First cut at full support for OUTER JOINs. There are still a few loose

ends to clean up (see my message of same date to pghackers), but mostly
it works.  INITDB REQUIRED!

src/backend/optimizer/plan/initsplan.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/initsplan.c,v 1.49 2000/08/13 02:50:07 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/initsplan.c,v 1.50 2000/09/12 21:06:54 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -26,13 +26,18 @@
 #include "optimizer/planmain.h"
 #include "optimizer/tlist.h"
 #include "optimizer/var.h"
+#include "parser/parsetree.h"
 #include "parser/parse_expr.h"
 #include "parser/parse_oper.h"
 #include "parser/parse_type.h"
 #include "utils/lsyscache.h"
 
 
-static void add_restrict_and_join_to_rel(Query *root, Node *clause);
+static void mark_baserels_for_outer_join(Query *root, Relids rels,
+										 Relids outerrels);
+static void add_restrict_and_join_to_rel(Query *root, Node *clause,
+										 bool isjoinqual,
+										 Relids outerjoinrelids);
 static void add_join_info_to_rels(Query *root, RestrictInfo *restrictinfo,
 					  Relids join_relids);
 static void add_vars_to_targetlist(Query *root, List *vars);
@@ -47,14 +52,14 @@ static void check_hashjoinable(RestrictInfo *restrictinfo);
  *****************************************************************************/
 
 /*
- * make_var_only_tlist
+ * build_base_rel_tlists
  *	  Creates rel nodes for every relation mentioned in the target list
  *	  'tlist' (if a node hasn't already been created) and adds them to
- *	  *query_relation_list*.  Creates targetlist entries for each member of
- *	  'tlist' and adds them to the tlist field of the appropriate rel node.
+ *	  root->base_rel_list.  Creates targetlist entries for each var seen
+ *	  in 'tlist' and adds them to the tlist of the appropriate rel node.
  */
 void
-make_var_only_tlist(Query *root, List *tlist)
+build_base_rel_tlists(Query *root, List *tlist)
 {
 	List	   *tlist_vars = pull_var_clause((Node *) tlist, false);
 
@@ -82,48 +87,75 @@ add_vars_to_targetlist(Query *root, List *vars)
 	}
 }
 
-/*
+/*----------
  * add_missing_rels_to_query
  *
- *	  If we have a range variable in the FROM clause that does not appear
+ *	  If we have a relation listed in the join tree that does not appear
  *	  in the target list nor qualifications, we must add it to the base
- *	  relation list so that it will be joined.	For instance, "select f.x
- *	  from foo f, foo f2" is a join of f and f2.  Note that if we have
- *	  "select foo.x from foo f", it also gets turned into a join (between
- *	  foo as foo and foo as f).
+ *	  relation list so that it can be processed.  For instance,
+ *			select f.x from foo f, foo f2
+ *	  is a join of f and f2.  Note that if we have
+ *			select foo.x from foo f
+ *	  this also gets turned into a join (between foo as foo and foo as f).
  *
  *	  To avoid putting useless entries into the per-relation targetlists,
  *	  this should only be called after all the variables in the targetlist
  *	  and quals have been processed by the routines above.
+ *
+ *	  Returns a list of all the base relations (RelOptInfo nodes) that appear
+ *	  in the join tree.  This list can be used for cross-checking in the
+ *	  reverse direction, ie, that we have a join tree entry for every
+ *	  relation used in the query.
+ *----------
  */
-void
-add_missing_rels_to_query(Query *root)
+List *
+add_missing_rels_to_query(Query *root, Node *jtnode)
 {
-	int			varno = 1;
-	List	   *l;
+	List	   *result = NIL;
 
-	foreach(l, root->rtable)
+	if (jtnode == NULL)
+		return NIL;
+	if (IsA(jtnode, List))
 	{
-		RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
+		List	   *l;
 
-		if (rte->inJoinSet)
+		foreach(l, (List *) jtnode)
 		{
-			RelOptInfo *rel = get_base_rel(root, varno);
-
-			/*
-			 * If the rel isn't otherwise referenced, give it a dummy
-			 * targetlist consisting of its own OID.
-			 */
-			if (rel->targetlist == NIL)
-			{
-				Var		   *var = makeVar(varno, ObjectIdAttributeNumber,
-										  OIDOID, -1, 0);
-
-				add_var_to_tlist(rel, var);
-			}
+			result = nconc(result,
+						   add_missing_rels_to_query(root, lfirst(l)));
 		}
-		varno++;
 	}
+	else if (IsA(jtnode, RangeTblRef))
+	{
+		int			varno = ((RangeTblRef *) jtnode)->rtindex;
+		RelOptInfo *rel = get_base_rel(root, varno);
+
+		/*
+		 * If the rel isn't otherwise referenced, give it a dummy
+		 * targetlist consisting of its own OID.
+		 */
+		if (rel->targetlist == NIL)
+		{
+			Var		   *var = makeVar(varno, ObjectIdAttributeNumber,
+									  OIDOID, -1, 0);
+
+			add_var_to_tlist(rel, var);
+		}
+
+		result = lcons(rel, NIL);
+	}
+	else if (IsA(jtnode, JoinExpr))
+	{
+		JoinExpr   *j = (JoinExpr *) jtnode;
+
+		result = add_missing_rels_to_query(root, j->larg);
+		result = nconc(result,
+					   add_missing_rels_to_query(root, j->rarg));
+	}
+	else
+		elog(ERROR, "add_missing_rels_to_query: unexpected node type %d",
+			 nodeTag(jtnode));
+	return result;
 }
 
 
@@ -134,11 +166,145 @@ add_missing_rels_to_query(Query *root)
  *****************************************************************************/
 
 
+/*
+ * add_join_quals_to_rels
+ *	  Recursively scan the join tree for JOIN/ON (and JOIN/USING) qual
+ *	  clauses, and add these to the appropriate JoinInfo lists.  Also,
+ *	  mark base RelOptInfos with outerjoinset information, which will
+ *	  be needed for proper placement of WHERE clauses during
+ *	  add_restrict_and_join_to_rels().
+ *
+ * NOTE: when dealing with inner joins, it is appropriate to let a qual clause
+ * be evaluated at the lowest level where all the variables it mentions are
+ * available.  However, we cannot do this within an outer join since the qual
+ * might eliminate matching rows and cause a NULL row to be added improperly.
+ * Therefore, rels appearing within (the nullable side of) an outer join
+ * are marked with outerjoinset = list of Relids used at the outer join node.
+ * This list will be added to the list of rels referenced by quals using
+ * such a rel, thereby forcing them up the join tree to the right level.
+ *
+ * To ease the calculation of these values, add_join_quals_to_rels() returns
+ * the list of Relids involved in its own level of join.  This is just an
+ * internal convenience; no outside callers pay attention to the result.
+ */
+Relids
+add_join_quals_to_rels(Query *root, Node *jtnode)
+{
+	Relids		result = NIL;
+
+	if (jtnode == NULL)
+		return result;
+	if (IsA(jtnode, List))
+	{
+		List	   *l;
+
+		/*
+		 * Note: we assume it's impossible to see same RT index from more
+		 * than one subtree, so nconc() is OK rather than LispUnioni().
+		 */
+		foreach(l, (List *) jtnode)
+			result = nconc(result,
+						   add_join_quals_to_rels(root, lfirst(l)));
+	}
+	else if (IsA(jtnode, RangeTblRef))
+	{
+		int			varno = ((RangeTblRef *) jtnode)->rtindex;
+
+		/* No quals to deal with, just return correct result */
+		result = lconsi(varno, NIL);
+	}
+	else if (IsA(jtnode, JoinExpr))
+	{
+		JoinExpr   *j = (JoinExpr *) jtnode;
+		Relids		leftids,
+					rightids,
+					outerjoinids;
+		List	   *qual;
+
+		/*
+		 * Order of operations here is subtle and critical.  First we recurse
+		 * to handle sub-JOINs.  Their join quals will be placed without
+		 * regard for whether this level is an outer join, which is correct.
+		 * Then, if we are an outer join, we mark baserels contained within
+		 * the nullable side(s) with our own rel list; this will restrict
+		 * placement of subsequent quals using those rels, including our own
+		 * quals, quals above us in the join tree, and WHERE quals.
+		 * Finally we place our own join quals.
+		 */
+		leftids = add_join_quals_to_rels(root, j->larg);
+		rightids = add_join_quals_to_rels(root, j->rarg);
+
+		result = nconc(listCopy(leftids), rightids);
+
+		outerjoinids = NIL;
+		switch (j->jointype)
+		{
+			case JOIN_INNER:
+				/* Inner join adds no restrictions for quals */
+				break;
+			case JOIN_LEFT:
+				mark_baserels_for_outer_join(root, rightids, result);
+				outerjoinids = result;
+				break;
+			case JOIN_FULL:
+				mark_baserels_for_outer_join(root, result, result);
+				outerjoinids = result;
+				break;
+			case JOIN_RIGHT:
+				mark_baserels_for_outer_join(root, leftids, result);
+				outerjoinids = result;
+				break;
+			case JOIN_UNION:
+				/*
+				 * This is where we fail if upper levels of planner haven't
+				 * rewritten UNION JOIN as an Append ...
+				 */
+				elog(ERROR, "UNION JOIN is not implemented yet");
+				break;
+			default:
+				elog(ERROR, "add_join_quals_to_rels: unsupported join type %d",
+					 (int) j->jointype);
+				break;
+		}
+
+		foreach(qual, (List *) j->quals)
+			add_restrict_and_join_to_rel(root, (Node *) lfirst(qual),
+										 true, outerjoinids);
+	}
+	else
+		elog(ERROR, "add_join_quals_to_rels: unexpected node type %d",
+			 nodeTag(jtnode));
+	return result;
+}
+
+/*
+ * mark_baserels_for_outer_join
+ *	  Mark all base rels listed in 'rels' as having the given outerjoinset.
+ */
+static void
+mark_baserels_for_outer_join(Query *root, Relids rels, Relids outerrels)
+{
+	List	   *relid;
+
+	foreach(relid, rels)
+	{
+		RelOptInfo *rel = get_base_rel(root, lfirsti(relid));
+
+		/*
+		 * Since we do this bottom-up, any outer-rels previously marked
+		 * should be within the new outer join set.
+		 */
+		Assert(is_subseti(rel->outerjoinset, outerrels));
+
+		rel->outerjoinset = outerrels;
+	}
+}
+
 /*
  * add_restrict_and_join_to_rels
  *	  Fill RestrictInfo and JoinInfo lists of relation entries for all
  *	  relations appearing within clauses.  Creates new relation entries if
- *	  necessary, adding them to *query_relation_list*.
+ *	  necessary, adding them to root->base_rel_list.
  *
  * 'clauses': the list of clauses in the cnfify'd query qualification.
  */
@@ -148,7 +314,8 @@ add_restrict_and_join_to_rels(Query *root, List *clauses)
 	List	   *clause;
 
 	foreach(clause, clauses)
-		add_restrict_and_join_to_rel(root, (Node *) lfirst(clause));
+		add_restrict_and_join_to_rel(root, (Node *) lfirst(clause),
+									 false, NIL);
 }
 
 /*
@@ -157,17 +324,31 @@ add_restrict_and_join_to_rels(Query *root, List *clauses)
  *	  (depending on whether the clause is a join) of each base relation
  *	  mentioned in the clause.	A RestrictInfo node is created and added to
  *	  the appropriate list for each rel.  Also, if the clause uses a
- *	  mergejoinable operator, enter the left- and right-side expressions
- *	  into the query's lists of equijoined vars.
+ *	  mergejoinable operator and is not an outer-join qual, enter the left-
+ *	  and right-side expressions into the query's lists of equijoined vars.
+ *
+ * isjoinqual is true if the clause came from JOIN/ON or JOIN/USING;
+ * we have to mark the created RestrictInfo accordingly.  If the JOIN
+ * is an OUTER join, the caller must set outerjoinrelids = all relids of join,
+ * which will override the joinrel identifiers extracted from the clause
+ * itself.  For inner join quals and WHERE clauses, set outerjoinrelids = NIL.
+ * (Passing the whole list, and not just an "isouterjoin" boolean, is simply
+ * a speed optimization: we could extract the same list from the base rels'
+ * outerjoinsets, but since add_join_quals_to_rels() already knows what we
+ * should use, might as well pass it in instead of recalculating it.)
  */
 static void
-add_restrict_and_join_to_rel(Query *root, Node *clause)
+add_restrict_and_join_to_rel(Query *root, Node *clause,
+							 bool isjoinqual,
+							 Relids outerjoinrelids)
 {
 	RestrictInfo *restrictinfo = makeNode(RestrictInfo);
 	Relids		relids;
 	List	   *vars;
+	bool		can_be_equijoin;
 
 	restrictinfo->clause = (Expr *) clause;
+	restrictinfo->isjoinqual = isjoinqual;
 	restrictinfo->subclauseindices = NIL;
 	restrictinfo->mergejoinoperator = InvalidOid;
 	restrictinfo->left_sortop = InvalidOid;
@@ -179,6 +360,44 @@ add_restrict_and_join_to_rel(Query *root, Node *clause)
 	 */
 	clause_get_relids_vars(clause, &relids, &vars);
 
+	/*
+	 * If caller has given us a join relid list, use it; otherwise, we must
+	 * scan the referenced base rels and add in any outer-join rel lists.
+	 * This prevents the clause from being applied at a lower level of joining
+	 * than any OUTER JOIN that should be evaluated before it.
+	 */
+	if (outerjoinrelids)
+	{
+		/* Safety check: parser should have enforced this to start with */
+		if (! is_subseti(relids, outerjoinrelids))
+			elog(ERROR, "JOIN qualification may not refer to other relations");
+		relids = outerjoinrelids;
+		can_be_equijoin = false;
+	}
+	else
+	{
+		Relids		newrelids = relids;
+		List	   *relid;
+
+		/* We rely on LispUnioni to be nondestructive of its input lists... */
+		can_be_equijoin = true;
+		foreach(relid, relids)
+		{
+			RelOptInfo *rel = get_base_rel(root, lfirsti(relid));
+
+			if (rel->outerjoinset)
+			{
+				newrelids = LispUnioni(newrelids, rel->outerjoinset);
+				/*
+				 * Because application of the qual will be delayed by outer
+				 * join, we mustn't assume its vars are equal everywhere.
+				 */
+				can_be_equijoin = false;
+			}
+		}
+		relids = newrelids;
+	}
+
 	if (length(relids) == 1)
 	{
 
@@ -199,7 +418,8 @@ add_restrict_and_join_to_rel(Query *root, Node *clause)
 		 * that "a.x = a.y AND a.x = b.z AND a.y = c.q" allows us to
 		 * consider z and q equal after their rels are joined.
 		 */
-		check_mergejoinable(restrictinfo);
+		if (can_be_equijoin)
+			check_mergejoinable(restrictinfo);
 	}
 	else if (relids != NIL)
 	{
@@ -209,11 +429,11 @@ add_restrict_and_join_to_rel(Query *root, Node *clause)
 		 * the relid list.	Set additional RestrictInfo fields for
 		 * joining.
 		 *
-		 * We need the merge info whether or not mergejoin is enabled (for
-		 * constructing equijoined-var lists), but we don't bother setting
-		 * hash info if hashjoin is disabled.
+		 * We don't bother setting the merge/hashjoin info if we're not
+		 * going to need it.
 		 */
-		check_mergejoinable(restrictinfo);
+		if (enable_mergejoin || can_be_equijoin)
+			check_mergejoinable(restrictinfo);
 		if (enable_hashjoin)
 			check_hashjoinable(restrictinfo);
 
@@ -223,7 +443,7 @@ add_restrict_and_join_to_rel(Query *root, Node *clause)
 		add_join_info_to_rels(root, restrictinfo, relids);
 
 		/*
-		 * Add vars used in the join clause to targetlists of member
+		 * Add vars used in the join clause to targetlists of their
 		 * relations, so that they will be emitted by the plan nodes that
 		 * scan those relations (else they won't be available at the join
 		 * node!).
@@ -241,12 +461,14 @@ add_restrict_and_join_to_rel(Query *root, Node *clause)
 	}
 
 	/*
-	 * If the clause has a mergejoinable operator, then the two sides
+	 * If the clause has a mergejoinable operator, and is not an outer-join
+	 * qualification nor bubbled up due to an outer join, then the two sides
 	 * represent equivalent PathKeyItems for path keys: any path that is
-	 * sorted by one side will also be sorted by the other (after joining,
-	 * that is).  Record the key equivalence for future use.
+	 * sorted by one side will also be sorted by the other (as soon as the
+	 * two rels are joined, that is).  Record the key equivalence for future
+	 * use.
 	 */
-	if (restrictinfo->mergejoinoperator != InvalidOid)
+	if (can_be_equijoin && restrictinfo->mergejoinoperator != InvalidOid)
 		add_equijoined_keys(root, restrictinfo);
 }
 
@@ -392,7 +614,8 @@ process_implied_equality(Query *root, Node *item1, Node *item2,
 									 BOOLOID); /* operator result type */
 	clause->args = lcons(item1, lcons(item2, NIL));
 
-	add_restrict_and_join_to_rel(root, (Node *) clause);
+	add_restrict_and_join_to_rel(root, (Node *) clause,
+								 false, NIL);
 }