Update comments about pathkeys.

src/backend/optimizer/path/pathkeys.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/path/pathkeys.c,v 1.12 1999/07/16 04:59:15 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/path/pathkeys.c,v 1.13 1999/08/13 01:17:16 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -33,18 +33,24 @@ static List *new_join_pathkey(List *pathkeys, List *join_rel_tlist,
  *	order of the result generated by the Path.
  *
  *	In single/base relation RelOptInfo's, the Path's represent various ways
- *	of generating the relation and the resulting ordering of the tuples.
+ *	of scanning the relation and the resulting ordering of the tuples.
  *	Sequential scan Paths have NIL pathkeys, indicating no known ordering.
- *	Index scans have Path.pathkeys that represent the chosen index.
+ *	Index scans have Path.pathkeys that represent the chosen index's ordering,
- *	A single-key index pathkeys would be { {tab1_indexkey1} }.	For a
+ *  if any.  A single-key index would create a pathkey with a single sublist,
- *	multi-key index pathkeys would be { {tab1_indexkey1}, {tab1_indexkey2} },
+ *	e.g. ( (tab1_indexkey1) ).  A multi-key index generates a sublist per key,
- *	indicating major sort by indexkey1 and minor sort by indexkey2.
+ *	e.g. ( (tab1_indexkey1) (tab1_indexkey2) ) which shows major sort by
+ *	indexkey1 and minor sort by indexkey2.
+ *
+ *	Note that a multi-pass indexscan (OR clause scan) has NIL pathkeys since
+ *	we can say nothing about the overall order of its result.  Also, an index
+ *	scan on an unordered type of index generates no useful pathkeys.  However,
+ *	we can always create a pathkey by doing an explicit sort.
  *
  *	Multi-relation RelOptInfo Path's are more complicated.  Mergejoins are
  *	only performed with equijoins ("=").  Because of this, the multi-relation
  *	path actually has more than one primary Var key.  For example, a
- *	mergejoin Path of "tab1.col1 = tab2.col1" would generate a pathkeys of
+ *	mergejoin Path of "tab1.col1 = tab2.col1" would generate pathkeys of
- *	{ {tab1.col1, tab2.col1} }, indicating that the major sort order of the
+ *	( (tab1.col1 tab2.col1) ), indicating that the major sort order of the
  *	Path can be taken to be *either* tab1.col1 or tab2.col1.
  *	They are equal, so they are both primary sort keys.  This allows future
  *	joins to use either Var as a pre-sorted key to prevent upper Mergejoins
@@ -53,21 +59,30 @@ static List *new_join_pathkey(List *pathkeys, List *join_rel_tlist,
  *	Note that while the order of the top list is meaningful (primary vs.
  *	secondary sort key), the order of each sublist is arbitrary.
  *
- *	For multi-key sorts, if the outer is sorted by a multi-key index, the
+ *	We can actually keep all of the keys of the outer path of a merge or
- *	multi-key index remains after the join.  If the inner has a multi-key
+ *	nestloop join, since the ordering of the outer path will be reflected
- *	sort, only the primary key of the inner is added to the result.
+ *	in the result.  We add to each pathkey sublist any inner vars that are
- *	Mergejoins only join on the primary key.  Currently, non-primary keys
+ *	equijoined to any of the outer vars in the sublist.  In the nestloop
- *	in the pathkeys List are of limited value.
+ *	case we have to be careful to consider only equijoin operators; the
+ *	nestloop's join clauses might include non-equijoin operators.
+ *	(Currently, we do this by considering only mergejoinable operators while
+ *	making the pathkeys, since we have no separate marking for operators that
+ *	are equijoins but aren't mergejoinable.)
  *
  *	Although Hashjoins also work only with equijoin operators, it is *not*
  *	safe to consider the output of a Hashjoin to be sorted in any particular
  *	order --- not even the outer path's order.  This is true because the
- *	executor might have to split the join into multiple batches.
+ *	executor might have to split the join into multiple batches.  Therefore
+ *	a Hashjoin is always given NIL pathkeys.
  *
- *	NestJoin does not perform sorting, and allows non-equijoins, so it does
+ *	Notice that pathkeys only say *what* is being ordered, and not *how*
- *	not allow useful pathkeys.	(But couldn't we use the outer path's order?)
+ *	it is ordered.  The actual sort ordering is indicated by a separate
+ *	data structure, the PathOrder.  The PathOrder provides a sort operator
+ *	OID for each of the sublists of the path key.  This is fairly bogus,
+ *	since in cross-datatype cases we really want to keep track of more than
+ *	one sort operator...
  *
- *	-- bjm
+ *	-- bjm & tgl
  *--------------------
  */
 
@@ -328,17 +343,32 @@ make_pathkeys_from_joinkeys(List *joinkeys,
 
 /*
  * new_join_pathkeys
- *	  Find the path keys for a join relation by finding all vars in the list
+ *	  Build the path keys for a join relation constructed by mergejoin or
- *	  of join clauses 'joinclauses' such that:
+ *	  nestloop join.  These keys should include all the path key vars of the
- *		(1) the var corresponding to the outer join relation is a
+ *	  outer path (since the join will retain the ordering of the outer path)
- *			key on the outer path
+ *	  plus any vars of the inner path that are mergejoined to the outer vars.
- *		(2) the var appears in the target list of the join relation
+ *
- *	  In other words, add to each outer path key the inner path keys that
+ *	  Per the discussion at the top of this file, mergejoined inner vars
- *	  are required for qualification.
+ *	  can be considered path keys of the result, just the same as the outer
+ *	  vars they were joined with.
+ *
+ *	  We can also use inner path vars as pathkeys of a nestloop join, but we
+ *	  must be careful that we only consider equijoin clauses and not general
+ *	  join clauses.  For example, "t1.a < t2.b" might be a join clause of a
+ *	  nestloop, but it doesn't result in b acquiring the ordering of a!
+ *	  joinpath.c handles that problem by only passing this routine clauses
+ *	  that are marked mergejoinable, even if a nestloop join is being built.
+ *	  Therefore we only have 't1.a = t2.b' style clauses, and can expect that
+ *	  the inner var will acquire the outer's ordering no matter which join
+ *	  method is actually used.
+ *
+ * All vars in the result are copied from the join relation's tlist, not from
+ * the given pathkeys or the join clauses.  (Is that necessary?  I suspect
+ * not --- tgl)
  *
  * 'outer_pathkeys' is the list of the outer path's path keys
  * 'join_rel_tlist' is the target list of the join relation
- * 'joinclauses' is the list of restricting join clauses
+ * 'joinclauses' is the list of mergejoinable join clauses
  *
  * Returns the list of new path keys.
  *
@@ -358,8 +388,13 @@ new_join_pathkeys(List *outer_pathkeys,
 
 		new_pathkey = new_join_pathkey(outer_pathkey, join_rel_tlist,
 									   joinclauses);
-		if (new_pathkey != NIL)
+		/* if we can find no sortable vars for the n'th sort key,
-			final_pathkeys = lappend(final_pathkeys, new_pathkey);
+		 * then we're done generating pathkeys; can't expect to order
+		 * subsequent vars.  Not clear that this can really happen.
+		 */
+		if (new_pathkey == NIL)
+			break;
+		final_pathkeys = lappend(final_pathkeys, new_pathkey);
 	}
 	return final_pathkeys;
 }
@@ -372,7 +407,7 @@ new_join_pathkeys(List *outer_pathkeys,
  *	  at the top of this file).
  *
  *	  Note that each returned pathkey is the var node found in
- *	  'join_rel_tlist' rather than the joinclause var node.
+ *	  'join_rel_tlist' rather than the input pathkey or joinclause var node.
  *	  (Is this important?)	Also, we return a fully copied list
  *	  that does not share any subnodes with existing data structures.
  *	  (Is that important, either?)