MWL#89: Cost-based choice between Materialization and IN->EXISTS transformation

- Changed the default optimizer switches to provide 5.1/5.2 compatible behavior
- Added a regression test file to test consistently all cases covered by MWL#89
- Added/corrected/improved comments.

mysql-test/t/subselect_mat_cost.test

+#
+# Tets of cost-based choice between the materialization and in-to-exists
+# subquery execution strategies (MWL#89)
+#
+
+--disable_warnings
+drop table if exists t1, t2, t3, t1i, t2i, t3i;
+--enable_warnings
+
+create table t1 (a1 char(8), a2 char(8));
+create table t2 (b1 char(8), b2 char(8));
+create table t3 (c1 char(8), c2 char(8));
+
+--echo Unindexed tables
+insert into t1 values ('1 - 00', '2 - 00');
+insert into t1 values ('1 - 01', '2 - 01');
+insert into t1 values ('1 - 02', '2 - 02');
+
+insert into t2 values ('1 - 01', '2 - 01');
+insert into t2 values ('1 - 01', '2 - 01');     
+insert into t2 values ('1 - 02', '2 - 02');
+insert into t2 values ('1 - 02', '2 - 02');
+insert into t2 values ('1 - 03', '2 - 03');
+
+insert into t3 values ('1 - 01', '2 - 01');
+insert into t3 values ('1 - 02', '2 - 02');
+insert into t3 values ('1 - 03', '2 - 03');
+insert into t3 values ('1 - 04', '2 - 04');
+
+--echo Indexed tables
+create table t1i (a1 char(8), a2 char(8));
+create table t2i (b1 char(8), b2 char(8));
+create table t3i (c1 char(8), c2 char(8));
+create index it1i1 on t1i (a1);
+create index it1i2 on t1i (a2);
+create index it1i3 on t1i (a1, a2);
+
+create index it2i1 on t2i (b1);
+create index it2i2 on t2i (b2);
+create index it2i3 on t2i (b1, b2);
+
+create index it3i1 on t3i (c1);
+create index it3i2 on t3i (c2);
+create index it3i3 on t3i (c1, c2);
+
+insert into t1i select * from t1;
+insert into t2i select * from t2;
+insert into t3i select * from t3;
+
+--echo 
+--echo 1. Both materialization and in-to-exists are possible to execute
+--echo 
+set @@optimizer_switch='materialization=on,in_to_exists=on';
+
+--echo 
+--echo 1.1 In-to-exists is cheaper
+
+set @@optimizer_switch='semijoin=off';
+explain extended
+select * from t1 where a1 in (select b1 from t2 where b1 > '0');
+select * from t1 where a1 in (select b1 from t2 where b1 > '0');
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0');
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0');
+
+set @@optimizer_switch='semijoin=on';
+explain extended
+select * from t1 where a1 in (select b1 from t2 where b1 > '0' group by b1);
+select * from t1 where a1 in (select b1 from t2 where b1 > '0' group by b1);
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2);
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2);
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2 having b2 <  '2 - 04');
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2 having b2 <  '2 - 04');
+
+--echo 
+--echo 1.2 Materialization is cheaper
+
+--echo 1.2.1 Materialization is executable
+
+insert into t1 values ('1 - 03', '2 - 03');
+insert into t1 values ('1 - 04', '2 - 04');
+insert into t1 values ('1 - 05', '2 - 05');
+
+set @@optimizer_switch='semijoin=off';
+explain extended
+select * from t1 where a1 in (select b1 from t2 where b1 > '0');
+select * from t1 where a1 in (select b1 from t2 where b1 > '0');
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0');
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0');
+
+set @@optimizer_switch='semijoin=on';
+explain extended
+select * from t1 where a1 in (select b1 from t2 where b1 > '0' group by b1);
+select * from t1 where a1 in (select b1 from t2 where b1 > '0' group by b1);
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2);
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2);
+explain extended
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2 having b2 <  '2 - 04');
+select * from t1 where (a1, a2) in (select b1, b2 from t2 where b1 > '0' group by b1, b2 having b2 <  '2 - 04');
+
+delete from t1 where a1 >= '1 - 03';
+
+--echo 1.2.2 Materialization is NOT executable, revert to in-to-exists
+
+--echo 
+--echo 2. Materialization is OFF, in-to-exists is ON
+--echo 
+set @@optimizer_switch='materialization=off,in_to_exists=on';
+
+--echo 
+--echo 3. Materialization is ON, in-to-exists is OFF
+--echo 
+set @@optimizer_switch='materialization=on,in_to_exists=off';
+
+--echo 3.1 Materialization is executable
+--echo 3.2 Materialization is NOT executable, revert to in-to-exists
+
+--echo 
+--echo 4. Edge cases
+--echo 
+--echo 4.0 Both materialization and in_to_exists cannot be off
+set @@optimizer_switch='materialization=off,in_to_exists=off';
+--error ER_ILLEGAL_SUBQUERY_OPTIMIZER_SWITCHES
+select * from t1 where a1 in (select b1 from t2 where b1 > '0' group by b1);
+
+--echo 4.1 Outer query with no tables
+--echo 4.2 Subquery with no tables
+--echo 4.3 optimize_cond detects FALSE where/having clause
+--echo 4.4 opt_sum_query detects no matching min/max row or substitutes MIN/MAX with a const
+--echo 4.4 make_join_select detects impossible WHERE
+--echo 4.5 constant optimization detects "no matching row in const table"
+
+--echo 5. UPDATE/DELETE with subqueries
+
+drop table t1, t2, t3, t1i, t2i, t3i;

sql/mysql_priv.h

@@ -581,6 +581,11 @@ class Default_object_creation_ctx : public Object_creation_ctx
 
 #ifdef DBUG_OFF 
 /* The following must be kept in sync with optimizer_switch_str in mysqld.cc */
+/*
+TODO: Materialization is off by default to mimic 5.1/5.2 behavior.
+Once cost based choice between materialization and in-to-exists should be
+enabled by default, add OPTIMIZER_SWITCH_MATERIALIZATION
+*/
 #  define OPTIMIZER_SWITCH_DEFAULT (OPTIMIZER_SWITCH_INDEX_MERGE | \
                                     OPTIMIZER_SWITCH_INDEX_MERGE_UNION | \
                                     OPTIMIZER_SWITCH_INDEX_MERGE_SORT_UNION | \
@@ -588,7 +593,6 @@ class Default_object_creation_ctx : public Object_creation_ctx
                                     OPTIMIZER_SWITCH_INDEX_COND_PUSHDOWN | \
                                     OPTIMIZER_SWITCH_FIRSTMATCH | \
                                     OPTIMIZER_SWITCH_LOOSE_SCAN | \
-                                    OPTIMIZER_SWITCH_MATERIALIZATION | \
                                     OPTIMIZER_SWITCH_IN_TO_EXISTS | \
                                     OPTIMIZER_SWITCH_SEMIJOIN | \
                                     OPTIMIZER_SWITCH_PARTIAL_MATCH_ROWID_MERGE|\
@@ -603,7 +607,6 @@ class Default_object_creation_ctx : public Object_creation_ctx
                                     OPTIMIZER_SWITCH_TABLE_ELIMINATION | \
                                     OPTIMIZER_SWITCH_FIRSTMATCH | \
                                     OPTIMIZER_SWITCH_LOOSE_SCAN | \
-                                    OPTIMIZER_SWITCH_MATERIALIZATION | \
                                     OPTIMIZER_SWITCH_IN_TO_EXISTS | \
                                     OPTIMIZER_SWITCH_SEMIJOIN | \
                                     OPTIMIZER_SWITCH_PARTIAL_MATCH_ROWID_MERGE|\

sql/mysqld.cc

@@ -412,7 +412,7 @@ static const char *optimizer_switch_str="index_merge=on,index_merge_union=on,"
                                         "index_condition_pushdown=on,"
                                         "firstmatch=on,"
                                         "loosescan=on,"
-                                        "materialization=on,"
+                                        "materialization=off,"
                                         "in_to_exists=on,"
                                         "semijoin=on,"
                                         "partial_match_rowid_merge=on,"

sql/opt_subselect.cc

@@ -3643,15 +3643,15 @@ bool JOIN::choose_subquery_plan(table_map join_tables)
     JOIN *outer_join= unit->outer_select() ? unit->outer_select()->join : NULL;
     JOIN *inner_join= this;
     /* Cost of the outer JOIN. */
-    double outer_read_time= 0, outer_record_count= 0;
+    double outer_read_time, outer_record_count;
     /* Cost of the unmodified subquery. */
-    double inner_read_time_1= 0, inner_record_count_1= 0;
+    double inner_read_time_1, inner_record_count_1;
     /* Cost of the subquery with injected IN-EXISTS predicates. */
-    double inner_read_time_2= 0, inner_record_count_2= 0;
+    double inner_read_time_2, inner_record_count_2;
     /* The cost to compute IN via materialization. */
-    double materialize_strategy_cost= 0;
+    double materialize_strategy_cost;
     /* The cost of the IN->EXISTS strategy. */
-    double in_exists_strategy_cost= 1;
+    double in_exists_strategy_cost;
 
     if (outer_join)
       get_partial_join_cost(outer_join, outer_join->tables,
@@ -3688,6 +3688,7 @@ bool JOIN::choose_subquery_plan(table_map join_tables)
     {
       /* Reoptimization would not produce any better plan. */
       inner_read_time_2= inner_read_time_1;
+      inner_record_count_2= inner_record_count_1;
     }
 
     /* Compute execution costs. */

sql/sql_select.cc

@@ -1308,7 +1308,6 @@ JOIN::optimize()
   if (!(select_options & SELECT_DESCRIBE))
     init_ftfuncs(thd, select_lex, test(order));
 
-  /* Create all structures needed for materialized subquery execution. */
   if (optimize_unflattened_subqueries())
     DBUG_RETURN(1);
   
@@ -1411,8 +1410,6 @@ JOIN::optimize()
   /*
     Even with zero matching rows, subqueries in the HAVING clause may
     need to be evaluated if there are aggregate functions in the query.
-    If we planned to materialize the subquery, we need to set it up
-    properly before prematurely leaving optimize().
   */
   if (optimize_unflattened_subqueries())
     DBUG_RETURN(1);