MDEV-37933: Rewrite [NATURAL] FULL OUTER to LEFT, RIGHT, or INNER JOIN

Rewrite FULL OUTER JOIN queries as either LEFT, RIGHT, or INNER JOIN
by checking if and how the WHERE clause rejects nulls.

For example, the following two queries are equivalent because the
WHERE condition rejects nulls from the left table and allows matches
in the right table (or NULL from the right table) for the remaining
rows:

  SELECT * FROM t1 FULL JOIN t2 ON t1.v = t2.v WHERE t1.v IS NOT NULL;
  SELECT * FROM t1 LEFT JOIN t2 ON t1.v = t2.v;

Author: DaveGosselin-MariaDB

mysql-test/main/join.result

@@ -2073,34 +2073,34 @@ explain extended select * from t1 natural full outer join t2;
 select * from t1 natural full join t2;
 explain extended select * from t1 natural full join t2;
 
-create view v1 as select * from t1 full join t2 on t1.a = t2.a;
+create view v1 as select t1.a as t1a, t2.a as t2a from t1 full join t2 on t1.a = t2.a;
 --error ER_NOT_SUPPORTED_YET
 select * from v1;
 explain extended select * from v1;
 drop view v1;
 
-create view v1 as select * from t1 full outer join t2 on t1.a = t2.a;
+create view v1 as select t1.a as t1a, t2.a as t2a from t1 full outer join t2 on t1.a = t2.a;
 --error ER_NOT_SUPPORTED_YET
 select * from v1;
 explain extended select * from v1;
 drop view v1;
 
-create view v1 as select * from t1 natural full join t2;
+create view v1 as select t1.a as t1a, t2.a as t2a from t1 natural full join t2;
 --error ER_NOT_SUPPORTED_YET
 select * from v1;
 explain extended select * from v1;
 drop view v1;
 
-create view v1 as select * from t1 natural full outer join t2;
+create view v1 as select t1.a as t1a, t2.a as t2a from t1 natural full outer join t2;
 --error ER_NOT_SUPPORTED_YET
 select * from v1;
 explain extended select * from v1;
 drop view v1;
 
---error ER_BAD_FIELD_ERROR
+--error ER_NOT_SUPPORTED_YET
 select * from (select t1.a from t1 full join t2 on t1.a = t2.a union select * from t1) dt;
 
---error ER_BAD_FIELD_ERROR
+--error ER_NOT_SUPPORTED_YET
 select * from (select t1.a from t1 full outer join t2 on t1.a = t2.a union select * from t1) dt;
 
 --error ER_NOT_SUPPORTED_YET
@@ -2127,7 +2127,104 @@ select * from t1, t2 natural full join t3;
 --error ER_NO_SUCH_TABLE
 select * from t1, t2 natural full outer join t3;
 
+--error ER_FULL_JOIN_BASE_TABLES_ONLY
+select * from (select * from t1) dt natural full join (select * from t2) du;
+
+--error ER_FULL_JOIN_BASE_TABLES_ONLY
+select * from (select * from t1) dt natural full join t2;
+
+--error ER_FULL_JOIN_BASE_TABLES_ONLY
+select * from t1 natural full join (select * from t2) du;
+
 drop table t1, t2;
+
+--echo # Exercise FULL JOIN rewrites to LEFT, RIGHT, and INNER JOIN.
+create table x (pk int auto_increment, x int, y int, primary key (pk));
+create table xsq (pk int auto_increment, x int, y int, primary key (pk));
+insert into x (x, y) values (-5,-5),(-4,-4),(-3,-3),(-2,-2),(-1,-1),(0,0),(1,1),(2,2),(3,3),(4,4),(5,5);
+insert into xsq (x, y) values (-5,25),(-4,16),(-3,9),(-2,4),(-1,1),(0,0),(1,1),(2,4),(3,9),(4,16),(5,25);
+
+--echo # FULL to RIGHT JOIN, these two queries should be equal:
+select * from x  full join xsq on x.y = xsq.y where xsq.pk is not null;
+select * from x right join xsq on x.y = xsq.y;
+
+--echo # FULL to RIGHT JOIN, these two queries should be equal:
+select * from x  full join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1 where xsq.pk is not null;
+select * from x right join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1;
+
+--echo # FULL to INNER JOIN, these two queries should be equal:
+select * from x  full join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1 where x.pk is not null and xsq.pk is not null;
+select * from x inner join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1;
+
+--echo # FULL to LEFT JOIN, these two queries should be equal:
+select * from x full join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1 where x.pk is not null;
+select * from x left join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1;
+
+--echo # FULL NATURAL to INNER JOIN, these two queries should be equal:
+select * from x natural full join xsq where x.pk is not null and xsq.pk is not null;
+select * from x inner join xsq on x.x = xsq.x and x.y = xsq.y;
+
+--echo # FULL NATURAL to LEFT JOIN, these two queries should be equal:
+select * from x natural full join xsq where x.pk is not null;
+select * from x left join xsq on xsq.pk = x.pk and xsq.x = x.x and xsq.y = x.y;
+
+--echo # FULL NATURAL to RIGHT JOIN
+select * from x natural full join xsq where xsq.pk is not null;
+select * from x right join xsq on x.pk = xsq.pk and x.x = xsq.x and x.y = xsq.y;
+
+--echo # These two will fail because it cannot be rewritten to a LEFT, RIGHT, nor INNER JOIN.
+--error ER_NOT_SUPPORTED_YET
+select * from x full join xsq on x.x >= 0 and x.x <= 1 and xsq.x >= 0 and xsq.x <= 1;
+--error ER_NOT_SUPPORTED_YET
+select * from x natural full join xsq;
+
+drop table x, xsq;
+
+--echo # Nested JOINs
+create table one (v int);
+insert into one (v) values (1);
+create table two (v int);
+insert into two (v) values (2);
+create table three (v int);
+insert into three (v) values (3);
+
+--echo # (FULL)FULL to (INNER)INNER JOIN
+select * from one full join two on one.v = two.v full join three on two.v = three.v where one.v is not null and two.v is not null and three.v is not null;
+select * from one inner join two on one.v = two.v inner join three on two.v = three.v;
+
+--echo # (FULL)FULL to (RIGHT)LEFT JOIN
+select * from one full join two on one.v = two.v full join three on one.v = three.v where two.v is not null;
+select * from one right join two on one.v = two.v left join three on one.v = three.v;
+
+--echo # (FULL)FULL to (LEFT)LEFT JOIN
+select * from one full join two on one.v = two.v full join three on one.v = three.v where one.v is not null;
+select * from one left join two on two.v = one.v left join three on three.v = one.v;
+
+--echo # (FULL)LEFT to (LEFT)LEFT JOIN
+select * from one full join two on one.v = two.v left join three on two.v = three.v where one.v is not null;
+select * from one left join two on one.v = two.v left join three on two.v = three.v;
+
+--echo # (FULL)LEFT to (RIGHT)LEFT JOIN
+select * from one full join two on one.v = two.v left join three on two.v = three.v where two.v is not null;
+select * from one right join two on one.v = two.v left join three on three.v = two.v;
+
+--echo # (LEFT)FULL to (LEFT)RIGHT JOIN
+select * from one left join two on one.v = two.v full join three on two.v = three.v where three.v is not null;
+select * from one left join two on one.v = two.v right join three on two.v = three.v;
+
+--echo # (LEFT)FULL to (LEFT)LEFT JOIN
+insert into one (v) values (2),(3);
+insert into two (v) values (1);
+truncate three;
+insert into three (v) values (1);
+select * from one;
+select * from two;
+select * from three;
+select * from one left join two on one.v = two.v full join three on two.v = three.v where three.v = 1;
+select * from three left join one on one.v = 1 left join two on two.v = 1;
+
+drop table one, two, three;
+
 # TODO fix PS protocol before end of FULL OUTER JOIN development
 --enable_ps_protocol
 --echo # End of 12.3 tests

mysql-test/main/join.test

@@ -12362,3 +12362,5 @@ ER_WARN_CONFLICTING_COMPOUND_INDEX_HINT_FOR_KEY
         eng "Hint %s is ignored as conflicting/duplicated (an index hint of the same type or opposite kind has already been specified for the key)"
 ER_WARN_NO_IMPLICIT_QB_NAMES_IN_VIEW
 	eng "Implicit query block names are ignored for hints specified within VIEWs"
+ER_FULL_JOIN_BASE_TABLES_ONLY
+        eng "FULL JOIN supported for base tables only, %s is not a base table"

sql/share/errmsg-utf8.txt

@@ -7998,6 +7998,23 @@ store_top_level_join_columns(THD *thd, TABLE_LIST *table_ref,
 }
 
 
+/*
+  If the tables involved participate in a FULL OUTER JOIN then don't
+  update their name resolution contexts at this point, or it will
+  break simplify_joins.
+ */
+static bool is_full_outer_join(Name_resolution_context *context,
+                               TABLE_LIST *right_neighbor)
+{
+  const uint ctx_outer_join= context->first_name_resolution_table->outer_join;
+  const uint rtn_outer_join=
+    right_neighbor->first_leaf_for_name_resolution()->outer_join;
+
+  return (ctx_outer_join & JOIN_TYPE_FULL) ||
+         (rtn_outer_join & JOIN_TYPE_FULL);
+}
+
+
 /*
   Compute and store the row types of the top-most NATURAL/USING joins
   in a FROM clause.
@@ -8085,8 +8102,9 @@ static bool setup_natural_join_row_types(THD *thd,
     FROM clause.
   */
   DBUG_ASSERT(right_neighbor);
-  context->first_name_resolution_table=
-    right_neighbor->first_leaf_for_name_resolution();
+  if (!is_full_outer_join(context, right_neighbor))
+    context->first_name_resolution_table=
+      right_neighbor->first_leaf_for_name_resolution();
   /*
     This is only to ensure that first_name_resolution_table doesn't
     change on re-execution

sql/sql_base.cc

@@ -597,6 +597,34 @@ template <class T> class List_iterator :public base_list_iterator
   inline void remove()      { base_list_iterator::remove(); }
   inline void after(T *a)   { base_list_iterator::after(a); }
   inline T** ref(void)	    { return (T**) base_list_iterator::ref(); }
+
+  /*
+    Swap the current element with the next one in the list.
+
+    If this iterator points to no element or to the last element, then this
+    method does nothing and returns nullptr.
+
+    If this iter points to B in the following list
+      A, B, C, D, ...
+    then after this method returns, the list will be
+      A, C, B, D, ...
+    and this method returns C.  This iter will point to the same location
+    in the list after this method returns as it did before, but the element at
+    that location will be C instead of B.
+
+    Other iterators pointing to the same list remain valid and will see the
+    updated list order.
+  */
+  T* swap_next()
+  {
+    if (!ref() || !*ref() || !peek())
+      return nullptr;
+    T* cur= *ref();
+    List_iterator<T> itr= *this;
+    T* nxt= itr++;
+    replace(nxt);
+    return itr.replace(cur);
+  }
 };
 
 

sql/sql_list.h

@@ -8672,12 +8672,10 @@ bool st_select_lex::add_cross_joined_table(TABLE_LIST *left_op,
 
 TABLE_LIST *st_select_lex::convert_right_join()
 {
-  TABLE_LIST *tab2= join_list->pop();
-  TABLE_LIST *tab1= join_list->pop();
   DBUG_ENTER("convert_right_join");
-
-  join_list->push_front(tab2, parent_lex->thd->mem_root);
-  join_list->push_front(tab1, parent_lex->thd->mem_root);
+  List_iterator<TABLE_LIST> li(*join_list);
+  li++; // points iterator at first element and returns it
+  TABLE_LIST* tab1= li.swap_next();
   tab1->outer_join|= JOIN_TYPE_RIGHT;
 
   DBUG_RETURN(tab1);