Fix for BUG#9213: GROUP BY returns wrong query results:

Make test_if_skip_sort_order() rebuild tab->ref if it decides to use an index
different from the index join optimizer has choosen.  

mysql-test/r/group_by.result

@@ -677,3 +677,28 @@ select sum(a)*sum(b) as d from t1 where a=1 group by c having d > 0;
 d
 10
 drop table t1;
+create table t1(a int);
+insert into t1 values (0),(1),(2),(3),(4),(5),(6),(8),(9);
+create table t2 (
+a int,
+b varchar(200) NOT NULL,
+c varchar(50) NOT NULL,
+d varchar(100) NOT NULL,
+primary key (a,b(132),c,d),
+key a (a,b)
+) charset=utf8;
+insert into t2 select 
+x3.a,  -- 3
+concat('val-', x3.a + 3*x4.a), -- 12
+concat('val-', @a:=x3.a + 3*x4.a + 12*C.a), -- 120
+concat('val-', @a + 120*D.a)
+from t1 x3, t1 x4, t1 C, t1 D where x3.a < 3 and x4.a < 4 and D.a < 4;
+delete from t2  where a = 2 and b = 'val-2' limit 30;
+explain select c from t2 where a = 2 and b = 'val-2' group by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t2	ref	PRIMARY,a	PRIMARY	400	const,const	6	Using where
+select c from t2 where a = 2 and b = 'val-2' group by c;
+c
+val-74
+val-98
+drop table t1,t2;

mysql-test/t/group_by.test

@@ -489,3 +489,28 @@ select a,sum(b) from t1 where a=1 group by c having a=1;
 select a as d,sum(b) from t1 where a=1 group by c having d=1;
 select sum(a)*sum(b) as d from t1 where a=1 group by c having d > 0;
 drop table t1;
+
+# Test for BUG#9213 GROUP BY query on utf-8 key returns wrong results
+create table t1(a int);
+insert into t1 values (0),(1),(2),(3),(4),(5),(6),(8),(9);
+create table t2 (
+  a int,
+  b varchar(200) NOT NULL,
+  c varchar(50) NOT NULL,
+  d varchar(100) NOT NULL,
+  primary key (a,b(132),c,d),
+  key a (a,b)
+) charset=utf8;
+
+insert into t2 select 
+   x3.a,  -- 3
+   concat('val-', x3.a + 3*x4.a), -- 12
+   concat('val-', @a:=x3.a + 3*x4.a + 12*C.a), -- 120
+   concat('val-', @a + 120*D.a)
+from t1 x3, t1 x4, t1 C, t1 D where x3.a < 3 and x4.a < 4 and D.a < 4;
+delete from t2  where a = 2 and b = 'val-2' limit 30;
+
+explain select c from t2 where a = 2 and b = 'val-2' group by c;
+select c from t2 where a = 2 and b = 'val-2' group by c;
+drop table t1,t2;
+

sql/sql_select.cc

@@ -7185,7 +7185,19 @@ test_if_skip_sort_order(JOIN_TAB *tab,ORDER *order,ha_rows select_limit,
 	/* Found key that can be used to retrieve data in sorted order */
 	if (tab->ref.key >= 0)
 	{
-	  tab->ref.key= new_ref_key;
+          /*
+            We'll use ref access method on key new_ref_key. In general case 
+            the index search tuple for new_ref_key will be different (e.g.
+            when one of the indexes only covers prefix of the field, see
+            BUG#9213 in group_by.test).
+            So we build tab->ref from scratch here.
+          */
+          KEYUSE *keyuse= tab->keyuse;
+          while (keyuse->key != new_ref_key && keyuse->table == tab->table)
+            keyuse++;
+          if (create_ref_for_key(tab->join, tab, keyuse, 
+                                 tab->join->const_table_map))
+            DBUG_RETURN(0);
 	}
 	else
 	{

sql/table.h

@@ -90,7 +90,9 @@ struct st_table {
   uint null_fields;			/* number of null fields */
   uint blob_fields;			/* number of blob fields */
   key_map keys_in_use, keys_for_keyread, read_only_keys;
-  key_map quick_keys, used_keys, keys_in_use_for_query;
+  key_map quick_keys;
+  key_map used_keys;  /* keys that cover all used table fields */
+  key_map keys_in_use_for_query;
   KEY  *key_info;			/* data of keys in database */
   TYPELIB keynames;			/* Pointers to keynames */
   ha_rows max_rows;			/* create information */