Bug#20487336: Assert 'length > 0 && keyparts != 0' in calc_length_and_keyparts

Preliminary patch:

Refactor TABLE_LIST::sj_on_expr:
Rename member to m_sj_cond (it is indeed a condition).
Add setter set_sj_cond() and getter sj_cond().

Author: roylyseng

sql/aggregate_check.cc

@@ -1028,13 +1028,13 @@ void Group_check::find_fd_in_joined_table(List<TABLE_LIST> *join_list)
   TABLE_LIST *table;
   while ((table= li++))
   {
-    if (table->sj_on_expr)
+    if (table->sj_cond())
     {
       /*
         We can ignore this nest as:
-        - the subquery's WHERE was copied to sj_on_expr
+        - the subquery's WHERE was copied to the semi-join condition
         - so where the IN equalities
-        - sj_on_expr is also present in the parent nest's join condition or in
+        - sj_cond() is also present in the parent nest's join condition or in
         the query block's WHERE condition, which we check somewhere else.
         Note that columns from sj-inner tables can help only as "intermediate
         link" in the graph of functional dependencies, as they are neither in

sql/sql_lex.cc

@@ -2886,7 +2886,7 @@ static void print_table_array(THD *thd, String *str, TABLE_LIST **table,
     }
     else if (curr->straight)
       str->append(STRING_WITH_LEN(" straight_join "));
-    else if (curr->sj_on_expr)
+    else if (curr->sj_cond())
       str->append(STRING_WITH_LEN(" semi join "));
     else
       str->append(STRING_WITH_LEN(" join "));
@@ -2952,12 +2952,12 @@ static void print_join(THD *thd,
     not a semi-join nest. This is necessary because "A SEMIJOIN B" is not the
     same as "B SEMIJOIN A".
   */
-  if ((*table)->sj_on_expr)
+  if ((*table)->sj_cond())
   {
     TABLE_LIST **end= table + non_const_tables;
     for (TABLE_LIST **t2= table; t2!=end; t2++)
     {
-      if (!(*t2)->sj_on_expr)
+      if (!(*t2)->sj_cond())
       {
         TABLE_LIST *tmp= *t2;
         *t2= *table;

sql/sql_optimizer.cc

@@ -4317,7 +4317,7 @@ uint build_bitmap_for_nested_joins(List<TABLE_LIST> *join_list,
     if ((nested_join= table->nested_join))
     {
       // We should have either a join condition or a semi-join condition
-      DBUG_ASSERT((table->join_cond() == NULL) == (table->sj_on_expr != NULL));
+      DBUG_ASSERT((table->join_cond() == NULL) == (table->sj_cond() != NULL));
 
       nested_join->nj_map= 0;
       nested_join->nj_total= 0;
@@ -4332,7 +4332,7 @@ uint build_bitmap_for_nested_joins(List<TABLE_LIST> *join_list,
         nested_join->nj_map= (nested_join_map) 1 << first_unused++;
         nested_join->nj_total= nested_join->join_list.elements;
       }
-      else if (table->sj_on_expr)
+      else if (table->sj_cond())
       {
         NESTED_JOIN *const outer_nest=
           table->embedding ? table->embedding->nested_join : NULL;
@@ -4981,7 +4981,7 @@ bool JOIN::extract_const_tables()
       */
       extract_method= extract_no_table;
     }
-    else if (tl->embedding && tl->embedding->sj_on_expr)
+    else if (tl->embedding && tl->embedding->sj_cond())
     {
       /*
         Table belongs to a semi-join.
@@ -5182,7 +5182,7 @@ bool JOIN::extract_func_dependent_tables()
           if (eq_part.is_prefix(table->key_info[key].user_defined_key_parts) &&
               !table->fulltext_searched &&                           // 1
               !tl->outer_join_nest() &&                              // 2
-              !(tl->embedding && tl->embedding->sj_on_expr) &&       // 3
+              !(tl->embedding && tl->embedding->sj_cond()) &&        // 3
               !(tab->join_cond() && tab->join_cond()->is_expensive()) &&// 4
               !(table->file->ha_table_flags() & HA_BLOCK_CONST_TABLE))  // 5
           {
@@ -5303,7 +5303,7 @@ bool JOIN::estimate_rowcount()
     TABLE_LIST *const tl= tab->table_ref;
     if (!tab->const_keys.is_clear_all() &&                        // (1)
         (!tl->embedding ||                                        // (2)
-         (tl->embedding && tl->embedding->sj_on_expr)))           // (3)
+         (tl->embedding && tl->embedding->sj_cond())))            // (3)
     {
       /*
         This call fills tab->quick() with the best QUICK access method
@@ -5323,7 +5323,7 @@ bool JOIN::estimate_rowcount()
       */
       if (records == 0 &&
           tab->table()->reginfo.impossible_range &&
-          (!(tl->embedding && tl->embedding->sj_on_expr)))
+          (!(tl->embedding && tl->embedding->sj_cond())))
       {
         /*
           Impossible WHERE condition or join condition
@@ -5388,7 +5388,7 @@ void JOIN::set_semijoin_embedding()
   {
     for (TABLE_LIST *tl= tab->table_ref; tl->embedding; tl= tl->embedding)
     {
-      if (tl->embedding->sj_on_expr)
+      if (tl->embedding->sj_cond())
       {
         tab->emb_sj_nest= tl->embedding;
         break;
@@ -6137,7 +6137,7 @@ static bool pull_out_semijoin_tables(JOIN *join)
               add("functionally_dependent", true);
             /*
               Pulling a table out of uncorrelated subquery in general makes
-              makes it correlated. See the NOTE to this function. 
+              it correlated. See the NOTE to this function. 
             */
             sj_nest->nested_join->sj_corr_tables|= tbl->map();
             sj_nest->nested_join->sj_depends_on|= tbl->map();
@@ -6432,7 +6432,7 @@ merge_key_fields(Key_field *start, Key_field *new_fields, Key_field *end,
 static uint get_semi_join_select_list_index(Item_field *item_field)
 {
   TABLE_LIST *emb_sj_nest= item_field->table_ref->embedding;
-  if (emb_sj_nest && emb_sj_nest->sj_on_expr)
+  if (emb_sj_nest && emb_sj_nest->sj_cond())
   {
     List<Item> &items= emb_sj_nest->nested_join->sj_inner_exprs;
     List_iterator<Item> it(items);

sql/sql_resolver.cc

@@ -1404,7 +1404,7 @@ SELECT_LEX::simplify_joins(THD *thd,
            the corresponding join condition is added to JC. 
 	*/ 
         if (simplify_joins(thd, &nested_join->join_list,
-                           false, in_sj || table->sj_on_expr,
+                           false, in_sj || table->sj_cond(),
                            &join_cond, changelog))
           DBUG_RETURN(true);
 
@@ -1417,7 +1417,7 @@ SELECT_LEX::simplify_joins(THD *thd,
       nested_join->used_tables= (table_map) 0;
       nested_join->not_null_tables=(table_map) 0;
       if (simplify_joins(thd, &nested_join->join_list, top,
-                         in_sj || table->sj_on_expr, cond, changelog))
+                         in_sj || table->sj_cond(), cond, changelog))
         DBUG_RETURN(true);
       used_tables= nested_join->used_tables;
       not_null_tables= nested_join->not_null_tables;  
@@ -1549,7 +1549,7 @@ SELECT_LEX::simplify_joins(THD *thd,
   while ((table= li++))
   {
     nested_join= table->nested_join;
-    if (table->sj_on_expr && !in_sj)
+    if (table->sj_cond() && !in_sj)
     {
        /*
          If this is a semi-join that is not contained within another semi-join, 
@@ -1635,9 +1635,10 @@ bool SELECT_LEX::record_join_nest_info(List<TABLE_LIST> *tables)
       This assignment is required in case pull_out_semijoin_tables()
       is not called.
     */
-    if (table->sj_on_expr)
+    if (table->sj_cond())
       table->sj_inner_tables= table->nested_join->used_tables;
-    if (table->sj_on_expr && sj_nests.push_back(table))
+
+    if (table->sj_cond() && sj_nests.push_back(table))
       DBUG_RETURN(true);
 
     if (table->join_cond())
@@ -1998,15 +1999,17 @@ SELECT_LEX::convert_subquery_to_semijoin(Item_exists_subselect *subq_pred)
       correlated tables to sj_corr_tables.
     */
     nested_join->sj_corr_tables= subq_pred->used_tables();
+
     /*
       sj_depends_on contains the set of outer tables referred in the
       subquery's WHERE clause as well as tables referred in the IN predicate's
       left-hand side.
     */
     nested_join->sj_depends_on=  subq_pred->used_tables() |
                                  in_subq_pred->left_expr->used_tables();
-    /* Put the subquery's WHERE into semi-join's condition. */
-    sj_nest->sj_on_expr= subq_select->where_cond();
+
+    // Put the subquery's WHERE into semi-join's condition.
+    Item *sj_cond= subq_select->where_cond();
 
     /*
     Create the IN-equalities and inject them into semi-join's ON condition.
@@ -2058,8 +2061,8 @@ SELECT_LEX::convert_subquery_to_semijoin(Item_exists_subselect *subq_pred)
       if (item_eq == NULL)
         DBUG_RETURN(true);      /* purecov: inspected */
 
-      sj_nest->sj_on_expr= and_items(sj_nest->sj_on_expr, item_eq);
-      if (sj_nest->sj_on_expr == NULL)
+      sj_cond= and_items(sj_cond, item_eq);
+      if (sj_cond == NULL)
         DBUG_RETURN(true);      /* purecov: inspected */
     }
     else
@@ -2076,18 +2079,21 @@ SELECT_LEX::convert_subquery_to_semijoin(Item_exists_subselect *subq_pred)
         if (item_eq == NULL)
           DBUG_RETURN(true);      /* purecov: inspected */
 
-        sj_nest->sj_on_expr= and_items(sj_nest->sj_on_expr, item_eq);
-        if (sj_nest->sj_on_expr == NULL)
+        sj_cond= and_items(sj_cond, item_eq);
+        if (sj_cond == NULL)
           DBUG_RETURN(true);      /* purecov: inspected */
       }
     }
-    /* Fix the created equality and AND */
+    // Fix the created equality and AND
 
     Opt_trace_array sj_on_trace(&thd->opt_trace,
                                 "evaluating_constant_semijoin_conditions");
-    sj_nest->sj_on_expr->top_level_item();
-    if (sj_nest->sj_on_expr->fix_fields(thd, &sj_nest->sj_on_expr))
+    sj_cond->top_level_item();
+    if (sj_cond->fix_fields(thd, &sj_cond))
       DBUG_RETURN(true);          /* purecov: inspected */
+
+    // Attach semi-join condition to semi-join nest
+    sj_nest->set_sj_cond(sj_cond);
   }
 
   // Unlink the subquery's query expression:
@@ -2114,21 +2120,21 @@ SELECT_LEX::convert_subquery_to_semijoin(Item_exists_subselect *subq_pred)
   repoint_contexts_of_join_nests(subq_select->top_join_list);
 
   // Update table map for the semi-join condition
-  sj_nest->sj_on_expr->fix_after_pullout(this, subq_select);
+  sj_nest->sj_cond()->fix_after_pullout(this, subq_select);
 
   // Update table map for semi-join nest's WHERE condition and join conditions
   fix_tables_after_pullout(this, subq_select,
                            &sj_nest->nested_join->join_list, 0);
 
   //TODO fix QT_
   DBUG_EXECUTE("where",
-               print_where(sj_nest->sj_on_expr,"SJ-EXPR", QT_ORDINARY););
+               print_where(sj_nest->sj_cond(),"SJ-COND", QT_ORDINARY););
 
   if (emb_tbl_nest)
   {
     // Inject semi-join condition into parent's join condition
     emb_tbl_nest->set_join_cond(and_items(emb_tbl_nest->join_cond(),
-                                          sj_nest->sj_on_expr));
+                                          sj_nest->sj_cond()));
     if (emb_tbl_nest->join_cond() == NULL)
       DBUG_RETURN(true);
     emb_tbl_nest->join_cond()->top_level_item();
@@ -2140,7 +2146,7 @@ SELECT_LEX::convert_subquery_to_semijoin(Item_exists_subselect *subq_pred)
   else
   {
     // Inject semi-join condition into parent's WHERE condition
-    m_where_cond= and_items(m_where_cond, sj_nest->sj_on_expr);
+    m_where_cond= and_items(m_where_cond, sj_nest->sj_cond());
     if (m_where_cond == NULL)
       DBUG_RETURN(true);
     m_where_cond->top_level_item();

sql/table.h

@@ -1704,6 +1704,16 @@ struct TABLE_LIST
   }
   Item **join_cond_optim_ref() { return &m_join_cond_optim; }
 
+  /// Get the semi-join condition for a semi-join nest, NULL otherwise
+  Item *sj_cond() const { return m_sj_cond; }
+
+  /// Set the semi-join condition for a semi-join nest
+  void set_sj_cond(Item *cond)
+  {
+    DBUG_ASSERT(m_sj_cond == NULL);
+    m_sj_cond= cond;
+  }
+
   /// Merge tables from a query block into a nested join structure
   bool merge_underlying_tables(class st_select_lex *select);
 
@@ -2017,7 +2027,7 @@ struct TABLE_LIST
   {
     if (!embedding)
       return NULL;
-    if (embedding->sj_on_expr)
+    if (embedding->sj_cond())
       return embedding->embedding;
     return embedding;
   }
@@ -2115,8 +2125,8 @@ struct TABLE_LIST
      once for all executions of a prepared statement).
   */
   Item		*m_join_cond;
+  Item          *m_sj_cond;               ///< Synthesized semijoin condition
 public:
-  Item          *sj_on_expr;            /* Synthesized semijoin condition */
   /*
     (Valid only for semi-join nests) Bitmap of tables that are within the
     semi-join (this is different from bitmap of all nest's children because