Subquery optimizations: non-semijoin materialization

- Backport into Maria DB 5.3, part 1

sql/item.h

@@ -878,6 +878,9 @@ class Item {
   static CHARSET_INFO *default_charset();
   virtual CHARSET_INFO *compare_collation() { return NULL; }
 
+  /* Cache of the result of is_expensive(). */
+  int8 is_expensive_cache;
+  
   virtual bool walk(Item_processor processor, bool walk_subquery, uchar *arg)
   {
     return (this->*processor)(arg);
@@ -1081,6 +1084,29 @@ class Item {
   */
   virtual bool result_as_longlong() { return FALSE; }
   bool is_datetime();
+
+  /*
+    Test whether an expression is expensive to compute. Used during
+    optimization to avoid computing expensive expressions during this
+    phase. Also used to force temp tables when sorting on expensive
+    functions.
+    TODO:
+    Normally we should have a method:
+      cost Item::execution_cost(),
+    where 'cost' is either 'double' or some structure of various cost
+    parameters.
+
+    NOTE
+      This function is now used to prevent evaluation of materialized IN
+      subquery predicates before it is allowed. grep for 
+      DontEvaluateMaterializedSubqueryTooEarly to see the uses.
+  */
+  virtual bool is_expensive()
+  {
+    if (is_expensive_cache < 0)
+      is_expensive_cache= walk(&Item::is_expensive_processor, 0, (uchar*)0);
+    return test(is_expensive_cache);
+  }
   virtual Field::geometry_type get_geometry_type() const
     { return Field::GEOM_GEOMETRY; };
   String *check_well_formed_result(String *str, bool send_error= 0);
@@ -2842,9 +2868,10 @@ class Cached_item_field :public Cached_item
   uint length;
 
 public:
-  Cached_item_field(Item_field *item)
+  Cached_item_field(Field *arg_field) : field(arg_field)
   {
-    field= item->field;
+    field= arg_field;
+    /* TODO: take the memory allocation below out of the constructor. */
     buff= (uchar*) sql_calloc(length=field->pack_length());
   }
   bool cmp(void);
@@ -3271,7 +3298,8 @@ void mark_select_range_as_dependent(THD *thd,
                                     Field *found_field, Item *found_item,
                                     Item_ident *resolved_item);
 
-extern Cached_item *new_Cached_item(THD *thd, Item *item);
+extern Cached_item *new_Cached_item(THD *thd, Item *item,
+                                    bool use_result_field);
 extern Item_result item_cmp_type(Item_result a,Item_result b);
 extern void resolve_const_item(THD *thd, Item **ref, Item *cmp_item);
 extern int stored_field_cmp_to_item(Field *field, Item *item);

sql/item_buff.cc

@@ -27,11 +27,16 @@
   Create right type of Cached_item for an item.
 */
 
-Cached_item *new_Cached_item(THD *thd, Item *item)
+Cached_item *new_Cached_item(THD *thd, Item *item, bool use_result_field)
 {
   if (item->real_item()->type() == Item::FIELD_ITEM &&
       !(((Item_field *) (item->real_item()))->field->flags & BLOB_FLAG))
-    return new Cached_item_field((Item_field *) (item->real_item()));
+  {
+    Item_field *real_item= (Item_field *) item->real_item();
+    Field *cached_field= use_result_field ? real_item->result_field :
+                                            real_item->field;
+    return new Cached_item_field(cached_field);
+  }
   switch (item->result_type()) {
   case STRING_RESULT:
     return new Cached_item_str(thd, (Item_field *) item);

sql/item_func.cc

@@ -490,12 +490,12 @@ Field *Item_func::tmp_table_field(TABLE *table)
   return field;
 }
 
-
+/*
 bool Item_func::is_expensive_processor(uchar *arg)
 {
   return is_expensive();
 }
-
+*/
 
 my_decimal *Item_func::val_decimal(my_decimal *decimal_value)
 {

sql/item_func.h

@@ -192,8 +192,8 @@ class Item_func :public Item_result_field
                 Item_transformer transformer, uchar *arg_t);
   void traverse_cond(Cond_traverser traverser,
                      void * arg, traverse_order order);
-  bool is_expensive_processor(uchar *arg);
-  virtual bool is_expensive() { return 0; }
+ // bool is_expensive_processor(uchar *arg);
+ // virtual bool is_expensive() { return 0; }
   inline double fix_result(double value)
   {
     if (isfinite(value))
@@ -1053,6 +1053,7 @@ class Item_udf_func :public Item_func
 {
 protected:
   udf_handler udf;
+  bool is_expensive_processor(uchar *arg) { return TRUE; }
 
 public:
   Item_udf_func(udf_func *udf_arg)
@@ -1677,6 +1678,9 @@ class Item_func_sp :public Item_func
   bool execute_impl(THD *thd);
   bool init_result_field(THD *thd);
 
+protected:
+  bool is_expensive_processor(uchar *arg) { return TRUE; }
+  
 public:
 
   Item_func_sp(Name_resolution_context *context_arg, sp_name *name);

sql/opt_range.cc

@@ -5493,7 +5493,7 @@ static SEL_TREE *get_mm_tree(RANGE_OPT_PARAM *param,COND *cond)
     DBUG_RETURN(tree);
   }
   /* Here when simple cond */
-  if (cond->const_item())
+  if (cond->const_item() && !cond->is_expensive())
   {
     /*
       During the cond->val_int() evaluation we can come across a subselect 

sql/sql_lex.cc

@@ -2076,16 +2076,28 @@ void st_select_lex::print_limit(THD *thd,
 {
   SELECT_LEX_UNIT *unit= master_unit();
   Item_subselect *item= unit->item;
-  if (item && unit->global_parameters == this &&
-      (item->substype() == Item_subselect::EXISTS_SUBS ||
-       item->substype() == Item_subselect::IN_SUBS ||
-       item->substype() == Item_subselect::ALL_SUBS))
+
+  if (item && unit->global_parameters == this)
+  {
+    Item_subselect::subs_type subs_type= item->substype();
+    if (subs_type == Item_subselect::EXISTS_SUBS ||
+        subs_type == Item_subselect::IN_SUBS ||
+        subs_type == Item_subselect::ALL_SUBS)
     {
       DBUG_ASSERT(!item->fixed ||
-                (select_limit->val_int() == LL(1) && offset_limit == 0));
+                  /*
+                    If not using materialization both:
+                    select_limit == 1, and there should be no offset_limit.
+                  */
+                  (((subs_type == Item_subselect::IN_SUBS) &&
+                    ((Item_in_subselect*)item)->exec_method ==
+                    Item_in_subselect::MATERIALIZATION) ?
+                   TRUE :
+                   (select_limit->val_int() == 1LL) &&
+                   offset_limit == 0));
       return;
     }
-
+  }
   if (explicit_limit)
   {
     str->append(STRING_WITH_LEN(" limit "));
@@ -2098,6 +2110,7 @@ void st_select_lex::print_limit(THD *thd,
   }
 }
 
+
 /**
   @brief Restore the LEX and THD in case of a parse error.
 

sql/sql_lex.h

@@ -559,7 +559,8 @@ class st_select_lex_unit: public st_select_lex_node {
   bool add_fake_select_lex(THD *thd);
   void init_prepare_fake_select_lex(THD *thd);
   inline bool is_prepared() { return prepared; }
-  bool change_result(select_subselect *result, select_subselect *old_result);
+  bool change_result(select_result_interceptor *result,
+                     select_result_interceptor *old_result);
   void set_limit(st_select_lex *values);
   void set_thd(THD *thd_arg) { thd= thd_arg; }
   inline bool is_union (); 

sql/sql_select.h

@@ -1199,6 +1199,14 @@ enum_nested_loop_state sub_select_sjm(JOIN *join, JOIN_TAB *join_tab,
                                       bool end_of_records);
 int do_sj_dups_weedout(THD *thd, SJ_TMP_TABLE *sjtbl);
 
+enum_nested_loop_state
+end_send_group(JOIN *join, JOIN_TAB *join_tab __attribute__((unused)),
+	       bool end_of_records);
+enum_nested_loop_state
+end_write_group(JOIN *join, JOIN_TAB *join_tab __attribute__((unused)),
+		bool end_of_records);
+
+
 /**
   Information about a position of table within a join order. Used in join
   optimization.
@@ -1945,6 +1953,7 @@ bool error_if_full_join(JOIN *join);
 int report_error(TABLE *table, int error);
 int safe_index_read(JOIN_TAB *tab);
 COND *remove_eq_conds(THD *thd, COND *cond, Item::cond_result *cond_value);
+int test_if_item_cache_changed(List<Cached_item> &list);
 void calc_used_field_length(THD *thd, JOIN_TAB *join_tab);
 int join_init_read_record(JOIN_TAB *tab);
 void set_position(JOIN *join,uint idx,JOIN_TAB *table,KEYUSE *key);

sql/sql_union.cc

@@ -715,8 +715,8 @@ void st_select_lex_unit::reinit_exec_mechanism()
     TRUE  - error
 */
 
-bool st_select_lex_unit::change_result(select_subselect *new_result,
-                                       select_subselect *old_result)
+bool st_select_lex_unit::change_result(select_result_interceptor *new_result,
+                                       select_result_interceptor *old_result)
 {
   bool res= FALSE;
   for (SELECT_LEX *sl= first_select(); sl; sl= sl->next_select())