Change build_index_pathkeys() so that the expressions it builds to represent

index key columns always have the type expected by the index's associated
operators, ie, we add RelabelType nodes when dealing with binary-compatible
index opclasses.  This is needed to get varchar indexes to play nicely with
the new EquivalenceClass machinery, as per recent gripe from Josh Berkus that
CVS HEAD was failing to match a varchar index column to a constant restriction
in the query.

It seems likely that this change will allow removal of a lot of ugly ad-hoc
RelabelType-stripping that the planner has traditionally done while matching
expressions to other expressions, but I'll worry about that some other day.

src/backend/optimizer/path/pathkeys.c

@@ -11,13 +11,14 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/pathkeys.c,v 1.84 2007/04/15 20:09:28 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/pathkeys.c,v 1.85 2007/05/31 16:57:34 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
 #include "postgres.h"
 
 #include "access/skey.h"
+#include "catalog/pg_type.h"
 #include "nodes/makefuncs.h"
 #include "nodes/plannodes.h"
 #include "optimizer/clauses.h"
@@ -493,6 +494,27 @@ build_index_pathkeys(PlannerInfo *root,
 			indexprs_item = lnext(indexprs_item);
 		}
 
+		/*
+		 * When dealing with binary-compatible indexes, we have to ensure that
+		 * the exposed type of the expression tree matches the declared input
+		 * type of the opclass, except when that is a polymorphic type
+		 * (compare the behavior of parse_coerce.c).  This ensures that we can
+		 * correctly match the indexkey expression to expressions we find in
+		 * the query, because arguments of operators that could match the
+		 * index will be cast likewise.
+		 */
+		if (exprType((Node *) indexkey) != index->opcintype[i] &&
+			!IsPolymorphicType(index->opcintype[i]))
+		{
+			/* Strip any existing RelabelType, and add a new one */
+			while (indexkey && IsA(indexkey, RelabelType))
+				indexkey = (Expr *) ((RelabelType *) indexkey)->arg;
+			indexkey = (Expr *) makeRelabelType(indexkey,
+												index->opcintype[i],
+												-1,
+												COERCE_DONTCARE);
+		}
+
 		/* OK, make a canonical pathkey for this sort key */
 		cpathkey = make_pathkey_from_sortinfo(root,
 											  indexkey,

src/backend/optimizer/util/plancat.c

@@ -9,7 +9,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/util/plancat.c,v 1.135 2007/05/25 17:54:25 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/util/plancat.c,v 1.136 2007/05/31 16:57:34 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -171,20 +171,23 @@ get_relation_info(PlannerInfo *root, Oid relationObjectId, bool inhparent,
 			info->ncolumns = ncolumns = index->indnatts;
 
 			/*
-			 * Need to make opfamily array large enough to put a terminating
-			 * zero at the end.
+			 * Allocate per-column info arrays.  To save a few palloc cycles
+			 * we allocate all the Oid-type arrays in one request.  Note that
+			 * the opfamily array needs an extra, terminating zero at the end.
+			 * We pre-zero the ordering info in case the index is unordered.
 			 */
 			info->indexkeys = (int *) palloc(sizeof(int) * ncolumns);
-			info->opfamily = (Oid *) palloc0(sizeof(Oid) * (ncolumns + 1));
-			/* initialize these to zeroes in case index is unordered */
-			info->fwdsortop = (Oid *) palloc0(sizeof(Oid) * ncolumns);
-			info->revsortop = (Oid *) palloc0(sizeof(Oid) * ncolumns);
+			info->opfamily = (Oid *) palloc0(sizeof(Oid) * (4 * ncolumns + 1));
+			info->opcintype = info->opfamily + (ncolumns + 1);
+			info->fwdsortop = info->opcintype + ncolumns;
+			info->revsortop = info->fwdsortop + ncolumns;
 			info->nulls_first = (bool *) palloc0(sizeof(bool) * ncolumns);
 
 			for (i = 0; i < ncolumns; i++)
 			{
-				info->opfamily[i] = indexRelation->rd_opfamily[i];
 				info->indexkeys[i] = index->indkey.values[i];
+				info->opfamily[i] = indexRelation->rd_opfamily[i];
+				info->opcintype[i] = indexRelation->rd_opcintype[i];
 			}
 
 			info->relam = indexRelation->rd_rel->relam;

src/include/nodes/relation.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.143 2007/05/22 23:23:57 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.144 2007/05/31 16:57:34 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -370,10 +370,11 @@ typedef struct RelOptInfo
  *		and indexes, but that created confusion without actually doing anything
  *		useful.  So now we have a separate IndexOptInfo struct for indexes.
  *
- *		opfamily[], indexkeys[], fwdsortop[], revsortop[], and nulls_first[]
- *		each have ncolumns entries.  Note: for historical reasons, the
- *		opfamily array has an extra entry that is always zero.  Some code
- *		scans until it sees a zero entry, rather than looking at ncolumns.
+ *		opfamily[], indexkeys[], opcintype[], fwdsortop[], revsortop[],
+ *		and nulls_first[] each have ncolumns entries.
+ *		Note: for historical reasons, the opfamily array has an extra entry
+ *		that is always zero.  Some code scans until it sees a zero entry,
+ *		rather than looking at ncolumns.
  *
  *		Zeroes in the indexkeys[] array indicate index columns that are
  *		expressions; there is one element in indexprs for each such column.
@@ -402,6 +403,7 @@ typedef struct IndexOptInfo
 	int			ncolumns;		/* number of columns in index */
 	Oid		   *opfamily;		/* OIDs of operator families for columns */
 	int		   *indexkeys;		/* column numbers of index's keys, or 0 */
+	Oid		   *opcintype;		/* OIDs of opclass declared input data types */
 	Oid		   *fwdsortop;		/* OIDs of sort operators for each column */
 	Oid		   *revsortop;		/* OIDs of sort operators for backward scan */
 	bool	   *nulls_first;	/* do NULLs come first in the sort order? */