bzr push into mysql-trunk branch (norvald.ryeng:3804 to 3805) Bug#13626019

[Norvald H. Ryeng]

 3805 Norvald H. Ryeng	2012-03-20
      Bug#13626019 DIFFERENCE IN OUTPUT BETWEEN 5.1 & 5.5 FOR A LAST_DAY IS
      NULL SELECT QUERY
      
      Problem: Some LAST_DAY expressions that should evaluate to NULL do not
      satisfy IS NULL.
      
      There are two bugs in the code that can lead to this problem,
      depending on circumstances:
      
      1. In make_join_select(), constant conditions are replaced with
      Item::COND_FALSE or Item::COND_TRUE. When the IS NULL expression is
      evaluated, Item_func_isnull::val_int() checks if the value of its
      argument is NULL by calling is_null(). The default implementation,
      Item::is_null(), always returns false, and all items that may return
      NULL must overload this function. For Item_func_last_day, this is done
      in Item_func::is_null(). However, the argument to Item_func_isnull may
      in some cases be an Item_cache_datetime, for which there is no
      overloading is_null(), which means that Item::is_null() is called and
      always returns false.
      
      2. In 5.1 and 5.5 there is a related problem where maybe_null is not
      set for Item_func_last_day, which makes Item_func_isnull::val_int()
      reason that there is no need to call is_null() on its argument since
      it already knows it can't be NULL. This is the cause for erroneous
      results in 5.1. The bug has already been fixed in trunk by
      revid:alexander.barkov@stripped.
      
      Bug 2 has been fixed in trunk, and bug 1 is masked: The
      Item_cache_datetime object is constructed as part of
      JOIN::cache_const_exprs(), which is called from JOIN::optimize(). In
      5.1 and 5.5 this call happens before the call to
      make_join_select(). In trunk, the order is reversed and the bug is
      masked. The erroneous result can't occur in trunk, but the behavior of
      is_null() is still wrong and should be fixed.
      
      Fix: Item_cache::is_null() should return the cached null_value or, if
      no value has been cached, example->is_null().
     @ mysql-test/r/type_date.result
        Add test case for bug #13626019.
     @ mysql-test/t/type_date.test
        Add test case for bug #13626019.
     @ sql/item.h
        Add Item_cache::is_null().

    modified:
      mysql-test/r/type_date.result
      mysql-test/t/type_date.test
      sql/item.h
 3804 Tor Didriksen	2012-03-20
      BUG#11748924 PARTITIONS: TOO-LONG COMMENT CAUSES NO WARNING
      
      Post-push fix: remove memory leak, and fix indentation.

    modified:
      sql/sql_table.cc
      sql/unireg.cc
=== modified file 'mysql-test/r/type_date.result'
--- a/mysql-test/r/type_date.result	2012-03-19 17:59:14 +0000
+++ b/mysql-test/r/type_date.result	2012-03-20 12:26:25 +0000
@@ -481,6 +481,28 @@ DATE(MIN(a))
 DROP TABLE t1;
 SET timestamp=DEFAULT;
 #
+# Bug #13626019 DIFFERENCE IN OUTPUT BETWEEN 5.1 & 5.5 FOR A LAST_DAY IS
+# NULL SELECT QUERY
+#
+CREATE TABLE t1(a INT) ENGINE=InnoDB;
+CREATE TABLE t2(b CHAR(19));
+INSERT INTO t1 VALUES(1);
+INSERT INTO t2 VALUES('0000-00-00 00:00:00');
+SELECT * FROM t1 WHERE LAST_DAY('0000-00-00 00:00:00') IS NULL;
+a
+1
+Warnings:
+Warning	1292	Incorrect datetime value: '0000-00-00'
+Warning	1292	Incorrect datetime value: '0000-00-00'
+Warning	1292	Incorrect datetime value: '0000-00-00'
+SELECT * FROM t1 WHERE
+LAST_DAY((SELECT b FROM t2 WHERE b = '0000-00-00 00:00:00')) IS NULL;
+a
+1
+Warnings:
+Warning	1292	Incorrect datetime value: '0000-00-00'
+DROP TABLE t1,t2;
+#
 # Test for bug#11747847 - 34280: create table fails if NO_ZERO_DATE
 # or NO_ZERO_IN_DATE SQL mode is set.
 DROP TABLE IF EXISTS t1, t2, t3;

=== modified file 'mysql-test/t/type_date.test'
--- a/mysql-test/t/type_date.test	2012-03-19 17:59:14 +0000
+++ b/mysql-test/t/type_date.test	2012-03-20 12:26:25 +0000
@@ -399,6 +399,20 @@ DROP TABLE t1;
 SET timestamp=DEFAULT;
 
 --echo #
+--echo # Bug #13626019 DIFFERENCE IN OUTPUT BETWEEN 5.1 & 5.5 FOR A LAST_DAY IS
+--echo # NULL SELECT QUERY
+--echo #
+
+CREATE TABLE t1(a INT) ENGINE=InnoDB;
+CREATE TABLE t2(b CHAR(19));
+INSERT INTO t1 VALUES(1);
+INSERT INTO t2 VALUES('0000-00-00 00:00:00');
+SELECT * FROM t1 WHERE LAST_DAY('0000-00-00 00:00:00') IS NULL;
+SELECT * FROM t1 WHERE
+LAST_DAY((SELECT b FROM t2 WHERE b = '0000-00-00 00:00:00')) IS NULL;
+DROP TABLE t1,t2;
+
+--echo #
 --echo # Test for bug#11747847 - 34280: create table fails if NO_ZERO_DATE
 --echo # or NO_ZERO_IN_DATE SQL mode is set.
 

=== modified file 'sql/item.h'
--- a/sql/item.h	2012-03-15 14:41:06 +0000
+++ b/sql/item.h	2012-03-20 12:26:25 +0000
@@ -3902,6 +3902,7 @@ public:
   { return test(example && example->basic_const_item());}
   bool walk (Item_processor processor, bool walk_subquery, uchar *argument);
   virtual void clear() { null_value= TRUE; value_cached= FALSE; }
+  bool is_null() { return value_cached ? null_value : example->is_null(); }
   Item_result result_type() const
   {
     if (!example)

No bundle (reason: useless for push emails).