MWL#90: code cleanup

- Remove deadcode
- Improve comments 
- Do small several small TODOs

sql/item_cmpfunc.cc

@@ -5731,10 +5731,8 @@ Item_field* Item_equal::get_first(Item_field *field)
   {
     /*
       It's a field from an materialized semi-join. We can substitute it only
-      for a field from the same semi-join.
+      for a field from the same semi-join. Find the first of such items.
     */
-
-    /* Find an item to substitute for. */
     while ((item= it++))
     {
       if (item->field->table->pos_in_table_list->embedding == emb_nest)
@@ -5762,7 +5760,7 @@ Item_field* Item_equal::get_first(Item_field *field)
       First let's note that either it1.col or it2.col participates in 
       subquery's IN-equality. It can't be otherwise, because materialization is
       only applicable to uncorrelated subqueries, so the only way we could
-      infer "it1.col=ot1.col" is from IN-equality. Ok, so IN-eqality has 
+      infer "it1.col=ot1.col" is from the IN-equality. Ok, so IN-eqality has 
       it1.col or it2.col on its inner side. it1.col is first such item in the
       join order, so it's not possible for SJ-Mat to be
       SJ-Materialization-lookup, it is SJ-Materialization-Scan. The scan part

sql/opt_subselect.cc

@@ -30,6 +30,8 @@
   3.1.1 Merged semi-join TABLE_LIST object
   3.1.2 Non-merged semi-join data structure
   3.2 Semi-joins and query optimization
+  3.2.1 Non-merged semi-joins and join optimization
+  3.2.2 Merged semi-joins and join optimization
   3.3 Semi-joins and query execution
 
   1. What is a semi-join subquery
@@ -99,8 +101,8 @@
   3. Code-level view of semi-join processing
   ------------------------------------------
   
-  3.1 Conversion
-  --------------
+  3.1 Conversion and pre-optimization data structures
+  ---------------------------------------------------
   * When doing JOIN::prepare for the subquery, we detect that it can be
     converted into a semi-join and register it in parent_join->sj_subselects
 
@@ -117,7 +119,7 @@
   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   Merged semi-join object is a TABLE_LIST that contains a sub-join of 
   subquery tables and the semi-join ON expression (in this respect it is 
-  ery similar to nested outer join representation)
+  very similar to nested outer join representation)
   Merged semi-join represents this SQL:
 
     ... SEMI JOIN (inner_tbl1 JOIN ... JOIN inner_tbl_n) ON sj_on_expr
@@ -135,14 +137,38 @@
   clause. (They do remain in PS-saved WHERE clause, and they replace themselves
   with Item_int(1) on subsequent re-executions).
 
-  3.2 Semi-joins and query optimization
-  -------------------------------------
-  Query optimizer operates on semi-join nests.
+  3.2 Semi-joins and join optimization
+  ------------------------------------
+  
+  3.2.1 Non-merged semi-joins and join optimization
+  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+  For join optimization purposes, non-merged semi-join nests are similar to
+  base tables - they've got one JOIN_TAB, which can be accessed with one of
+  two methods:
+   - full table scan (representing SJ-Materialization-Scan strategy)
+   - eq_ref-like table lookup (representing SJ-Materialization-Lookup)
+
+  Unlike regular base tables, non-merged semi-joins have:
+   - non-zero JOIN_TAB::startup_cost, and
+   - join_tab->table->is_filled_at_execution()==TRUE, which means one
+     cannot do const table detection or range analysis or other table data-
+     dependent inferences
+  // instead, get_delayed_table_estimates() runs optimization on the nest so that 
+  // we get an idea about temptable size
+  
+  3.2.2 Merged semi-joins and join optimization
+  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+   - optimize_semijoin_nests() does pre-optimization 
+   - during join optimization, the join has one JOIN_TAB (or is it POSITION?) 
+     array, and suffix-based detection is used, see advance_sj_state()
+   - after join optimization is done, get_best_combination() switches 
+     the data-structure to prefix-based, multiple JOIN_TAB ranges format.
 
   3.3 Semi-joins and query execution
   ----------------------------------
   * Join executor has hooks for all semi-join strategies.
-    TODO elaborate
+    TODO elaborate.
+
 */
 
 
@@ -2682,8 +2708,6 @@ bool setup_sj_materialization(JOIN_TAB *sjm_tab)
       temptable.
     */
     TABLE_REF *tab_ref;
-    //if (!(tab_ref= (TABLE_REF*) thd->alloc(sizeof(TABLE_REF))))
-    //  DBUG_RETURN(TRUE); /* purecov: inspected */
     tab_ref= &sjm_tab->ref;
     tab_ref->key= 0; /* The only temp table index. */
     tab_ref->key_length= tmp_key->key_length;
@@ -3966,17 +3990,19 @@ static void remove_subq_pushed_predicates(JOIN *join, Item **where)
     pointers.
   
   RETURN 
-    FALSE  Ok
-    TRUE   Error, join execution is not possible.
+    NESTED_LOOP_OK - OK
+    NESTED_LOOP_ERROR| NESTED_LOOP_KILLED - Error, abort the join execution
 */
 
-bool join_tab_execution_startup(JOIN_TAB *tab)
+enum_nested_loop_state join_tab_execution_startup(JOIN_TAB *tab)
 {
   Item_in_subselect *in_subs;
   DBUG_ENTER("join_tab_execution_startup");
+
   if (tab->table->pos_in_table_list && 
       (in_subs= tab->table->pos_in_table_list->jtbm_subselect))
   {
+    /* It's a non-merged SJM nest */
     DBUG_ASSERT(in_subs->engine->engine_type() ==
                 subselect_engine::HASH_SJ_ENGINE);
 
@@ -3988,13 +4014,13 @@ bool join_tab_execution_startup(JOIN_TAB *tab)
       hash_sj_engine->is_materialized= TRUE; 
 
       if (hash_sj_engine->materialize_join->error || tab->join->thd->is_fatal_error)
-        DBUG_RETURN(TRUE);
+        DBUG_RETURN(NESTED_LOOP_ERROR);
     }
   }
   else if (tab->bush_children)
   {
     /* It's a merged SJM nest */
-    int rc; // psergey3: todo: error codes!
+    enum_nested_loop_state rc;
     JOIN *join= tab->join;
     SJ_MATERIALIZATION_INFO *sjm= tab->bush_children->start->emb_sj_nest->sj_mat_info;
     JOIN_TAB *join_tab= tab->bush_children->start;
@@ -4018,6 +4044,6 @@ bool join_tab_execution_startup(JOIN_TAB *tab)
     }
   }
 
-  DBUG_RETURN(0);
+  DBUG_RETURN(NESTED_LOOP_OK);
 }
 

sql/opt_subselect.h

@@ -372,5 +372,5 @@ void get_delayed_table_estimates(TABLE *table,
                                  double *scan_time,
                                  double *startup_cost);
 
-bool join_tab_execution_startup(JOIN_TAB *tab);
+enum_nested_loop_state join_tab_execution_startup(JOIN_TAB *tab);
 

sql/sql_join_cache.cc

@@ -1809,11 +1809,8 @@ enum_nested_loop_state JOIN_CACHE_BNL::join_matching_records(bool skip_last)
 
   /* Start retrieving all records of the joined table */
   
-  if (join_tab_execution_startup(join_tab))
-  {
-    rc= NESTED_LOOP_ERROR;
+  if ((rc= join_tab_execution_startup(join_tab)) < 0)
     goto finish;
-  }
 
   if ((error= join_init_read_record(join_tab))) 
   {

sql/sql_select.cc

@@ -8100,6 +8100,8 @@ void JOIN::cleanup(bool full)
     {
       for (tab= top_jtrange_tables?join_tab:NULL; tab; tab= next_linear_tab(this, tab, TRUE))
 	tab->cleanup();
+      //psergey4: how is the above supposed to work when
+      //top_jtrange_tables==FALSE? It will crash right away!
       table= 0;
     }
     else
@@ -12745,133 +12747,6 @@ int rr_sequential_and_unpack(READ_RECORD *info)
 }
 
 
-/*
-  Semi-join materialization join function
-
-  SYNOPSIS
-    sub_select_sjm()
-      join            The join
-      join_tab        The first table in the materialization nest
-      end_of_records  FALSE <=> This call is made to pass another record 
-                                combination
-                      TRUE  <=> EOF
-
-  DESCRIPTION
-    This is a join execution function that does materialization of a join
-    suborder before joining it to the rest of the join.
-
-    The table pointed by join_tab is the first of the materialized tables.
-    This function first creates the materialized table and then switches to
-    joining the materialized table with the rest of the join.
-
-    The materialized table can be accessed in two ways:
-     - index lookups
-     - full table scan
-
-  RETURN
-    One of enum_nested_loop_state values
-*/
-#if 0 
-enum_nested_loop_state
-sub_select_sjm(JOIN *join, JOIN_TAB *join_tab, bool end_of_records)
-{
-  int res;
-  enum_nested_loop_state rc;
-
-  DBUG_ENTER("sub_select_sjm");
-
-  if (!join_tab->emb_sj_nest)
-  {
-    /*
-      We're handling GROUP BY/ORDER BY, this is the first table, and we've
-      actually executed the join already and now we're just reading the
-      result of the join from the temporary table.
-      Bypass to regular join handling.
-      Yes, it would be nicer if sub_select_sjm wasn't called at all in this
-      case but there's no easy way to arrange this.
-    */
-    rc= sub_select(join, join_tab, end_of_records);
-    DBUG_RETURN(rc);
-  }
-
-  SJ_MATERIALIZATION_INFO *sjm= join_tab->emb_sj_nest->sj_mat_info;
-  if (end_of_records)
-  {
-    rc= (*join_tab[sjm->tables - 1].next_select)(join,
-                                                 join_tab + sjm->tables,
-                                                 end_of_records);
-    DBUG_RETURN(rc);
-  }
-  if (!sjm->materialized)
-  {
-    /*
-      Do the materialization. First, put end_sj_materialize after the last
-      inner table so we can catch record combinations of sj-inner tables.
-    */
-    Next_select_func next_func= join_tab[sjm->tables - 1].next_select;
-    join_tab[sjm->tables - 1].next_select= end_sj_materialize;
-
-    /*
-      Now run the join for the inner tables. The first call is to run the
-      join, the second one is to signal EOF (this is essential for some
-      join strategies, e.g. it will make join buffering flush the records)
-    */
-    if ((rc= sub_select(join, join_tab, FALSE)) < 0 ||
-        (rc= sub_select(join, join_tab, TRUE/*EOF*/)) < 0)
-    {
-      join_tab[sjm->tables - 1].next_select= next_func;
-      DBUG_RETURN(rc); /* it's NESTED_LOOP_(ERROR|KILLED)*/
-    }
-    join_tab[sjm->tables - 1].next_select= next_func;
-
-    /*
-      Ok, materialization finished. Initialize the access to the temptable
-    */
-    sjm->materialized= TRUE;
-    join_tab->read_record.read_record= join_no_more_records;
-    if (sjm->is_sj_scan)
-    {
-      /* Initialize full scan */
-      JOIN_TAB *last_tab= join_tab + (sjm->tables - 1);
-      init_read_record(&last_tab->read_record, join->thd,
-                       sjm->table, NULL, TRUE, TRUE, FALSE);
-
-      DBUG_ASSERT(last_tab->read_record.read_record == rr_sequential);
-      last_tab->read_first_record= join_read_record_no_init;
-      last_tab->read_record.copy_field= sjm->copy_field;
-      last_tab->read_record.copy_field_end= sjm->copy_field +
-                                            sjm->sjm_table_cols.elements;
-      last_tab->read_record.read_record= rr_sequential_and_unpack;
-    }
-  }
-
-  if (sjm->is_sj_scan)
-  {
-    /* Do full scan of the materialized table */
-    JOIN_TAB *last_tab= join_tab + (sjm->tables - 1);
-
-    Item *save_cond= last_tab->select_cond;
-    last_tab->set_select_cond(sjm->join_cond, __LINE__);
-
-    rc= sub_select(join, last_tab, end_of_records);
-    last_tab->set_select_cond(save_cond, __LINE__);
-    DBUG_RETURN(rc);
-  }
-  else
-  {
-    /* Do index lookup in the materialized table */
-    if ((res= join_read_key2(join_tab, sjm->table, sjm->tab_ref)) == 1)
-      DBUG_RETURN(NESTED_LOOP_ERROR); /* purecov: inspected */
-    if (res || !sjm->in_equality->val_int())
-      DBUG_RETURN(NESTED_LOOP_NO_MORE_ROWS);
-  }
-  rc= (*join_tab[sjm->tables - 1].next_select)(join,
-                                               join_tab + sjm->tables,
-                                               end_of_records);
-  DBUG_RETURN(rc);
-}
-#endif
-
 /*
   Fill the join buffer with partial records, retrieve all full  matches for them   
 
@@ -13122,8 +12997,8 @@ sub_select(JOIN *join,JOIN_TAB *join_tab,bool end_of_records)
     }
     join->thd->row_count= 0;
     
-    if (join_tab_execution_startup(join_tab))
-      DBUG_RETURN(NESTED_LOOP_ERROR);
+    if ((rc= join_tab_execution_startup(join_tab)) < 0)
+      DBUG_RETURN(rc);
 
     error= (*join_tab->read_first_record)(join_tab);
 

sql/sql_select.h

@@ -1247,9 +1247,6 @@ enum_nested_loop_state sub_select_cache(JOIN *join, JOIN_TAB *join_tab, bool
                                         end_of_records);
 enum_nested_loop_state sub_select(JOIN *join,JOIN_TAB *join_tab, bool
                                   end_of_records);
-enum_nested_loop_state sub_select_sjm(JOIN *join, JOIN_TAB *join_tab, 
-                                      bool end_of_records);
-
 enum_nested_loop_state
 end_send_group(JOIN *join, JOIN_TAB *join_tab __attribute__((unused)),
 	       bool end_of_records);