WL#926 "AVG(DISTINCT) and other distincts", part 2 (out of 3): clean up

Item_sum_count_distinct, and
deploy Unique for use with COUNT(DISTINCT) if there is no blob
column in the list of DISTINCT arguments.


mysql-test/r/count_distinct2.result:
  Test results fixed.
mysql-test/r/func_group.result:
  Updated.
mysql-test/r/sum_distinct.result:
  Updated.
mysql-test/t/func_group.test:
  Add a test for COUNT(DISTINCT) and true varchar and case-insensitive
  collation. The table in the test contains only two distinct values.
mysql-test/t/sum_distinct.test:
  Since now we support INSERT INTO t1 (a) SELECT a+1 FROM t1, shorten
  the test.
  Add a nominal test for AVG(DISTINCT)
sql/item_sum.cc:
  Implementation of cleaned up Item_sum_count_distinct.
  Fixed a bug with COUNT(DISTINCT) and new VARCHAR and collations.
  Fixed a bug wiht AVG(DISTINCT) and wrong number of output digits
  after decimal point.
sql/item_sum.h:
  Cleanup for Item_sum_count_distinct.
  Now if the list of distinct arguments doesn't contain a blob column,
  we always use Unique and merge-sort to find distinct values.
sql/sql_class.h:
  Added a short-cut to find number of elements in Unique if all elements fit
  into memory.

mysql-test/r/count_distinct2.result

@@ -116,7 +116,7 @@ count(distinct n)
 5000
 show status like 'Created_tmp_disk_tables';
 Variable_name	Value
-Created_tmp_disk_tables	1
+Created_tmp_disk_tables	0
 drop table t1;
 create table t1 (s text);
 flush status;

mysql-test/r/func_group.result

@@ -881,3 +881,10 @@ SELECT MAX(id) FROM t1 WHERE id < 3 AND a=2 AND b=6;
 MAX(id)
 NULL
 DROP TABLE t1;
+CREATE TABLE t1 (a VARCHAR(400));
+INSERT INTO t1 (a) VALUES ("A"), ("a"), ("a "), ("a   "),
+("B"), ("b"), ("b "), ("b   ");
+SELECT COUNT(DISTINCT a) FROM t1;
+COUNT(DISTINCT a)
+2
+DROP TABLE t1;

mysql-test/r/sum_distinct.result

@@ -98,60 +98,60 @@ DROP TABLE t1;
 CREATE TABLE t1 (id INTEGER);
 CREATE TABLE t2 (id INTEGER);
 INSERT INTO t1 (id) VALUES (1), (1), (1),(1);
-INSERT INTO t2 (id) SELECT id FROM t1;
-INSERT INTO t1 (id) SELECT id FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1;
 /* 8 */
-INSERT INTO t1 (id) SELECT id FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1;
 /* 12 */
-INSERT INTO t1 (id) SELECT id FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1;
 /* 16 */
-INSERT INTO t1 (id) SELECT id FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1;
 /* 20 */
-INSERT INTO t1 (id) SELECT id FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1;
 /* 24 */
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+1 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+2 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+4 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+8 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+16 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+32 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+64 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+128 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+256 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+512 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+1024 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+2048 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+4096 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+8192 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
+INSERT INTO t1 SELECT id+1 FROM t1;
+INSERT INTO t1 SELECT id+2 FROM t1;
+INSERT INTO t1 SELECT id+4 FROM t1;
+INSERT INTO t1 SELECT id+8 FROM t1;
+INSERT INTO t1 SELECT id+16 FROM t1;
+INSERT INTO t1 SELECT id+32 FROM t1;
+INSERT INTO t1 SELECT id+64 FROM t1;
+INSERT INTO t1 SELECT id+128 FROM t1;
+INSERT INTO t1 SELECT id+256 FROM t1;
+INSERT INTO t1 SELECT id+512 FROM t1;
+SELECT AVG(DISTINCT id) FROM t1 GROUP BY id % 13;
+AVG(DISTINCT id)
+513.5000
+508.0000
+509.0000
+510.0000
+511.0000
+512.0000
+513.0000
+514.0000
+515.0000
+516.0000
+517.0000
+511.5000
+512.5000
+SELECT SUM(DISTINCT id)/COUNT(DISTINCT id) FROM t1 GROUP BY id % 13;
+SUM(DISTINCT id)/COUNT(DISTINCT id)
+513.50000
+508.00000
+509.00000
+510.00000
+511.00000
+512.00000
+513.00000
+514.00000
+515.00000
+516.00000
+517.00000
+511.50000
+512.50000
+INSERT INTO t1 SELECT id+1024 FROM t1;
+INSERT INTO t1 SELECT id+2048 FROM t1;
+INSERT INTO t1 SELECT id+4096 FROM t1;
+INSERT INTO t1 SELECT id+8192 FROM t1;
 INSERT INTO t2 SELECT id FROM t1 ORDER BY id*rand();
 SELECT SUM(DISTINCT id) sm FROM t1;
 sm

mysql-test/t/func_group.test

@@ -591,3 +591,13 @@ INSERT INTO t1 VALUES
   (1,1,4), (2,2,1), (3,1,3), (4,2,1), (5,1,1);
 SELECT MAX(id) FROM t1 WHERE id < 3 AND a=2 AND b=6;
 DROP TABLE t1;
+
+#
+# Test that new VARCHAR correctly works with COUNT(DISTINCT)
+#
+
+CREATE TABLE t1 (a VARCHAR(400));
+INSERT INTO t1 (a) VALUES ("A"), ("a"), ("a "), ("a   "),
+                          ("B"), ("b"), ("b "), ("b   ");
+SELECT COUNT(DISTINCT a) FROM t1;
+DROP TABLE t1;

mysql-test/t/sum_distinct.test

@@ -103,64 +103,30 @@ CREATE TABLE t1 (id INTEGER);
 CREATE TABLE t2 (id INTEGER);
 
 INSERT INTO t1 (id) VALUES (1), (1), (1),(1);
-INSERT INTO t2 (id) SELECT id FROM t1;
-INSERT INTO t1 (id) SELECT id FROM t2; /* 8 */
-INSERT INTO t1 (id) SELECT id FROM t2; /* 12 */
-INSERT INTO t1 (id) SELECT id FROM t2; /* 16 */
-INSERT INTO t1 (id) SELECT id FROM t2; /* 20 */
-INSERT INTO t1 (id) SELECT id FROM t2; /* 24 */
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+1 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+2 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+4 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+8 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+16 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+32 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+64 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+128 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+256 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+512 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+1024 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+2048 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+4096 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-INSERT INTO t2 (id) SELECT id+8192 FROM t1;
-INSERT INTO t1 SELECT id FROM t2;
-DELETE FROM t2;
-#INSERT INTO t2 (id) SELECT id+16384 FROM t1;
-#INSERT INTO t1 SELECT id FROM t2;
-#DELETE FROM t2;
-#INSERT INTO t2 (id) SELECT id+32768 FROM t1;
-#INSERT INTO t1 SELECT id FROM t2;
-#DELETE FROM t2;
-#INSERT INTO t2 (id) SELECT id+65536 FROM t1;
-#INSERT INTO t1 SELECT id FROM t2;
-#DELETE FROM t2;
+INSERT INTO t1 (id) SELECT id FROM t1; /* 8 */
+INSERT INTO t1 (id) SELECT id FROM t1; /* 12 */
+INSERT INTO t1 (id) SELECT id FROM t1; /* 16 */
+INSERT INTO t1 (id) SELECT id FROM t1; /* 20 */
+INSERT INTO t1 (id) SELECT id FROM t1; /* 24 */
+INSERT INTO t1 SELECT id+1 FROM t1;
+INSERT INTO t1 SELECT id+2 FROM t1;
+INSERT INTO t1 SELECT id+4 FROM t1;
+INSERT INTO t1 SELECT id+8 FROM t1;
+INSERT INTO t1 SELECT id+16 FROM t1;
+INSERT INTO t1 SELECT id+32 FROM t1;
+INSERT INTO t1 SELECT id+64 FROM t1;
+INSERT INTO t1 SELECT id+128 FROM t1;
+INSERT INTO t1 SELECT id+256 FROM t1;
+INSERT INTO t1 SELECT id+512 FROM t1;
+
+# Just test that AVG(DISTINCT) is there
+SELECT AVG(DISTINCT id) FROM t1 GROUP BY id % 13;
+SELECT SUM(DISTINCT id)/COUNT(DISTINCT id) FROM t1 GROUP BY id % 13;
+
+INSERT INTO t1 SELECT id+1024 FROM t1;
+INSERT INTO t1 SELECT id+2048 FROM t1;
+INSERT INTO t1 SELECT id+4096 FROM t1;
+INSERT INTO t1 SELECT id+8192 FROM t1;
 INSERT INTO t2 SELECT id FROM t1 ORDER BY id*rand();
 
 # SELECT '++++++++++++++++++++++++++++++++++++++++++++++++++';

sql/item_sum.h

@@ -239,6 +239,7 @@ class Item_sum_avg_distinct: public Item_sum_distinct
 public:
   Item_sum_avg_distinct(Item *item_arg) : Item_sum_distinct(item_arg) {}
 
+  void fix_length_and_dec();
   virtual void calculate_val_and_count();
   enum Sumfunctype sum_func () const { return AVG_DISTINCT_FUNC; }
   const char *func_name() const { return "avg_distinct"; }
@@ -280,68 +281,44 @@ class TMP_TABLE_PARAM;
 class Item_sum_count_distinct :public Item_sum_int
 {
   TABLE *table;
-  table_map used_table_cache;
   uint32 *field_lengths;
   TMP_TABLE_PARAM *tmp_table_param;
-  TREE tree_base;
-  TREE *tree;
+  /*
+    If there are no blobs, we can use a tree, which
+    is faster than heap table. In that case, we still use the table
+    to help get things set up, but we insert nothing in it
+  */
+  Unique *tree;
   /*
     Following is 0 normal object and pointer to original one for copy 
     (to correctly free resources)
   */
   Item_sum_count_distinct *original;
+  uint tree_key_length;
 
-  uint key_length;
-  CHARSET_INFO *key_charset;
-  
-  /*
-    Calculated based on max_heap_table_size. If reached,
-    walk the tree and dump it into MyISAM table
-  */
-  uint max_elements_in_tree;
-
-  /*
-    The first few bytes of record ( at least one)
-    are just markers for deleted and NULLs. We want to skip them since
-    they will just bloat the tree without providing any valuable info
-  */
-  int rec_offset;
 
-  /*
-    If there are no blobs, we can use a tree, which
-    is faster than heap table. In that case, we still use the table
-    to help get things set up, but we insert nothing in it
-  */
-  bool use_tree;
   bool always_null;		// Set to 1 if the result is always NULL
 
-  int tree_to_myisam();
 
   friend int composite_key_cmp(void* arg, byte* key1, byte* key2);
   friend int simple_str_key_cmp(void* arg, byte* key1, byte* key2);
-  friend int simple_raw_key_cmp(void* arg, byte* key1, byte* key2);
-  friend int dump_leaf(byte* key, uint32 count __attribute__((unused)),
-		       Item_sum_count_distinct* item);
 
-  public:
+public:
   Item_sum_count_distinct(List<Item> &list)
-    :Item_sum_int(list), table(0), used_table_cache(~(table_map) 0),
-     tmp_table_param(0), tree(&tree_base), original(0), use_tree(0),
-     always_null(0)
+    :Item_sum_int(list), table(0), field_lengths(0), tmp_table_param(0),
+     tree(0), original(0), always_null(FALSE)
   { quick_group= 0; }
   Item_sum_count_distinct(THD *thd, Item_sum_count_distinct *item)
     :Item_sum_int(thd, item), table(item->table),
-     used_table_cache(item->used_table_cache),
      field_lengths(item->field_lengths),
      tmp_table_param(item->tmp_table_param),
-     tree(item->tree), original(item), key_length(item->key_length),
-     max_elements_in_tree(item->max_elements_in_tree),
-     rec_offset(item->rec_offset), use_tree(item->use_tree),
+     tree(item->tree), original(item), tree_key_length(item->tree_key_length),
      always_null(item->always_null)
   {}
+  ~Item_sum_count_distinct();
+
   void cleanup();
 
-  table_map used_tables() const { return used_table_cache; }
   enum Sumfunctype sum_func () const { return COUNT_DISTINCT_FUNC; }
   void clear();
   bool add();

sql/sql_class.h

@@ -1831,6 +1831,7 @@ class Unique :public Sql_alloc
   Unique(qsort_cmp2 comp_func, void *comp_func_fixed_arg,
 	 uint size_arg, ulong max_in_memory_size_arg);
   ~Unique();
+  ulong elements_in_tree() { return tree.elements_in_tree; }
   inline bool unique_add(void *ptr)
   {
     DBUG_ENTER("unique_add");