Fix best_inner_indexscan to return both the cheapest-total-cost and

cheapest-startup-cost innerjoin indexscans, and make joinpath.c consider
both of these (when different) as the inside of a nestloop join.  The
original design was based on the assumption that indexscan paths always
have negligible startup cost, and so total cost is the only important
figure of merit; an assumption that's obviously broken by bitmap
indexscans.  This oversight could lead to choosing poor plans in cases
where fast-start behavior is more important than total cost, such as
LIMIT and IN queries.  8.1-vintage brain fade exposed by an example from
Chuck D.

1 files changed, 7 insertions, 6 deletions

diff --git a/src/include/nodes/relation.h b/src/include/nodes/relation.h
index dcbcc6b3cee..ad31d536e0b 100644
--- a/src/include/nodes/relation.h
+++ b/src/include/nodes/relation.h
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2006, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.128.2.1 2007/02/16 20:57:26 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.128.2.2 2007/05/22 01:40:42 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -806,20 +806,20 @@ typedef struct RestrictInfo
  * relation includes all other relids appearing in those joinclauses.
  * The set of usable joinclauses, and thus the best inner indexscan,
  * thus varies depending on which outer relation we consider; so we have
- * to recompute the best such path for every join.	To avoid lots of
+ * to recompute the best such paths for every join.  To avoid lots of
  * redundant computation, we cache the results of such searches.  For
  * each relation we compute the set of possible otherrelids (all relids
  * appearing in joinquals that could become indexquals for this table).
  * Two outer relations whose relids have the same intersection with this
  * set will have the same set of available joinclauses and thus the same
- * best inner indexscan for the inner relation.  By taking the intersection
+ * best inner indexscans for the inner relation.  By taking the intersection
  * before scanning the cache, we avoid recomputing when considering
  * join rels that differ only by the inclusion of irrelevant other rels.
  *
  * The search key also includes a bool showing whether the join being
  * considered is an outer join.  Since we constrain the join order for
  * outer joins, I believe that this bool can only have one possible value
- * for any particular base relation; but store it anyway to avoid confusion.
+ * for any particular lookup key; but store it anyway to avoid confusion.
  */
 
 typedef struct InnerIndexscanInfo
@@ -828,8 +828,9 @@ typedef struct InnerIndexscanInfo
 	/* The lookup key: */
 	Relids		other_relids;	/* a set of relevant other relids */
 	bool		isouterjoin;	/* true if join is outer */
-	/* Best path for this lookup key: */
-	Path	   *best_innerpath; /* best inner indexscan, or NULL if none */
+	/* Best paths for this lookup key (NULL if no available indexscans): */
+	Path	   *cheapest_startup_innerpath;	/* cheapest startup cost */
+	Path	   *cheapest_total_innerpath;	/* cheapest total cost */
 } InnerIndexscanInfo;
 
 /*