First cut at making indexscan cost estimates depend on correlation

between index order and table order.

doc/src/sgml/indexcost.sgml

@@ -1,5 +1,5 @@
 <!--
-$Header: /cvsroot/pgsql/doc/src/sgml/Attic/indexcost.sgml,v 2.6 2000/12/22 21:51:57 petere Exp $
+$Header: /cvsroot/pgsql/doc/src/sgml/Attic/indexcost.sgml,v 2.7 2001/05/09 23:13:34 tgl Exp $
 -->
 
  <chapter id="indexcost">
@@ -57,7 +57,8 @@ amcostestimate (Query *root,
                 List *indexQuals,
                 Cost *indexStartupCost,
                 Cost *indexTotalCost,
-                Selectivity *indexSelectivity);
+                Selectivity *indexSelectivity,
+                double *indexCorrelation);
    </programlisting>
 
    The first four parameters are inputs:
@@ -103,7 +104,7 @@ amcostestimate (Query *root,
   </para>
 
   <para>
-   The last three parameters are pass-by-reference outputs:
+   The last four parameters are pass-by-reference outputs:
 
    <variablelist>
     <varlistentry>
@@ -132,6 +133,16 @@ amcostestimate (Query *root,
       </para>
      </listitem>
     </varlistentry>
+
+    <varlistentry>
+     <term>*indexCorrelation</term>
+     <listitem>
+      <para>
+       Set to correlation coefficient between index scan order and
+       underlying table's order
+      </para>
+     </listitem>
+    </varlistentry>
    </variablelist>
   </para>
 
@@ -172,6 +183,13 @@ amcostestimate (Query *root,
    tuples that actually pass the given qual conditions.
   </para>
 
+  <para>
+   The indexCorrelation should be set to the correlation (ranging between
+   -1.0 and 1.0) between the index order and the table order.  This is used
+   to adjust the estimate for the cost of fetching tuples from the main
+   table.
+  </para>
+
   <procedure>
    <title>Cost Estimation</title>
    <para>
@@ -224,6 +242,14 @@ amcostestimate (Query *root,
      </programlisting>
     </para>
    </step>
+
+   <step>
+    <para>
+     Estimate the index correlation.  For a simple ordered index on a single
+     field, this can be retrieved from pg_statistic.  If the correlation
+     is not known, the conservative estimate is zero (no correlation).
+    </para>
+   </step>
   </procedure>
 
   <para>
@@ -237,8 +263,8 @@ amcostestimate (Query *root,
 
    <programlisting>
 prorettype = 0
-pronargs = 7
+pronargs = 8
-proargtypes = 0 0 0 0 0 0 0
+proargtypes = 0 0 0 0 0 0 0 0
    </programlisting>
 
    We use zero ("opaque") for all the arguments since none of them have types

src/backend/optimizer/path/costsize.c

@@ -31,17 +31,18 @@
  * result by interpolating between startup_cost and total_cost.  In detail:
  *		actual_cost = startup_cost +
  *			(total_cost - startup_cost) * tuples_to_fetch / path->parent->rows;
- * Note that a relation's rows count (and, by extension, a Plan's plan_rows)
+ * Note that a base relation's rows count (and, by extension, plan_rows for
- * are set without regard to any LIMIT, so that this equation works properly.
+ * plan nodes below the LIMIT node) are set without regard to any LIMIT, so
- * (Also, these routines guarantee not to set the rows count to zero, so there
+ * that this equation works properly.  (Also, these routines guarantee not to
- * will be no zero divide.)  The LIMIT is applied as a separate Plan node.
+ * set the rows count to zero, so there will be no zero divide.)  The LIMIT is
+ * applied as a top-level plan node.
  *
  *
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/path/costsize.c,v 1.72 2001/05/09 00:35:09 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/path/costsize.c,v 1.73 2001/05/09 23:13:34 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -205,12 +206,18 @@ cost_index(Path *path, Query *root,
 {
 	Cost		startup_cost = 0;
 	Cost		run_cost = 0;
-	Cost		cpu_per_tuple;
 	Cost		indexStartupCost;
 	Cost		indexTotalCost;
 	Selectivity indexSelectivity;
+	double		indexCorrelation,
+				csquared;
+	Cost		min_IO_cost,
+				max_IO_cost;
+	Cost		cpu_per_tuple;
 	double		tuples_fetched;
 	double		pages_fetched;
+	double		T,
+				b;
 
 	/* Should only be applied to base relations */
 	Assert(IsA(baserel, RelOptInfo) &&IsA(index, IndexOptInfo));
@@ -224,38 +231,52 @@ cost_index(Path *path, Query *root,
 	 * Call index-access-method-specific code to estimate the processing
 	 * cost for scanning the index, as well as the selectivity of the
 	 * index (ie, the fraction of main-table tuples we will have to
-	 * retrieve).
+	 * retrieve) and its correlation to the main-table tuple order.
 	 */
-	OidFunctionCall7(index->amcostestimate,
+	OidFunctionCall8(index->amcostestimate,
 					 PointerGetDatum(root),
 					 PointerGetDatum(baserel),
 					 PointerGetDatum(index),
 					 PointerGetDatum(indexQuals),
 					 PointerGetDatum(&indexStartupCost),
 					 PointerGetDatum(&indexTotalCost),
-					 PointerGetDatum(&indexSelectivity));
+					 PointerGetDatum(&indexSelectivity),
+					 PointerGetDatum(&indexCorrelation));
 
 	/* all costs for touching index itself included here */
 	startup_cost += indexStartupCost;
 	run_cost += indexTotalCost - indexStartupCost;
 
-	/*
+	/*----------
 	 * Estimate number of main-table tuples and pages fetched.
 	 *
-	 * If the number of tuples is much smaller than the number of pages in
+	 * When the index ordering is uncorrelated with the table ordering,
-	 * the relation, each tuple will cost a separate nonsequential fetch.
+	 * we use an approximation proposed by Mackert and Lohman, "Index Scans
-	 * If it is comparable or larger, then probably we will be able to
+	 * Using a Finite LRU Buffer: A Validated I/O Model", ACM Transactions
-	 * avoid some fetches.	We use a growth rate of log(#tuples/#pages +
+	 * on Database Systems, Vol. 14, No. 3, September 1989, Pages 401-424.
-	 * 1) --- probably totally bogus, but intuitively it gives the right
+	 * The Mackert and Lohman approximation is that the number of pages
-	 * shape of curve at least.
+	 * fetched is
+	 *	PF =
+	 *		min(2TNs/(2T+Ns), T)			when T <= b
+	 *		2TNs/(2T+Ns)					when T > b and Ns <= 2Tb/(2T-b)
+	 *		b + (Ns - 2Tb/(2T-b))*(T-b)/T	when T > b and Ns > 2Tb/(2T-b)
+	 * where
+	 *		T = # pages in table
+	 *		N = # tuples in table
+	 *		s = selectivity = fraction of table to be scanned
+	 *		b = # buffer pages available (we include kernel space here)
 	 *
-	 * XXX if the relation has recently been "clustered" using this index,
+	 * When the index ordering is exactly correlated with the table ordering
-	 * then in fact the target tuples will be highly nonuniformly
+	 * (just after a CLUSTER, for example), the number of pages fetched should
-	 * distributed, and we will be seriously overestimating the scan cost!
+	 * be just sT.  What's more, these will be sequential fetches, not the
-	 * Currently we have no way to know whether the relation has been
+	 * random fetches that occur in the uncorrelated case.  So, depending on
-	 * clustered, nor how much it's been modified since the last
+	 * the extent of correlation, we should estimate the actual I/O cost
-	 * clustering, so we ignore this effect.  Would be nice to do better
+	 * somewhere between s * T * 1.0 and PF * random_cost.  We currently
-	 * someday.
+	 * interpolate linearly between these two endpoints based on the
+	 * correlation squared (XXX is that appropriate?).
+	 *
+	 * In any case the number of tuples fetched is Ns.
+	 *----------
 	 */
 
 	tuples_fetched = indexSelectivity * baserel->tuples;
@@ -263,24 +284,56 @@ cost_index(Path *path, Query *root,
 	if (tuples_fetched < 1.0)
 		tuples_fetched = 1.0;
 
-	if (baserel->pages > 0)
+	/* This part is the Mackert and Lohman formula */
-		pages_fetched = ceil(baserel->pages *
+
-							 log(tuples_fetched / baserel->pages + 1.0));
+	T = (baserel->pages > 1) ? (double) baserel->pages : 1.0;
+	b = (effective_cache_size > 1) ? effective_cache_size : 1.0;
+
+	if (T <= b)
+	{
+		pages_fetched =
+			(2.0 * T * tuples_fetched) / (2.0 * T + tuples_fetched);
+		if (pages_fetched > T)
+			pages_fetched = T;
+	}
 	else
-		pages_fetched = tuples_fetched;
+	{
+		double	lim;
+
+		lim = (2.0 * T * b) / (2.0 * T - b);
+		if (tuples_fetched <= lim)
+		{
+			pages_fetched =
+				(2.0 * T * tuples_fetched) / (2.0 * T + tuples_fetched);
+		}
+		else
+		{
+			pages_fetched =
+				b + (tuples_fetched - lim) * (T - b) / T;
+		}
+	}
 
 	/*
-	 * Now estimate one nonsequential access per page fetched, plus
+	 * min_IO_cost corresponds to the perfectly correlated case (csquared=1),
-	 * appropriate CPU costs per tuple.
+	 * max_IO_cost to the perfectly uncorrelated case (csquared=0).  Note
+	 * that we just charge random_page_cost per page in the uncorrelated
+	 * case, rather than using cost_nonsequential_access, since we've already
+	 * accounted for caching effects by using the Mackert model.
 	 */
-
+	min_IO_cost = ceil(indexSelectivity * T);
-	/* disk costs for main table */
+	max_IO_cost = pages_fetched * random_page_cost;
-	run_cost += pages_fetched * cost_nonsequential_access(baserel->pages);
-
-	/* CPU costs */
-	cpu_per_tuple = cpu_tuple_cost + baserel->baserestrictcost;
 
 	/*
+	 * Now interpolate based on estimated index order correlation
+	 * to get total disk I/O cost for main table accesses.
+	 */
+	csquared = indexCorrelation * indexCorrelation;
+
+	run_cost += max_IO_cost + csquared * (min_IO_cost - max_IO_cost);
+
+	/*
+	 * Estimate CPU costs per tuple.
+	 *
 	 * Normally the indexquals will be removed from the list of
 	 * restriction clauses that we have to evaluate as qpquals, so we
 	 * should subtract their costs from baserestrictcost.  For a lossy
@@ -290,6 +343,8 @@ cost_index(Path *path, Query *root,
 	 * Rather than work out exactly how much to subtract, we don't
 	 * subtract anything in that case either.
 	 */
+	cpu_per_tuple = cpu_tuple_cost + baserel->baserestrictcost;
+
 	if (!index->lossy && !is_injoin)
 		cpu_per_tuple -= cost_qual_eval(indexQuals);
 

src/backend/optimizer/util/var.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/util/var.c,v 1.31 2001/04/18 20:42:55 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/util/var.c,v 1.32 2001/05/09 23:13:35 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -28,8 +28,9 @@ typedef struct
 typedef struct
 {
 	int			varno;
+	int			varattno;
 	int			sublevels_up;
-} contain_whole_tuple_var_context;
+} contain_var_reference_context;
 
 typedef struct
 {
@@ -39,8 +40,8 @@ typedef struct
 
 static bool pull_varnos_walker(Node *node,
 				   pull_varnos_context *context);
-static bool contain_whole_tuple_var_walker(Node *node,
+static bool contain_var_reference_walker(Node *node,
-				   contain_whole_tuple_var_context *context);
+				   contain_var_reference_context *context);
 static bool contain_var_clause_walker(Node *node, void *context);
 static bool pull_var_clause_walker(Node *node,
 					   pull_var_clause_context *context);
@@ -129,10 +130,10 @@ pull_varnos_walker(Node *node, pull_varnos_context *context)
 
 
 /*
- *		contain_whole_tuple_var
+ *		contain_var_reference
  *
- *		Detect whether a parsetree contains any references to the whole
+ *		Detect whether a parsetree contains any references to a specified
- *		tuple of a given rtable entry (ie, a Var with varattno = 0).
+ *		attribute of a specified rtable entry.
  *
  * NOTE: this is used on not-yet-planned expressions.  It may therefore find
  * bare SubLinks, and if so it needs to recurse into them to look for uplevel
@@ -140,11 +141,12 @@ pull_varnos_walker(Node *node, pull_varnos_context *context)
  * SubPlan, we only need to look at the parameters passed to the subplan.
  */
 bool
-contain_whole_tuple_var(Node *node, int varno, int levelsup)
+contain_var_reference(Node *node, int varno, int varattno, int levelsup)
 {
-	contain_whole_tuple_var_context context;
+	contain_var_reference_context context;
 
 	context.varno = varno;
+	context.varattno = varattno;
 	context.sublevels_up = levelsup;
 
 	/*
@@ -154,15 +156,15 @@ contain_whole_tuple_var(Node *node, int varno, int levelsup)
 	 */
 	if (node && IsA(node, Query))
 		return query_tree_walker((Query *) node,
-								 contain_whole_tuple_var_walker,
+								 contain_var_reference_walker,
 								 (void *) &context, true);
 	else
-		return contain_whole_tuple_var_walker(node, &context);
+		return contain_var_reference_walker(node, &context);
 }
 
 static bool
-contain_whole_tuple_var_walker(Node *node,
+contain_var_reference_walker(Node *node,
-							   contain_whole_tuple_var_context *context)
+							 contain_var_reference_context *context)
 {
 	if (node == NULL)
 		return false;
@@ -171,8 +173,8 @@ contain_whole_tuple_var_walker(Node *node,
 		Var		   *var = (Var *) node;
 
 		if (var->varno == context->varno &&
-			var->varlevelsup == context->sublevels_up &&
+			var->varattno == context->varattno &&
-			var->varattno == InvalidAttrNumber)
+			var->varlevelsup == context->sublevels_up)
 			return true;
 		return false;
 	}
@@ -187,10 +189,10 @@ contain_whole_tuple_var_walker(Node *node,
 		 */
 		Expr	   *expr = (Expr *) node;
 
-		if (contain_whole_tuple_var_walker((Node *) ((SubPlan *) expr->oper)->sublink->oper,
+		if (contain_var_reference_walker((Node *) ((SubPlan *) expr->oper)->sublink->oper,
 										 context))
 			return true;
-		if (contain_whole_tuple_var_walker((Node *) expr->args,
+		if (contain_var_reference_walker((Node *) expr->args,
 										 context))
 			return true;
 		return false;
@@ -202,16 +204,29 @@ contain_whole_tuple_var_walker(Node *node,
 
 		context->sublevels_up++;
 		result = query_tree_walker((Query *) node,
-								   contain_whole_tuple_var_walker,
+								   contain_var_reference_walker,
 								   (void *) context, true);
 		context->sublevels_up--;
 		return result;
 	}
-	return expression_tree_walker(node, contain_whole_tuple_var_walker,
+	return expression_tree_walker(node, contain_var_reference_walker,
 								  (void *) context);
 }
 
 
+/*
+ *		contain_whole_tuple_var
+ *
+ *		Detect whether a parsetree contains any references to the whole
+ *		tuple of a given rtable entry (ie, a Var with varattno = 0).
+ */
+bool
+contain_whole_tuple_var(Node *node, int varno, int levelsup)
+{
+	return contain_var_reference(node, varno, InvalidAttrNumber, levelsup);
+}
+
+
 /*
  * contain_var_clause
  *	  Recursively scan a clause to discover whether it contains any Var nodes

src/backend/utils/adt/selfuncs.c

@@ -15,7 +15,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/selfuncs.c,v 1.88 2001/05/07 00:43:23 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/selfuncs.c,v 1.89 2001/05/09 23:13:35 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -39,6 +39,7 @@
 #include "optimizer/cost.h"
 #include "parser/parse_func.h"
 #include "parser/parse_oper.h"
+#include "parser/parsetree.h"
 #include "utils/builtins.h"
 #include "utils/date.h"
 #include "utils/int8.h"
@@ -818,7 +819,6 @@ eqjoinsel(PG_FUNCTION_ARGS)
 {
 #ifdef NOT_USED					/* see neqjoinsel() before removing me! */
 	Oid			opid = PG_GETARG_OID(0);
-
 #endif
 	Oid			relid1 = PG_GETARG_OID(1);
 	AttrNumber	attno1 = PG_GETARG_INT16(2);
@@ -2230,16 +2230,14 @@ string_to_datum(const char *str, Oid datatype)
  *-------------------------------------------------------------------------
  */
 
-static Datum
+static void
-genericcostestimate(PG_FUNCTION_ARGS)
+genericcostestimate(Query *root, RelOptInfo *rel,
+					IndexOptInfo *index, List *indexQuals,
+					Cost *indexStartupCost,
+					Cost *indexTotalCost,
+					Selectivity *indexSelectivity,
+					double *indexCorrelation)
 {
-	Query	   *root = (Query *) PG_GETARG_POINTER(0);
-	RelOptInfo *rel = (RelOptInfo *) PG_GETARG_POINTER(1);
-	IndexOptInfo *index = (IndexOptInfo *) PG_GETARG_POINTER(2);
-	List	   *indexQuals = (List *) PG_GETARG_POINTER(3);
-	Cost	   *indexStartupCost = (Cost *) PG_GETARG_POINTER(4);
-	Cost	   *indexTotalCost = (Cost *) PG_GETARG_POINTER(5);
-	Selectivity *indexSelectivity = (Selectivity *) PG_GETARG_POINTER(6);
 	double		numIndexTuples;
 	double		numIndexPages;
 
@@ -2275,33 +2273,134 @@ genericcostestimate(PG_FUNCTION_ARGS)
 	*indexTotalCost = numIndexPages +
 		(cpu_index_tuple_cost + cost_qual_eval(indexQuals)) * numIndexTuples;
 
-	PG_RETURN_VOID();
+	/*
+	 * Generic assumption about index correlation: there isn't any.
+	 */
+	*indexCorrelation = 0.0;
 }
 
-/*
- * For first cut, just use generic function for all index types.
- */
 
 Datum
 btcostestimate(PG_FUNCTION_ARGS)
 {
-	return genericcostestimate(fcinfo);
+	Query	   *root = (Query *) PG_GETARG_POINTER(0);
+	RelOptInfo *rel = (RelOptInfo *) PG_GETARG_POINTER(1);
+	IndexOptInfo *index = (IndexOptInfo *) PG_GETARG_POINTER(2);
+	List	   *indexQuals = (List *) PG_GETARG_POINTER(3);
+	Cost	   *indexStartupCost = (Cost *) PG_GETARG_POINTER(4);
+	Cost	   *indexTotalCost = (Cost *) PG_GETARG_POINTER(5);
+	Selectivity *indexSelectivity = (Selectivity *) PG_GETARG_POINTER(6);
+	double	   *indexCorrelation = (double *) PG_GETARG_POINTER(7);
+
+	genericcostestimate(root, rel, index, indexQuals,
+						indexStartupCost, indexTotalCost,
+						indexSelectivity, indexCorrelation);
+
+	/*
+	 * If it's a functional index, leave the default zero-correlation
+	 * estimate in place.  If not, and if we can get an estimate for
+	 * the first variable's ordering correlation C from pg_statistic,
+	 * estimate the index correlation as C / number-of-columns.
+	 * (The idea here is that multiple columns dilute the importance
+	 * of the first column's ordering, but don't negate it entirely.)
+	 */
+	if (index->indproc == InvalidOid)
+	{
+		Oid			relid;
+		HeapTuple	tuple;
+
+		relid = getrelid(lfirsti(rel->relids), root->rtable);
+		Assert(relid != InvalidOid);
+		tuple = SearchSysCache(STATRELATT,
+							   ObjectIdGetDatum(relid),
+							   Int16GetDatum(index->indexkeys[0]),
+							   0, 0);
+		if (HeapTupleIsValid(tuple))
+		{
+			Oid		typid;
+			int32	typmod;
+			float4 *numbers;
+			int		nnumbers;
+
+			get_atttypetypmod(relid, index->indexkeys[0],
+							  &typid, &typmod);
+			if (get_attstatsslot(tuple, typid, typmod,
+								 STATISTIC_KIND_CORRELATION,
+								 index->ordering[0],
+								 NULL, NULL, &numbers, &nnumbers))
+			{
+				double	varCorrelation;
+				int		nKeys;
+
+				Assert(nnumbers == 1);
+				varCorrelation = numbers[0];
+				for (nKeys = 1; index->indexkeys[nKeys] != 0; nKeys++)
+					/*skip*/;
+
+				*indexCorrelation = varCorrelation / nKeys;
+
+				free_attstatsslot(typid, NULL, 0, numbers, nnumbers);
+			}
+			ReleaseSysCache(tuple);
+		}
+	}
+
+	PG_RETURN_VOID();
 }
 
 Datum
 rtcostestimate(PG_FUNCTION_ARGS)
 {
-	return genericcostestimate(fcinfo);
+	Query	   *root = (Query *) PG_GETARG_POINTER(0);
+	RelOptInfo *rel = (RelOptInfo *) PG_GETARG_POINTER(1);
+	IndexOptInfo *index = (IndexOptInfo *) PG_GETARG_POINTER(2);
+	List	   *indexQuals = (List *) PG_GETARG_POINTER(3);
+	Cost	   *indexStartupCost = (Cost *) PG_GETARG_POINTER(4);
+	Cost	   *indexTotalCost = (Cost *) PG_GETARG_POINTER(5);
+	Selectivity *indexSelectivity = (Selectivity *) PG_GETARG_POINTER(6);
+	double	   *indexCorrelation = (double *) PG_GETARG_POINTER(7);
+
+	genericcostestimate(root, rel, index, indexQuals,
+						indexStartupCost, indexTotalCost,
+						indexSelectivity, indexCorrelation);
+
+	PG_RETURN_VOID();
 }
 
 Datum
 hashcostestimate(PG_FUNCTION_ARGS)
 {
-	return genericcostestimate(fcinfo);
+	Query	   *root = (Query *) PG_GETARG_POINTER(0);
+	RelOptInfo *rel = (RelOptInfo *) PG_GETARG_POINTER(1);
+	IndexOptInfo *index = (IndexOptInfo *) PG_GETARG_POINTER(2);
+	List	   *indexQuals = (List *) PG_GETARG_POINTER(3);
+	Cost	   *indexStartupCost = (Cost *) PG_GETARG_POINTER(4);
+	Cost	   *indexTotalCost = (Cost *) PG_GETARG_POINTER(5);
+	Selectivity *indexSelectivity = (Selectivity *) PG_GETARG_POINTER(6);
+	double	   *indexCorrelation = (double *) PG_GETARG_POINTER(7);
+
+	genericcostestimate(root, rel, index, indexQuals,
+						indexStartupCost, indexTotalCost,
+						indexSelectivity, indexCorrelation);
+
+	PG_RETURN_VOID();
 }
 
 Datum
 gistcostestimate(PG_FUNCTION_ARGS)
 {
-	return genericcostestimate(fcinfo);
+	Query	   *root = (Query *) PG_GETARG_POINTER(0);
+	RelOptInfo *rel = (RelOptInfo *) PG_GETARG_POINTER(1);
+	IndexOptInfo *index = (IndexOptInfo *) PG_GETARG_POINTER(2);
+	List	   *indexQuals = (List *) PG_GETARG_POINTER(3);
+	Cost	   *indexStartupCost = (Cost *) PG_GETARG_POINTER(4);
+	Cost	   *indexTotalCost = (Cost *) PG_GETARG_POINTER(5);
+	Selectivity *indexSelectivity = (Selectivity *) PG_GETARG_POINTER(6);
+	double	   *indexCorrelation = (double *) PG_GETARG_POINTER(7);
+
+	genericcostestimate(root, rel, index, indexQuals,
+						indexStartupCost, indexTotalCost,
+						indexSelectivity, indexCorrelation);
+
+	PG_RETURN_VOID();
 }

src/backend/utils/cache/lsyscache.c

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/lsyscache.c,v 1.54 2001/05/09 00:35:09 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/lsyscache.c,v 1.55 2001/05/09 23:13:35 tgl Exp $
  *
  * NOTES
  *	  Eventually, the index information should go through here, too.
@@ -185,6 +185,36 @@ get_atttypmod(Oid relid, AttrNumber attnum)
 		return -1;
 }
 
+/*
+ * get_atttypetypmod
+ *
+ *		A two-fer: given the relation id and the attribute number,
+ *		fetch both type OID and atttypmod in a single cache lookup.
+ *
+ * Unlike the otherwise-similar get_atttype/get_atttypmod, this routine
+ * raises an error if it can't obtain the information.
+ */
+void
+get_atttypetypmod(Oid relid, AttrNumber attnum,
+				  Oid *typid, int32 *typmod)
+{
+	HeapTuple	tp;
+	Form_pg_attribute att_tup;
+
+	tp = SearchSysCache(ATTNUM,
+						ObjectIdGetDatum(relid),
+						Int16GetDatum(attnum),
+						0, 0);
+	if (!HeapTupleIsValid(tp))
+		elog(ERROR, "cache lookup failed for relation %u attribute %d",
+			 relid, attnum);
+	att_tup = (Form_pg_attribute) GETSTRUCT(tp);
+
+	*typid = att_tup->atttypid;
+	*typmod = att_tup->atttypmod;
+	ReleaseSysCache(tp);
+}
+
 /*				---------- INDEX CACHE ----------						 */
 
 /*		watch this space...

src/include/catalog/pg_proc.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pg_proc.h,v 1.184 2001/03/22 04:00:39 momjian Exp $
+ * $Id: pg_proc.h,v 1.185 2001/05/09 23:13:35 tgl Exp $
  *
  * NOTES
  *	  The script catalog/genbki.sh reads this file and generates .bki
@@ -210,9 +210,9 @@ DESCR("not equal");
 DATA(insert OID =  89 (  version		   PGUID 12 f t f t 0 f 25 "" 100 0 0 100 pgsql_version - ));
 DESCR("PostgreSQL version string");
 
-DATA(insert OID = 1265 (  rtcostestimate   PGUID 12 f t f t 7 f 0 "0 0 0 0 0 0 0" 100 0 0 100  rtcostestimate - ));
+DATA(insert OID = 1265 (  rtcostestimate   PGUID 12 f t f t 8 f 0 "0 0 0 0 0 0 0 0" 100 0 0 100  rtcostestimate - ));
 DESCR("r-tree cost estimator");
-DATA(insert OID = 1268 (  btcostestimate   PGUID 12 f t f t 7 f 0 "0 0 0 0 0 0 0" 100 0 0 100  btcostestimate - ));
+DATA(insert OID = 1268 (  btcostestimate   PGUID 12 f t f t 8 f 0 "0 0 0 0 0 0 0 0" 100 0 0 100  btcostestimate - ));
 DESCR("btree cost estimator");
 
 /* OIDS 100 - 199 */
@@ -789,7 +789,7 @@ DESCR("convert name to char()");
 DATA(insert OID =  409 (  name			   PGUID 12 f t t t 1 f 19 "1042" 100 0 0 100	bpchar_name - ));
 DESCR("convert char() to name");
 
-DATA(insert OID =  438 (  hashcostestimate PGUID 12 f t f t 7 f 0 "0 0 0 0 0 0 0" 100 0 0 100  hashcostestimate - ));
+DATA(insert OID =  438 (  hashcostestimate PGUID 12 f t f t 8 f 0 "0 0 0 0 0 0 0 0" 100 0 0 100  hashcostestimate - ));
 DESCR("hash index cost estimator");
 
 DATA(insert OID = 440 (  hashgettuple	   PGUID 12 f t f t 2 f 23 "0 0" 100 0 0 100  hashgettuple - ));
@@ -1014,7 +1014,7 @@ DESCR("larger of two");
 DATA(insert OID = 771 (  int2smaller	   PGUID 12 f t t t 2 f 21 "21 21" 100 0 0 100	int2smaller - ));
 DESCR("smaller of two");
 
-DATA(insert OID = 772 (  gistcostestimate  PGUID 12 f t f t 7 f 0 "0 0 0 0 0 0 0" 100 0 0 100  gistcostestimate - ));
+DATA(insert OID = 772 (  gistcostestimate  PGUID 12 f t f t 8 f 0 "0 0 0 0 0 0 0 0" 100 0 0 100  gistcostestimate - ));
 DESCR("gist cost estimator");
 DATA(insert OID = 774 (  gistgettuple	   PGUID 12 f t f t 2 f 23 "0 0" 100 0 0 100  gistgettuple - ));
 DESCR("gist(internal)");

src/include/optimizer/var.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: var.h,v 1.13 2001/04/18 20:42:55 tgl Exp $
+ * $Id: var.h,v 1.14 2001/05/09 23:13:36 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -17,6 +17,8 @@
 #include "nodes/primnodes.h"
 
 extern List *pull_varnos(Node *node);
+extern bool contain_var_reference(Node *node, int varno, int varattno,
+								  int levelsup);
 extern bool contain_whole_tuple_var(Node *node, int varno, int levelsup);
 extern bool contain_var_clause(Node *node);
 extern List *pull_var_clause(Node *node, bool includeUpperVars);

src/include/utils/lsyscache.h

@@ -6,7 +6,7 @@
  * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: lsyscache.h,v 1.32 2001/05/09 00:35:09 tgl Exp $
+ * $Id: lsyscache.h,v 1.33 2001/05/09 23:13:37 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -21,6 +21,8 @@ extern AttrNumber get_attnum(Oid relid, char *attname);
 extern Oid	get_atttype(Oid relid, AttrNumber attnum);
 extern bool get_attisset(Oid relid, char *attname);
 extern int32 get_atttypmod(Oid relid, AttrNumber attnum);
+extern void get_atttypetypmod(Oid relid, AttrNumber attnum,
+							  Oid *typid, int32 *typmod);
 extern RegProcedure get_opcode(Oid opno);
 extern char *get_opname(Oid opno);
 extern bool op_mergejoinable(Oid opno, Oid ltype, Oid rtype,