLUMNS_READ), as this bitmap contains
+       only those columns that are used in the SET clause. I.e they are being
+       set here. See multi_update::prepare()
+    */
     if (bitmap_fast_test_and_set(current_bitmap, field->field_index))
     {
       if (thd->mark_used_columns == MARK_COLUMNS_WRITE)