Allow row comparisons to be used as indexscan qualifications.

This completes the project to upgrade our handling of row comparisons.

11 files changed, 996 insertions, 142 deletions

diff --git a/src/backend/access/nbtree/nbtsearch.c b/src/backend/access/nbtree/nbtsearch.c
index 8805d32de9b..618b643d7e0 100644
--- a/src/backend/access/nbtree/nbtsearch.c
+++ b/src/backend/access/nbtree/nbtsearch.c
@@ -8,7 +8,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.101 2006/01/23 22:31:40 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.102 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -551,6 +551,10 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	 * one we use --- by definition, they are either redundant or
 	 * contradictory.
 	 *
+	 * In this loop, row-comparison keys are treated the same as keys on their
+	 * first (leftmost) columns.  We'll add on lower-order columns of the row
+	 * comparison below, if possible.
+	 *
 	 * The selected scan keys (at most one per index column) are remembered by
 	 * storing their addresses into the local startKeys[] array.
 	 *----------
@@ -657,44 +661,91 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	{
 		ScanKey		cur = startKeys[i];
 
-		/*
-		 * _bt_preprocess_keys disallows it, but it's place to add some code
-		 * later
-		 */
-		if (cur->sk_flags & SK_ISNULL)
-			elog(ERROR, "btree doesn't support is(not)null, yet");
+		Assert(cur->sk_attno == i+1);
 
-		/*
-		 * If scankey operator is of default subtype, we can use the cached
-		 * comparison procedure; otherwise gotta look it up in the catalogs.
-		 */
-		if (cur->sk_subtype == InvalidOid)
+		if (cur->sk_flags & SK_ROW_HEADER)
 		{
-			FmgrInfo   *procinfo;
-
-			procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
-			ScanKeyEntryInitializeWithInfo(scankeys + i,
-										   cur->sk_flags,
-										   i + 1,
-										   InvalidStrategy,
-										   InvalidOid,
-										   procinfo,
-										   cur->sk_argument);
+			/*
+			 * Row comparison header: look to the first row member instead.
+			 *
+			 * The member scankeys are already in insertion format (ie, they
+			 * have sk_func = 3-way-comparison function), but we have to
+			 * watch out for nulls, which _bt_preprocess_keys didn't check.
+			 * A null in the first row member makes the condition unmatchable,
+			 * just like qual_ok = false.
+			 */
+			cur = (ScanKey) DatumGetPointer(cur->sk_argument);
+			Assert(cur->sk_flags & SK_ROW_MEMBER);
+			if (cur->sk_flags & SK_ISNULL)
+				return false;
+			memcpy(scankeys + i, cur, sizeof(ScanKeyData));
+			/*
+			 * If the row comparison is the last positioning key we accepted,
+			 * try to add additional keys from the lower-order row members.
+			 * (If we accepted independent conditions on additional index
+			 * columns, we use those instead --- doesn't seem worth trying to
+			 * determine which is more restrictive.)  Note that this is OK
+			 * even if the row comparison is of ">" or "<" type, because the
+			 * condition applied to all but the last row member is effectively
+			 * ">=" or "<=", and so the extra keys don't break the positioning
+			 * scheme.
+			 */
+			if (i == keysCount - 1)
+			{
+				while (!(cur->sk_flags & SK_ROW_END))
+				{
+					cur++;
+					Assert(cur->sk_flags & SK_ROW_MEMBER);
+					if (cur->sk_attno != keysCount + 1)
+						break;	/* out-of-sequence, can't use it */
+					if (cur->sk_flags & SK_ISNULL)
+						break;	/* can't use null keys */
+					Assert(keysCount < INDEX_MAX_KEYS);
+					memcpy(scankeys + keysCount, cur, sizeof(ScanKeyData));
+					keysCount++;
+				}
+				break;			/* done with outer loop */
+			}
 		}
 		else
 		{
-			RegProcedure cmp_proc;
-
-			cmp_proc = get_opclass_proc(rel->rd_indclass->values[i],
-										cur->sk_subtype,
-										BTORDER_PROC);
-			ScanKeyEntryInitialize(scankeys + i,
-								   cur->sk_flags,
-								   i + 1,
-								   InvalidStrategy,
-								   cur->sk_subtype,
-								   cmp_proc,
-								   cur->sk_argument);
+			/*
+			 * Ordinary comparison key.  Transform the search-style scan key
+			 * to an insertion scan key by replacing the sk_func with the
+			 * appropriate btree comparison function.
+			 *
+			 * If scankey operator is of default subtype, we can use the
+			 * cached comparison function; otherwise gotta look it up in the
+			 * catalogs.
+			 */
+			if (cur->sk_subtype == InvalidOid)
+			{
+				FmgrInfo   *procinfo;
+
+				procinfo = index_getprocinfo(rel, cur->sk_attno, BTORDER_PROC);
+				ScanKeyEntryInitializeWithInfo(scankeys + i,
+											   cur->sk_flags,
+											   cur->sk_attno,
+											   InvalidStrategy,
+											   InvalidOid,
+											   procinfo,
+											   cur->sk_argument);
+			}
+			else
+			{
+				RegProcedure cmp_proc;
+
+				cmp_proc = get_opclass_proc(rel->rd_indclass->values[i],
+											cur->sk_subtype,
+											BTORDER_PROC);
+				ScanKeyEntryInitialize(scankeys + i,
+									   cur->sk_flags,
+									   cur->sk_attno,
+									   InvalidStrategy,
+									   cur->sk_subtype,
+									   cmp_proc,
+									   cur->sk_argument);
+			}
 		}
 	}
 
diff --git a/src/backend/access/nbtree/nbtutils.c b/src/backend/access/nbtree/nbtutils.c
index b715383871a..90f2f64c81f 100644
--- a/src/backend/access/nbtree/nbtutils.c
+++ b/src/backend/access/nbtree/nbtutils.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtutils.c,v 1.69 2006/01/23 22:31:40 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtutils.c,v 1.70 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -21,6 +21,12 @@
 #include "executor/execdebug.h"
 
 
+static void _bt_mark_scankey_required(ScanKey skey);
+static bool _bt_check_rowcompare(ScanKey skey,
+								 IndexTuple tuple, TupleDesc tupdesc,
+								 ScanDirection dir, bool *continuescan);
+
+
 /*
  * _bt_mkscankey
  *		Build an insertion scan key that contains comparison data from itup
@@ -218,6 +224,17 @@ _bt_formitem(IndexTuple itup)
  * equality quals for all columns.	In this case there can be at most one
  * (visible) matching tuple.  index_getnext uses this to avoid uselessly
  * continuing the scan after finding one match.
+ *
+ * Row comparison keys are treated the same as comparisons to nondefault
+ * datatypes: we just transfer them into the preprocessed array without any
+ * editorialization.  We can treat them the same as an ordinary inequality
+ * comparison on the row's first index column, for the purposes of the logic
+ * about required keys.
+ *
+ * Note: the reason we have to copy the preprocessed scan keys into private
+ * storage is that we are modifying the array based on comparisons of the
+ * key argument values, which could change on a rescan.  Therefore we can't
+ * overwrite the caller's data structure.
  *----------
  */
 void
@@ -273,26 +290,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 		memcpy(outkeys, inkeys, sizeof(ScanKeyData));
 		so->numberOfKeys = 1;
 		/* We can mark the qual as required if it's for first index col */
-		if (outkeys->sk_attno == 1)
-		{
-			switch (outkeys->sk_strategy)
-			{
-				case BTLessStrategyNumber:
-				case BTLessEqualStrategyNumber:
-					outkeys->sk_flags |= SK_BT_REQFWD;
-					break;
-				case BTEqualStrategyNumber:
-					outkeys->sk_flags |= (SK_BT_REQFWD | SK_BT_REQBKWD);
-					break;
-				case BTGreaterEqualStrategyNumber:
-				case BTGreaterStrategyNumber:
-					outkeys->sk_flags |= SK_BT_REQBKWD;
-					break;
-				default:
-					elog(ERROR, "unrecognized StrategyNumber: %d",
-						 (int) outkeys->sk_strategy);
-			}
-		}
+		if (cur->sk_attno == 1)
+			_bt_mark_scankey_required(outkeys);
 		return;
 	}
 
@@ -325,6 +324,7 @@ _bt_preprocess_keys(IndexScanDesc scan)
 		if (i < numberOfKeys)
 		{
 			/* See comments above: any NULL implies cannot match qual */
+			/* Note: we assume SK_ISNULL is never set in a row header key */
 			if (cur->sk_flags & SK_ISNULL)
 			{
 				so->qual_ok = false;
@@ -432,26 +432,7 @@ _bt_preprocess_keys(IndexScanDesc scan)
 
 					memcpy(outkey, xform[j], sizeof(ScanKeyData));
 					if (priorNumberOfEqualCols == attno - 1)
-					{
-						switch (outkey->sk_strategy)
-						{
-							case BTLessStrategyNumber:
-							case BTLessEqualStrategyNumber:
-								outkey->sk_flags |= SK_BT_REQFWD;
-								break;
-							case BTEqualStrategyNumber:
-								outkey->sk_flags |= (SK_BT_REQFWD |
-													 SK_BT_REQBKWD);
-								break;
-							case BTGreaterEqualStrategyNumber:
-							case BTGreaterStrategyNumber:
-								outkey->sk_flags |= SK_BT_REQBKWD;
-								break;
-							default:
-								elog(ERROR, "unrecognized StrategyNumber: %d",
-									 (int) outkey->sk_strategy);
-						}
-					}
+						_bt_mark_scankey_required(outkey);
 				}
 			}
 
@@ -470,11 +451,14 @@ _bt_preprocess_keys(IndexScanDesc scan)
 		/* check strategy this key's operator corresponds to */
 		j = cur->sk_strategy - 1;
 
-		/* if wrong RHS data type, punt */
-		if (cur->sk_subtype != InvalidOid)
+		/* if row comparison or wrong RHS data type, punt */
+		if ((cur->sk_flags & SK_ROW_HEADER) || cur->sk_subtype != InvalidOid)
 		{
-			memcpy(&outkeys[new_numberOfKeys++], cur,
-				   sizeof(ScanKeyData));
+			ScanKey		outkey = &outkeys[new_numberOfKeys++];
+
+			memcpy(outkey, cur, sizeof(ScanKeyData));
+			if (numberOfEqualCols == attno - 1)
+				_bt_mark_scankey_required(outkey);
 			if (j == (BTEqualStrategyNumber - 1))
 				hasOtherTypeEqual = true;
 			continue;
@@ -515,6 +499,73 @@ _bt_preprocess_keys(IndexScanDesc scan)
 }
 
 /*
+ * Mark a scankey as "required to continue the scan".
+ *
+ * Depending on the operator type, the key may be required for both scan
+ * directions or just one.  Also, if the key is a row comparison header,
+ * we have to mark the appropriate subsidiary ScanKeys as required.  In
+ * such cases, the first subsidiary key is required, but subsequent ones
+ * are required only as long as they correspond to successive index columns.
+ * Otherwise the row comparison ordering is different from the index ordering
+ * and so we can't stop the scan on the basis of those lower-order columns.
+ *
+ * Note: when we set required-key flag bits in a subsidiary scankey, we are
+ * scribbling on a data structure belonging to the index AM's caller, not on
+ * our private copy.  This should be OK because the marking will not change
+ * from scan to scan within a query, and so we'd just re-mark the same way
+ * anyway on a rescan.  Something to keep an eye on though.
+ */
+static void
+_bt_mark_scankey_required(ScanKey skey)
+{
+	int		addflags;
+
+	switch (skey->sk_strategy)
+	{
+		case BTLessStrategyNumber:
+		case BTLessEqualStrategyNumber:
+			addflags = SK_BT_REQFWD;
+			break;
+		case BTEqualStrategyNumber:
+			addflags = SK_BT_REQFWD | SK_BT_REQBKWD;
+			break;
+		case BTGreaterEqualStrategyNumber:
+		case BTGreaterStrategyNumber:
+			addflags = SK_BT_REQBKWD;
+			break;
+		default:
+			elog(ERROR, "unrecognized StrategyNumber: %d",
+				 (int) skey->sk_strategy);
+			addflags = 0;		/* keep compiler quiet */
+			break;
+	}
+
+	skey->sk_flags |= addflags;
+
+	if (skey->sk_flags & SK_ROW_HEADER)
+	{
+		ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
+		AttrNumber attno = skey->sk_attno;
+
+		/* First subkey should be same as the header says */
+		Assert(subkey->sk_attno == attno);
+
+		for (;;)
+		{
+			Assert(subkey->sk_flags & SK_ROW_MEMBER);
+			Assert(subkey->sk_strategy == skey->sk_strategy);
+			if (subkey->sk_attno != attno)
+				break;			/* non-adjacent key, so not required */
+			subkey->sk_flags |= addflags;
+			if (subkey->sk_flags & SK_ROW_END)
+				break;
+			subkey++;
+			attno++;
+		}
+	}
+}
+
+/*
  * Test whether an indextuple satisfies all the scankey conditions.
  *
  * If so, copy its TID into scan->xs_ctup.t_self, and return TRUE.
@@ -595,6 +646,14 @@ _bt_checkkeys(IndexScanDesc scan,
 		bool		isNull;
 		Datum		test;
 
+		/* row-comparison keys need special processing */
+		if (key->sk_flags & SK_ROW_HEADER)
+		{
+			if (_bt_check_rowcompare(key, tuple, tupdesc, dir, continuescan))
+				continue;
+			return false;
+		}
+
 		datum = index_getattr(tuple,
 							  key->sk_attno,
 							  tupdesc,
@@ -660,3 +719,136 @@ _bt_checkkeys(IndexScanDesc scan,
 
 	return tuple_valid;
 }
+
+/*
+ * Test whether an indextuple satisfies a row-comparison scan condition.
+ *
+ * Return true if so, false if not.  If not, also clear *continuescan if
+ * it's not possible for any future tuples in the current scan direction
+ * to pass the qual.
+ *
+ * This is a subroutine for _bt_checkkeys, which see for more info.
+ */
+static bool
+_bt_check_rowcompare(ScanKey skey, IndexTuple tuple, TupleDesc tupdesc,
+					 ScanDirection dir, bool *continuescan)
+{
+	ScanKey		subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
+	int32		cmpresult = 0;
+	bool		result;
+
+	/* First subkey should be same as the header says */
+	Assert(subkey->sk_attno == skey->sk_attno);
+
+	/* Loop over columns of the row condition */
+	for (;;)
+	{
+		Datum		datum;
+		bool		isNull;
+
+		Assert(subkey->sk_flags & SK_ROW_MEMBER);
+		Assert(subkey->sk_strategy == skey->sk_strategy);
+
+		datum = index_getattr(tuple,
+							  subkey->sk_attno,
+							  tupdesc,
+							  &isNull);
+
+		if (isNull)
+		{
+			/*
+			 * Since NULLs are sorted after non-NULLs, we know we have reached
+			 * the upper limit of the range of values for this index attr.	On
+			 * a forward scan, we can stop if this qual is one of the "must
+			 * match" subset.  On a backward scan, however, we should keep
+			 * going.
+			 */
+			if ((subkey->sk_flags & SK_BT_REQFWD) &&
+				ScanDirectionIsForward(dir))
+				*continuescan = false;
+
+			/*
+			 * In any case, this indextuple doesn't match the qual.
+			 */
+			return false;
+		}
+
+		if (subkey->sk_flags & SK_ISNULL)
+		{
+			/*
+			 * Unlike the simple-scankey case, this isn't a disallowed case.
+			 * But it can never match.  If all the earlier row comparison
+			 * columns are required for the scan direction, we can stop
+			 * the scan, because there can't be another tuple that will
+			 * succeed.
+			 */
+			if (subkey != (ScanKey) DatumGetPointer(skey->sk_argument))
+				subkey--;
+			if ((subkey->sk_flags & SK_BT_REQFWD) &&
+				ScanDirectionIsForward(dir))
+				*continuescan = false;
+			else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
+					 ScanDirectionIsBackward(dir))
+				*continuescan = false;
+			return false;
+		}
+
+		/* Perform the test --- three-way comparison not bool operator */
+		cmpresult = DatumGetInt32(FunctionCall2(&subkey->sk_func,
+												datum,
+												subkey->sk_argument));
+
+		/* Done comparing if unequal, else advance to next column */
+		if (cmpresult != 0)
+			break;
+
+		if (subkey->sk_flags & SK_ROW_END)
+			break;
+		subkey++;
+	}
+
+	/*
+	 * At this point cmpresult indicates the overall result of the row
+	 * comparison, and subkey points to the deciding column (or the last
+	 * column if the result is "=").
+	 */
+	switch (subkey->sk_strategy)
+	{
+		/* EQ and NE cases aren't allowed here */
+		case BTLessStrategyNumber:
+			result = (cmpresult < 0);
+			break;
+		case BTLessEqualStrategyNumber:
+			result = (cmpresult <= 0);
+			break;
+		case BTGreaterEqualStrategyNumber:
+			result = (cmpresult >= 0);
+			break;
+		case BTGreaterStrategyNumber:
+			result = (cmpresult > 0);
+			break;
+		default:
+			elog(ERROR, "unrecognized RowCompareType: %d",
+				 (int) subkey->sk_strategy);
+			result = 0;			/* keep compiler quiet */
+			break;
+	}
+
+	if (!result)
+	{
+		/*
+		 * Tuple fails this qual.  If it's a required qual for the current
+		 * scan direction, then we can conclude no further tuples will
+		 * pass, either.  Note we have to look at the deciding column, not
+		 * necessarily the first or last column of the row condition.
+		 */
+		if ((subkey->sk_flags & SK_BT_REQFWD) &&
+			ScanDirectionIsForward(dir))
+			*continuescan = false;
+		else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
+				 ScanDirectionIsBackward(dir))
+			*continuescan = false;
+	}
+
+	return result;
+}
diff --git a/src/backend/executor/nodeBitmapIndexscan.c b/src/backend/executor/nodeBitmapIndexscan.c
index 114de29ba8e..37839c0255b 100644
--- a/src/backend/executor/nodeBitmapIndexscan.c
+++ b/src/backend/executor/nodeBitmapIndexscan.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/executor/nodeBitmapIndexscan.c,v 1.14 2005/12/03 05:51:01 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/executor/nodeBitmapIndexscan.c,v 1.15 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -245,6 +245,20 @@ ExecInitBitmapIndexScan(BitmapIndexScan *node, EState *estate)
 #define BITMAPINDEXSCAN_NSLOTS 0
 
 	/*
+	 * We do not open or lock the base relation here.  We assume that an
+	 * ancestor BitmapHeapScan node is holding AccessShareLock (or better)
+	 * on the heap relation throughout the execution of the plan tree.
+	 */
+
+	indexstate->ss.ss_currentRelation = NULL;
+	indexstate->ss.ss_currentScanDesc = NULL;
+
+	/*
+	 * Open the index relation.
+	 */
+	indexstate->biss_RelationDesc = index_open(node->indexid);
+
+	/*
 	 * Initialize index-specific scan state
 	 */
 	indexstate->biss_RuntimeKeysReady = false;
@@ -255,6 +269,7 @@ ExecInitBitmapIndexScan(BitmapIndexScan *node, EState *estate)
 	 * build the index scan keys from the index qualification
 	 */
 	ExecIndexBuildScanKeys((PlanState *) indexstate,
+						   indexstate->biss_RelationDesc,
 						   node->indexqual,
 						   node->indexstrategy,
 						   node->indexsubtype,
@@ -286,16 +301,8 @@ ExecInitBitmapIndexScan(BitmapIndexScan *node, EState *estate)
 	}
 
 	/*
-	 * We do not open or lock the base relation here.  We assume that an
-	 * ancestor BitmapHeapScan node is holding AccessShareLock (or better)
-	 * on the heap relation throughout the execution of the plan tree.
-	 */
-
-	indexstate->ss.ss_currentRelation = NULL;
-	indexstate->ss.ss_currentScanDesc = NULL;
-
-	/*
-	 * Open the index relation and initialize relation and scan descriptors.
+	 * Initialize scan descriptor.
+	 *
 	 * Note we acquire no locks here; the index machinery does its own locks
 	 * and unlocks.  (We rely on having a lock on the parent table to
 	 * ensure the index won't go away!)  Furthermore, if the parent table
@@ -303,7 +310,6 @@ ExecInitBitmapIndexScan(BitmapIndexScan *node, EState *estate)
 	 * opened and write-locked the index, so we can tell the index machinery
 	 * not to bother getting an extra lock.
 	 */
-	indexstate->biss_RelationDesc = index_open(node->indexid);
 	relistarget = ExecRelationIsTargetRelation(estate, node->scan.scanrelid);
 	indexstate->biss_ScanDesc =
 		index_beginscan_multi(indexstate->biss_RelationDesc,
diff --git a/src/backend/executor/nodeIndexscan.c b/src/backend/executor/nodeIndexscan.c
index 94495b4bc71..16406c784f4 100644
--- a/src/backend/executor/nodeIndexscan.c
+++ b/src/backend/executor/nodeIndexscan.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/executor/nodeIndexscan.c,v 1.109 2005/12/03 05:51:02 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/executor/nodeIndexscan.c,v 1.110 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -26,6 +26,7 @@
 
 #include "access/genam.h"
 #include "access/heapam.h"
+#include "access/nbtree.h"
 #include "executor/execdebug.h"
 #include "executor/nodeIndexscan.h"
 #include "miscadmin.h"
@@ -505,6 +506,24 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
 	ExecInitScanTupleSlot(estate, &indexstate->ss);
 
 	/*
+	 * open the base relation and acquire appropriate lock on it.
+	 */
+	currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid);
+
+	indexstate->ss.ss_currentRelation = currentRelation;
+	indexstate->ss.ss_currentScanDesc = NULL;	/* no heap scan here */
+
+	/*
+	 * get the scan type from the relation descriptor.
+	 */
+	ExecAssignScanType(&indexstate->ss, RelationGetDescr(currentRelation), false);
+
+	/*
+	 * Open the index relation.
+	 */
+	indexstate->iss_RelationDesc = index_open(node->indexid);
+
+	/*
 	 * Initialize index-specific scan state
 	 */
 	indexstate->iss_RuntimeKeysReady = false;
@@ -515,6 +534,7 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
 	 * build the index scan keys from the index qualification
 	 */
 	ExecIndexBuildScanKeys((PlanState *) indexstate,
+						   indexstate->iss_RelationDesc,
 						   node->indexqual,
 						   node->indexstrategy,
 						   node->indexsubtype,
@@ -545,20 +565,8 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
 	}
 
 	/*
-	 * open the base relation and acquire appropriate lock on it.
-	 */
-	currentRelation = ExecOpenScanRelation(estate, node->scan.scanrelid);
-
-	indexstate->ss.ss_currentRelation = currentRelation;
-	indexstate->ss.ss_currentScanDesc = NULL;	/* no heap scan here */
-
-	/*
-	 * get the scan type from the relation descriptor.
-	 */
-	ExecAssignScanType(&indexstate->ss, RelationGetDescr(currentRelation), false);
-
-	/*
-	 * Open the index relation and initialize relation and scan descriptors.
+	 * Initialize scan descriptor.
+	 *
 	 * Note we acquire no locks here; the index machinery does its own locks
 	 * and unlocks.  (We rely on having a lock on the parent table to
 	 * ensure the index won't go away!)  Furthermore, if the parent table
@@ -566,7 +574,6 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
 	 * opened and write-locked the index, so we can tell the index machinery
 	 * not to bother getting an extra lock.
 	 */
-	indexstate->iss_RelationDesc = index_open(node->indexid);
 	relistarget = ExecRelationIsTargetRelation(estate, node->scan.scanrelid);
 	indexstate->iss_ScanDesc = index_beginscan(currentRelation,
 											   indexstate->iss_RelationDesc,
@@ -595,7 +602,7 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
  * The index quals are passed to the index AM in the form of a ScanKey array.
  * This routine sets up the ScanKeys, fills in all constant fields of the
  * ScanKeys, and prepares information about the keys that have non-constant
- * comparison values.  We divide index qual expressions into three types:
+ * comparison values.  We divide index qual expressions into four types:
  *
  * 1. Simple operator with constant comparison value ("indexkey op constant").
  * For these, we just fill in a ScanKey containing the constant value.
@@ -605,7 +612,12 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
  * expression value, and set up an IndexRuntimeKeyInfo struct to drive
  * evaluation of the expression at the right times.
  *
- * 3. ScalarArrayOpExpr ("indexkey op ANY (array-expression)").  For these,
+ * 3. RowCompareExpr ("(indexkey, indexkey, ...) op (expr, expr, ...)").
+ * For these, we create a header ScanKey plus a subsidiary ScanKey array,
+ * as specified in access/skey.h.  The elements of the row comparison
+ * can have either constant or non-constant comparison values.
+ *
+ * 4. ScalarArrayOpExpr ("indexkey op ANY (array-expression)").  For these,
  * we create a ScanKey with everything filled in except the comparison value,
  * and set up an IndexArrayKeyInfo struct to drive processing of the qual.
  * (Note that we treat all array-expressions as requiring runtime evaluation,
@@ -614,10 +626,15 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
  * Input params are:
  *
  * planstate: executor state node we are working for
+ * index: the index we are building scan keys for
  * quals: indexquals expressions
  * strategies: associated operator strategy numbers
  * subtypes: associated operator subtype OIDs
  *
+ * (Any elements of the strategies and subtypes lists that correspond to
+ * RowCompareExpr quals are not used here; instead we look up the info
+ * afresh.)
+ *
  * Output params are:
  *
  * *scanKeys: receives ptr to array of ScanKeys
@@ -631,8 +648,8 @@ ExecInitIndexScan(IndexScan *node, EState *estate)
  * ScalarArrayOpExpr quals are not supported.
  */
 void
-ExecIndexBuildScanKeys(PlanState *planstate, List *quals,
-					   List *strategies, List *subtypes,
+ExecIndexBuildScanKeys(PlanState *planstate, Relation index,
+					   List *quals, List *strategies, List *subtypes,
 					   ScanKey *scanKeys, int *numScanKeys,
 					   IndexRuntimeKeyInfo **runtimeKeys, int *numRuntimeKeys,
 					   IndexArrayKeyInfo **arrayKeys, int *numArrayKeys)
@@ -644,21 +661,43 @@ ExecIndexBuildScanKeys(PlanState *planstate, List *quals,
 	IndexRuntimeKeyInfo *runtime_keys;
 	IndexArrayKeyInfo *array_keys;
 	int			n_scan_keys;
+	int			extra_scan_keys;
 	int			n_runtime_keys;
 	int			n_array_keys;
 	int			j;
 
+	/*
+	 * If there are any RowCompareExpr quals, we need extra ScanKey entries
+	 * for them, and possibly extra runtime-key entries.  Count up what's
+	 * needed.  (The subsidiary ScanKey arrays for the RowCompareExprs could
+	 * be allocated as separate chunks, but we have to count anyway to make
+	 * runtime_keys large enough, so might as well just do one palloc.)
+	 */
 	n_scan_keys = list_length(quals);
-	scan_keys = (ScanKey) palloc(n_scan_keys * sizeof(ScanKeyData));
+	extra_scan_keys = 0;
+	foreach(qual_cell, quals)
+	{
+		if (IsA(lfirst(qual_cell), RowCompareExpr))
+			extra_scan_keys +=
+				list_length(((RowCompareExpr *) lfirst(qual_cell))->opnos);
+	}
+	scan_keys = (ScanKey)
+		palloc((n_scan_keys + extra_scan_keys) * sizeof(ScanKeyData));
 	/* Allocate these arrays as large as they could possibly need to be */
 	runtime_keys = (IndexRuntimeKeyInfo *)
-		palloc(n_scan_keys * sizeof(IndexRuntimeKeyInfo));
+		palloc((n_scan_keys + extra_scan_keys) * sizeof(IndexRuntimeKeyInfo));
 	array_keys = (IndexArrayKeyInfo *)
 		palloc0(n_scan_keys * sizeof(IndexArrayKeyInfo));
 	n_runtime_keys = 0;
 	n_array_keys = 0;
 
 	/*
+	 * Below here, extra_scan_keys is index of first cell to use for next
+	 * RowCompareExpr
+	 */
+	extra_scan_keys = n_scan_keys;
+
+	/*
 	 * for each opclause in the given qual, convert each qual's opclause into
 	 * a single scan key
 	 */
@@ -749,6 +788,119 @@ ExecIndexBuildScanKeys(PlanState *planstate, List *quals,
 								   opfuncid,	/* reg proc to use */
 								   scanvalue);	/* constant */
 		}
+		else if (IsA(clause, RowCompareExpr))
+		{
+			/* (indexkey, indexkey, ...) op (expression, expression, ...) */
+			RowCompareExpr *rc = (RowCompareExpr *) clause;
+			ListCell *largs_cell = list_head(rc->largs);
+			ListCell *rargs_cell = list_head(rc->rargs);
+			ListCell *opnos_cell = list_head(rc->opnos);
+			ScanKey		first_sub_key = &scan_keys[extra_scan_keys];
+
+			/* Scan RowCompare columns and generate subsidiary ScanKey items */
+			while (opnos_cell != NULL)
+			{
+				ScanKey		this_sub_key = &scan_keys[extra_scan_keys];
+				int			flags = SK_ROW_MEMBER;
+				Datum		scanvalue;
+				Oid			opno;
+				Oid			opclass;
+				int			op_strategy;
+				Oid			op_subtype;
+				bool		op_recheck;
+
+				/*
+				 * leftop should be the index key Var, possibly relabeled
+				 */
+				leftop = (Expr *) lfirst(largs_cell);
+				largs_cell = lnext(largs_cell);
+
+				if (leftop && IsA(leftop, RelabelType))
+					leftop = ((RelabelType *) leftop)->arg;
+
+				Assert(leftop != NULL);
+
+				if (!(IsA(leftop, Var) &&
+					  var_is_rel((Var *) leftop)))
+					elog(ERROR, "indexqual doesn't have key on left side");
+
+				varattno = ((Var *) leftop)->varattno;
+
+				/*
+				 * rightop is the constant or variable comparison value
+				 */
+				rightop = (Expr *) lfirst(rargs_cell);
+				rargs_cell = lnext(rargs_cell);
+
+				if (rightop && IsA(rightop, RelabelType))
+					rightop = ((RelabelType *) rightop)->arg;
+
+				Assert(rightop != NULL);
+
+				if (IsA(rightop, Const))
+				{
+					/* OK, simple constant comparison value */
+					scanvalue = ((Const *) rightop)->constvalue;
+					if (((Const *) rightop)->constisnull)
+						flags |= SK_ISNULL;
+				}
+				else
+				{
+					/* Need to treat this one as a runtime key */
+					runtime_keys[n_runtime_keys].scan_key = this_sub_key;
+					runtime_keys[n_runtime_keys].key_expr =
+						ExecInitExpr(rightop, planstate);
+					n_runtime_keys++;
+					scanvalue = (Datum) 0;
+				}
+
+				/*
+				 * We have to look up the operator's associated btree support
+				 * function
+				 */
+				opno = lfirst_oid(opnos_cell);
+				opnos_cell = lnext(opnos_cell);
+
+				if (index->rd_rel->relam != BTREE_AM_OID ||
+					varattno < 1 || varattno > index->rd_index->indnatts)
+					elog(ERROR, "bogus RowCompare index qualification");
+				opclass = index->rd_indclass->values[varattno - 1];
+
+				get_op_opclass_properties(opno, opclass,
+									&op_strategy, &op_subtype, &op_recheck);
+
+				if (op_strategy != rc->rctype)
+					elog(ERROR, "RowCompare index qualification contains wrong operator");
+
+				opfuncid = get_opclass_proc(opclass, op_subtype, BTORDER_PROC);
+
+				/*
+				 * initialize the subsidiary scan key's fields appropriately
+				 */
+				ScanKeyEntryInitialize(this_sub_key,
+									   flags,
+									   varattno,	/* attribute number */
+									   op_strategy,	/* op's strategy */
+									   op_subtype,	/* strategy subtype */
+									   opfuncid,	/* reg proc to use */
+									   scanvalue);	/* constant */
+				extra_scan_keys++;
+			}
+
+			/* Mark the last subsidiary scankey correctly */
+			scan_keys[extra_scan_keys - 1].sk_flags |= SK_ROW_END;
+
+			/*
+			 * We don't use ScanKeyEntryInitialize for the header because
+			 * it isn't going to contain a valid sk_func pointer.
+			 */
+			MemSet(this_scan_key, 0, sizeof(ScanKeyData));
+			this_scan_key->sk_flags = SK_ROW_HEADER;
+			this_scan_key->sk_attno = first_sub_key->sk_attno;
+			this_scan_key->sk_strategy = rc->rctype;
+			/* sk_subtype, sk_func not used in a header */
+			this_scan_key->sk_argument = PointerGetDatum(first_sub_key);
+		}
 		else if (IsA(clause, ScalarArrayOpExpr))
 		{
 			/* indexkey op ANY (array-expression) */
diff --git a/src/backend/optimizer/path/indxpath.c b/src/backend/optimizer/path/indxpath.c
index 159960764fb..9ec5911403f 100644
--- a/src/backend/optimizer/path/indxpath.c
+++ b/src/backend/optimizer/path/indxpath.c
@@ -9,7 +9,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/path/indxpath.c,v 1.196 2005/12/06 16:50:36 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/path/indxpath.c,v 1.197 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -61,6 +61,11 @@ static bool match_clause_to_indexcol(IndexOptInfo *index,
 						 SaOpControl saop_control);
 static bool is_indexable_operator(Oid expr_op, Oid opclass,
 								  bool indexkey_on_left);
+static bool match_rowcompare_to_indexcol(IndexOptInfo *index,
+							 int indexcol,
+							 Oid opclass,
+							 RowCompareExpr *clause,
+							 Relids outer_relids);
 static Relids indexable_outerrelids(RelOptInfo *rel);
 static bool matches_any_index(RestrictInfo *rinfo, RelOptInfo *rel,
 				  Relids outer_relids);
@@ -82,7 +87,10 @@ static bool match_special_index_operator(Expr *clause, Oid opclass,
 							 bool indexkey_on_left);
 static Expr *expand_boolean_index_clause(Node *clause, int indexcol,
 							IndexOptInfo *index);
-static List *expand_indexqual_condition(RestrictInfo *rinfo, Oid opclass);
+static List *expand_indexqual_opclause(RestrictInfo *rinfo, Oid opclass);
+static RestrictInfo *expand_indexqual_rowcompare(RestrictInfo *rinfo,
+							IndexOptInfo *index,
+							int indexcol);
 static List *prefix_quals(Node *leftop, Oid opclass,
 			 Const *prefix, Pattern_Prefix_Status pstatus);
 static List *network_prefix_quals(Node *leftop, Oid expr_op, Oid opclass,
@@ -900,6 +908,14 @@ group_clauses_by_indexkey(IndexOptInfo *index,
  *	  We do not actually do the commuting here, but we check whether a
  *	  suitable commutator operator is available.
  *
+ *	  It is also possible to match RowCompareExpr clauses to indexes (but
+ *	  currently, only btree indexes handle this).  In this routine we will
+ *	  report a match if the first column of the row comparison matches the
+ *	  target index column.  This is sufficient to guarantee that some index
+ *	  condition can be constructed from the RowCompareExpr --- whether the
+ *	  remaining columns match the index too is considered in
+ *	  expand_indexqual_rowcompare().
+ *
  *	  It is also possible to match ScalarArrayOpExpr clauses to indexes, when
  *	  the clause is of the form "indexkey op ANY (arrayconst)".  Since the
  *	  executor can only handle these in the context of bitmap index scans,
@@ -944,7 +960,8 @@ match_clause_to_indexcol(IndexOptInfo *index,
 
 	/*
 	 * Clause must be a binary opclause, or possibly a ScalarArrayOpExpr
-	 * (which is always binary, by definition).
+	 * (which is always binary, by definition).  Or it could be a
+	 * RowCompareExpr, which we pass off to match_rowcompare_to_indexcol().
 	 */
 	if (is_opclause(clause))
 	{
@@ -972,6 +989,12 @@ match_clause_to_indexcol(IndexOptInfo *index,
 		expr_op = saop->opno;
 		plain_op = false;
 	}
+	else if (clause && IsA(clause, RowCompareExpr))
+	{
+		return match_rowcompare_to_indexcol(index, indexcol, opclass,
+											(RowCompareExpr *) clause,
+											outer_relids);
+	}
 	else
 		return false;
 
@@ -1039,6 +1062,74 @@ is_indexable_operator(Oid expr_op, Oid opclass, bool indexkey_on_left)
 	return op_in_opclass(expr_op, opclass);
 }
 
+/*
+ * match_rowcompare_to_indexcol()
+ *	  Handles the RowCompareExpr case for match_clause_to_indexcol(),
+ *	  which see for comments.
+ */
+static bool
+match_rowcompare_to_indexcol(IndexOptInfo *index,
+							 int indexcol,
+							 Oid opclass,
+							 RowCompareExpr *clause,
+							 Relids outer_relids)
+{
+	Node	   *leftop,
+			   *rightop;
+	Oid			expr_op;
+
+	/* Forget it if we're not dealing with a btree index */
+	if (index->relam != BTREE_AM_OID)
+		return false;
+
+	/*
+	 * We could do the matching on the basis of insisting that the opclass
+	 * shown in the RowCompareExpr be the same as the index column's opclass,
+	 * but that does not work well for cross-type comparisons (the opclass
+	 * could be for the other datatype).  Also it would fail to handle indexes
+	 * using reverse-sort opclasses.  Instead, match if the operator listed in
+	 * the RowCompareExpr is the < <= > or >= member of the index opclass
+	 * (after commutation, if the indexkey is on the right).
+	 */
+	leftop = (Node *) linitial(clause->largs);
+	rightop = (Node *) linitial(clause->rargs);
+	expr_op = linitial_oid(clause->opnos);
+
+	/*
+	 * These syntactic tests are the same as in match_clause_to_indexcol()
+	 */
+	if (match_index_to_operand(leftop, indexcol, index) &&
+		bms_is_subset(pull_varnos(rightop), outer_relids) &&
+		!contain_volatile_functions(rightop))
+	{
+		/* OK, indexkey is on left */
+	}
+	else if (match_index_to_operand(rightop, indexcol, index) &&
+			 bms_is_subset(pull_varnos(leftop), outer_relids) &&
+			 !contain_volatile_functions(leftop))
+	{
+		/* indexkey is on right, so commute the operator */
+		expr_op = get_commutator(expr_op);
+		if (expr_op == InvalidOid)
+			return false;
+	}
+	else
+		return false;
+
+	/* We're good if the operator is the right type of opclass member */
+	switch (get_op_opclass_strategy(expr_op, opclass))
+	{
+		case BTLessStrategyNumber:
+		case BTLessEqualStrategyNumber:
+		case BTGreaterEqualStrategyNumber:
+		case BTGreaterStrategyNumber:
+			return true;
+	}
+
+	return false;
+}
+
+
 /****************************************************************************
  *				----  ROUTINES TO DO PARTIAL INDEX PREDICATE TESTS	----
  ****************************************************************************/
@@ -2014,7 +2105,8 @@ match_special_index_operator(Expr *clause, Oid opclass,
  *	  of index qual clauses.  Standard qual clauses (those in the index's
  *	  opclass) are passed through unchanged.  Boolean clauses and "special"
  *	  index operators are expanded into clauses that the indexscan machinery
- *	  will know what to do with.
+ *	  will know what to do with.  RowCompare clauses are simplified if
+ *	  necessary to create a clause that is fully checkable by the index.
  *
  * The input list is ordered by index key, and so the output list is too.
  * (The latter is not depended on by any part of the core planner, I believe,
@@ -2041,13 +2133,14 @@ expand_indexqual_conditions(IndexOptInfo *index, List *clausegroups)
 		foreach(l, (List *) lfirst(clausegroup_item))
 		{
 			RestrictInfo *rinfo = (RestrictInfo *) lfirst(l);
+			Expr   *clause = rinfo->clause;
 
 			/* First check for boolean cases */
 			if (IsBooleanOpclass(curClass))
 			{
 				Expr	   *boolqual;
 
-				boolqual = expand_boolean_index_clause((Node *) rinfo->clause,
+				boolqual = expand_boolean_index_clause((Node *) clause,
 													   indexcol,
 													   index);
 				if (boolqual)
@@ -2061,16 +2154,31 @@ expand_indexqual_conditions(IndexOptInfo *index, List *clausegroups)
 				}
 			}
 
-			/* Next check for ScalarArrayOp cases */
-			if (IsA(rinfo->clause, ScalarArrayOpExpr))
+			/*
+			 * Else it must be an opclause (usual case), ScalarArrayOp, or
+			 * RowCompare
+			 */
+			if (is_opclause(clause))
 			{
+				resultquals = list_concat(resultquals,
+										  expand_indexqual_opclause(rinfo,
+																	curClass));
+			}
+			else if (IsA(clause, ScalarArrayOpExpr))
+			{
+				/* no extra work at this time */
 				resultquals = lappend(resultquals, rinfo);
-				continue;
 			}
-
-			resultquals = list_concat(resultquals,
-									  expand_indexqual_condition(rinfo,
-																 curClass));
+			else if (IsA(clause, RowCompareExpr))
+			{
+				resultquals = lappend(resultquals,
+									  expand_indexqual_rowcompare(rinfo,
+																  index,
+																  indexcol));
+			}
+			else
+				elog(ERROR, "unsupported indexqual type: %d",
+					 (int) nodeTag(clause));
 		}
 
 		clausegroup_item = lnext(clausegroup_item);
@@ -2145,16 +2253,15 @@ expand_boolean_index_clause(Node *clause,
 }
 
 /*
- * expand_indexqual_condition --- expand a single indexqual condition
- *		(other than a boolean-qual or ScalarArrayOp case)
+ * expand_indexqual_opclause --- expand a single indexqual condition
+ *		that is an operator clause
  *
  * The input is a single RestrictInfo, the output a list of RestrictInfos
  */
 static List *
-expand_indexqual_condition(RestrictInfo *rinfo, Oid opclass)
+expand_indexqual_opclause(RestrictInfo *rinfo, Oid opclass)
 {
 	Expr	   *clause = rinfo->clause;
-
 	/* we know these will succeed */
 	Node	   *leftop = get_leftop(clause);
 	Node	   *rightop = get_rightop(clause);
@@ -2225,6 +2332,204 @@ expand_indexqual_condition(RestrictInfo *rinfo, Oid opclass)
 }
 
 /*
+ * expand_indexqual_rowcompare --- expand a single indexqual condition
+ *		that is a RowCompareExpr
+ *
+ * It's already known that the first column of the row comparison matches
+ * the specified column of the index.  We can use additional columns of the
+ * row comparison as index qualifications, so long as they match the index
+ * in the "same direction", ie, the indexkeys are all on the same side of the
+ * clause and the operators are all the same-type members of the opclasses.
+ * If all the columns of the RowCompareExpr match in this way, we just use it
+ * as-is.  Otherwise, we build a shortened RowCompareExpr (if more than one
+ * column matches) or a simple OpExpr (if the first-column match is all
+ * there is).  In these cases the modified clause is always "<=" or ">="
+ * even when the original was "<" or ">" --- this is necessary to match all
+ * the rows that could match the original.  (We are essentially building a
+ * lossy version of the row comparison when we do this.)
+ */
+static RestrictInfo *
+expand_indexqual_rowcompare(RestrictInfo *rinfo,
+							IndexOptInfo *index,
+							int indexcol)
+{
+	RowCompareExpr *clause = (RowCompareExpr *) rinfo->clause;
+	bool	var_on_left;
+	int		op_strategy;
+	Oid		op_subtype;
+	bool	op_recheck;
+	int		matching_cols;
+	Oid		expr_op;
+	List   *opclasses;
+	List   *subtypes;
+	List   *new_ops;
+	ListCell *largs_cell;
+	ListCell *rargs_cell;
+	ListCell *opnos_cell;
+
+	/* We have to figure out (again) how the first col matches */
+	var_on_left = match_index_to_operand((Node *) linitial(clause->largs),
+										 indexcol, index);
+	Assert(var_on_left ||
+		   match_index_to_operand((Node *) linitial(clause->rargs),
+								  indexcol, index));
+	expr_op = linitial_oid(clause->opnos);
+	if (!var_on_left)
+		expr_op = get_commutator(expr_op);
+	get_op_opclass_properties(expr_op, index->classlist[indexcol],
+							  &op_strategy, &op_subtype, &op_recheck);
+	/* Build lists of the opclasses and operator subtypes in case needed */
+	opclasses = list_make1_oid(index->classlist[indexcol]);
+	subtypes = list_make1_oid(op_subtype);
+
+	/*
+	 * See how many of the remaining columns match some index column
+	 * in the same way.  A note about rel membership tests: we assume
+	 * that the clause as a whole is already known to use only Vars from
+	 * the indexed relation and possibly some acceptable outer relations.
+	 * So the "other" side of any potential index condition is OK as long
+	 * as it doesn't use Vars from the indexed relation.
+	 */
+	matching_cols = 1;
+	largs_cell = lnext(list_head(clause->largs));
+	rargs_cell = lnext(list_head(clause->rargs));
+	opnos_cell = lnext(list_head(clause->opnos));
+
+	while (largs_cell != NULL)
+	{
+		Node	   *varop;
+		Node	   *constop;
+		int			i;
+
+		expr_op = lfirst_oid(opnos_cell);
+		if (var_on_left)
+		{
+			varop = (Node *) lfirst(largs_cell);
+			constop = (Node *) lfirst(rargs_cell);
+		}
+		else
+		{
+			varop = (Node *) lfirst(rargs_cell);
+			constop = (Node *) lfirst(largs_cell);
+			/* indexkey is on right, so commute the operator */
+			expr_op = get_commutator(expr_op);
+			if (expr_op == InvalidOid)
+				break;			/* operator is not usable */
+		}
+		if (bms_is_member(index->rel->relid, pull_varnos(constop)))
+			break;				/* no good, Var on wrong side */
+		if (contain_volatile_functions(constop))
+			break;				/* no good, volatile comparison value */
+
+		/*
+		 * The Var side can match any column of the index.  If the user
+		 * does something weird like having multiple identical index
+		 * columns, we insist the match be on the first such column,
+		 * to avoid confusing the executor.
+		 */
+		for (i = 0; i < index->ncolumns; i++)
+		{
+			if (match_index_to_operand(varop, i, index))
+				break;
+		}
+		if (i >= index->ncolumns)
+			break;				/* no match found */
+
+		/* Now, do we have the right operator for this column? */
+		if (get_op_opclass_strategy(expr_op, index->classlist[i])
+			!= op_strategy)
+			break;
+
+		/* Add opclass and subtype to lists */
+		get_op_opclass_properties(expr_op, index->classlist[i],
+								  &op_strategy, &op_subtype, &op_recheck);
+		opclasses = lappend_oid(opclasses, index->classlist[i]);
+		subtypes = lappend_oid(subtypes, op_subtype);
+
+		/* This column matches, keep scanning */
+		matching_cols++;
+		largs_cell = lnext(largs_cell);
+		rargs_cell = lnext(rargs_cell);
+		opnos_cell = lnext(opnos_cell);
+	}
+
+	/* Return clause as-is if it's all usable as index quals */
+	if (matching_cols == list_length(clause->opnos))
+		return rinfo;
+
+	/*
+	 * We have to generate a subset rowcompare (possibly just one OpExpr).
+	 * The painful part of this is changing < to <= or > to >=, so deal with
+	 * that first.
+	 */
+	if (op_strategy == BTLessEqualStrategyNumber ||
+		op_strategy == BTGreaterEqualStrategyNumber)
+	{
+		/* easy, just use the same operators */
+		new_ops = list_truncate(list_copy(clause->opnos), matching_cols);
+	}
+	else
+	{
+		ListCell *opclasses_cell;
+		ListCell *subtypes_cell;
+
+		if (op_strategy == BTLessStrategyNumber)
+			op_strategy = BTLessEqualStrategyNumber;
+		else if (op_strategy == BTGreaterStrategyNumber)
+			op_strategy = BTGreaterEqualStrategyNumber;
+		else
+			elog(ERROR, "unexpected strategy number %d", op_strategy);
+		new_ops = NIL;
+		forboth(opclasses_cell, opclasses, subtypes_cell, subtypes)
+		{
+			expr_op = get_opclass_member(lfirst_oid(opclasses_cell),
+										 lfirst_oid(subtypes_cell),
+										 op_strategy);
+			if (!OidIsValid(expr_op))				/* should not happen */
+				elog(ERROR, "could not find member %d of opclass %u",
+					 op_strategy, lfirst_oid(opclasses_cell));
+			if (!var_on_left)
+			{
+				expr_op = get_commutator(expr_op);
+				if (!OidIsValid(expr_op))			/* should not happen */
+					elog(ERROR, "could not find commutator of member %d of opclass %u",
+						 op_strategy, lfirst_oid(opclasses_cell));
+			}
+			new_ops = lappend_oid(new_ops, expr_op);
+		}
+	}
+
+	/* If we have more than one matching col, create a subset rowcompare */
+	if (matching_cols > 1)
+	{
+		RowCompareExpr *rc = makeNode(RowCompareExpr);
+
+		if (var_on_left)
+			rc->rctype = (RowCompareType) op_strategy;
+		else
+			rc->rctype = (op_strategy == BTLessEqualStrategyNumber) ?
+				ROWCOMPARE_GE : ROWCOMPARE_LE;
+		rc->opnos = new_ops;
+		rc->opclasses = list_truncate(list_copy(clause->opclasses),
+									  matching_cols);
+		rc->largs = list_truncate((List *) copyObject(clause->largs),
+								  matching_cols);
+		rc->rargs = list_truncate((List *) copyObject(clause->rargs),
+								  matching_cols);
+		return make_restrictinfo((Expr *) rc, true, false, NULL);
+	}
+	else
+	{
+		Expr *opexpr;
+
+		opexpr = make_opclause(linitial_oid(new_ops), BOOLOID, false,
+							   copyObject(linitial(clause->largs)),
+							   copyObject(linitial(clause->rargs)));
+		return make_restrictinfo(opexpr, true, false, NULL);
+	}
+}
+
+/*
  * Given a fixed prefix that all the "leftop" values must have,
  * generate suitable indexqual condition(s).  opclass is the index
  * operator class; we use it to deduce the appropriate comparison
diff --git a/src/backend/optimizer/plan/createplan.c b/src/backend/optimizer/plan/createplan.c
index 4acac8421c8..e5355340c17 100644
--- a/src/backend/optimizer/plan/createplan.c
+++ b/src/backend/optimizer/plan/createplan.c
@@ -10,7 +10,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/plan/createplan.c,v 1.205 2005/11/26 22:14:56 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/plan/createplan.c,v 1.206 2006/01/25 20:29:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1583,7 +1583,7 @@ fix_indexqual_references(List *indexquals, IndexPath *index_path,
 			 * (only) the base relation.
 			 */
 			if (!bms_equal(rinfo->left_relids, index->rel->relids))
-				CommuteClause(op);
+				CommuteOpExpr(op);
 
 			/*
 			 * Now, determine which index attribute this is, change the
@@ -1594,6 +1594,41 @@ fix_indexqual_references(List *indexquals, IndexPath *index_path,
 													   &opclass);
 			clause_op = op->opno;
 		}
+		else if (IsA(clause, RowCompareExpr))
+		{
+			RowCompareExpr *rc = (RowCompareExpr *) clause;
+			ListCell *lc;
+
+			/*
+			 * Check to see if the indexkey is on the right; if so, commute
+			 * the clause. The indexkey should be the side that refers to
+			 * (only) the base relation.
+			 */
+			if (!bms_overlap(pull_varnos(linitial(rc->largs)),
+							 index->rel->relids))
+				CommuteRowCompareExpr(rc);
+
+			/*
+			 * For each column in the row comparison, determine which index
+			 * attribute this is and change the indexkey operand as needed.
+			 *
+			 * Save the index opclass for only the first column.  We will
+			 * return the operator and opclass info for just the first
+			 * column of the row comparison; the executor will have to
+			 * look up the rest if it needs them.
+			 */
+			foreach(lc, rc->largs)
+			{
+				Oid		tmp_opclass;
+
+				lfirst(lc) = fix_indexqual_operand(lfirst(lc),
+												   index,
+												   &tmp_opclass);
+				if (lc == list_head(rc->largs))
+					opclass = tmp_opclass;
+			}
+			clause_op = linitial_oid(rc->opnos);
+		}
 		else if (IsA(clause, ScalarArrayOpExpr))
 		{
 			ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
@@ -1745,7 +1780,7 @@ get_switched_clauses(List *clauses, Relids outerrelids)
 			temp->opretset = clause->opretset;
 			temp->args = list_copy(clause->args);
 			/* Commute it --- note this modifies the temp node in-place. */
-			CommuteClause(temp);
+			CommuteOpExpr(temp);
 			t_list = lappend(t_list, temp);
 		}
 		else
diff --git a/src/backend/optimizer/util/clauses.c b/src/backend/optimizer/util/clauses.c
index 2b6583c1dad..5266ff85d82 100644
--- a/src/backend/optimizer/util/clauses.c
+++ b/src/backend/optimizer/util/clauses.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/optimizer/util/clauses.c,v 1.205 2005/12/28 01:30:00 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/optimizer/util/clauses.c,v 1.206 2006/01/25 20:29:23 tgl Exp $
  *
  * HISTORY
  *	  AUTHOR			DATE			MAJOR EVENT
@@ -1167,12 +1167,12 @@ NumRelids(Node *clause)
 }
 
 /*
- * CommuteClause: commute a binary operator clause
+ * CommuteOpExpr: commute a binary operator clause
  *
  * XXX the clause is destructively modified!
  */
 void
-CommuteClause(OpExpr *clause)
+CommuteOpExpr(OpExpr *clause)
 {
 	Oid			opoid;
 	Node	   *temp;
@@ -1201,6 +1201,73 @@ CommuteClause(OpExpr *clause)
 }
 
 /*
+ * CommuteRowCompareExpr: commute a RowCompareExpr clause
+ *
+ * XXX the clause is destructively modified!
+ */
+void
+CommuteRowCompareExpr(RowCompareExpr *clause)
+{
+	List	   *newops;
+	List	   *temp;
+	ListCell   *l;
+
+	/* Sanity checks: caller is at fault if these fail */
+	if (!IsA(clause, RowCompareExpr))
+		elog(ERROR, "expected a RowCompareExpr");
+
+	/* Build list of commuted operators */
+	newops = NIL;
+	foreach(l, clause->opnos)
+	{
+		Oid			opoid = lfirst_oid(l);
+
+		opoid = get_commutator(opoid);
+		if (!OidIsValid(opoid))
+			elog(ERROR, "could not find commutator for operator %u",
+				 lfirst_oid(l));
+		newops = lappend_oid(newops, opoid);
+	}
+
+	/*
+	 * modify the clause in-place!
+	 */
+	switch (clause->rctype)
+	{
+		case ROWCOMPARE_LT:
+			clause->rctype = ROWCOMPARE_GT;
+			break;
+		case ROWCOMPARE_LE:
+			clause->rctype = ROWCOMPARE_GE;
+			break;
+		case ROWCOMPARE_GE:
+			clause->rctype = ROWCOMPARE_LE;
+			break;
+		case ROWCOMPARE_GT:
+			clause->rctype = ROWCOMPARE_LT;
+			break;
+		default:
+			elog(ERROR, "unexpected RowCompare type: %d",
+				 (int) clause->rctype);
+			break;
+	}
+
+	clause->opnos = newops;
+	/*
+	 * Note: we don't bother to update the opclasses list, but just set
+	 * it to empty.  This is OK since this routine is currently only used
+	 * for index quals, and the index machinery won't use the opclass
+	 * information.  The original opclass list is NOT valid if we have
+	 * commuted any cross-type comparisons, so don't leave it in place.
+	 */
+	clause->opclasses = NIL;	/* XXX */
+
+	temp = clause->largs;
+	clause->largs = clause->rargs;
+	clause->rargs = temp;
+}
+
+/*
  * strip_implicit_coercions: remove implicit coercions at top level of tree
  *
  * Note: there isn't any useful thing we can do with a RowExpr here, so
diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index 336c1deaeaa..cb9acf2d8a3 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -15,7 +15,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/selfuncs.c,v 1.196 2006/01/14 00:14:11 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/selfuncs.c,v 1.197 2006/01/25 20:29:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -4657,6 +4657,9 @@ btcostestimate(PG_FUNCTION_ARGS)
 	 * to find out which ones count as boundary quals.	We rely on the
 	 * knowledge that they are given in index column order.
 	 *
+	 * For a RowCompareExpr, we consider only the first column, just as
+	 * rowcomparesel() does.
+	 *
 	 * If there's a ScalarArrayOpExpr in the quals, we'll actually perform
 	 * N index scans not one, but the ScalarArrayOpExpr's operator can be
 	 * considered to act the same as it normally does.
@@ -4682,6 +4685,14 @@ btcostestimate(PG_FUNCTION_ARGS)
 			rightop = get_rightop(clause);
 			clause_op = ((OpExpr *) clause)->opno;
 		}
+		else if (IsA(clause, RowCompareExpr))
+		{
+			RowCompareExpr *rc = (RowCompareExpr *) clause;
+
+			leftop = (Node *) linitial(rc->largs);
+			rightop = (Node *) linitial(rc->rargs);
+			clause_op = linitial_oid(rc->opnos);
+		}
 		else if (IsA(clause, ScalarArrayOpExpr))
 		{
 			ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) clause;
diff --git a/src/include/access/skey.h b/src/include/access/skey.h
index f3845e55184..ecca1b84cff 100644
--- a/src/include/access/skey.h
+++ b/src/include/access/skey.h
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/access/skey.h,v 1.30 2006/01/14 22:03:35 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/access/skey.h,v 1.31 2006/01/25 20:29:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -70,6 +70,36 @@ typedef struct ScanKeyData
 typedef ScanKeyData *ScanKey;
 
 /*
+ * About row comparisons:
+ *
+ * The ScanKey data structure also supports row comparisons, that is ordered
+ * tuple comparisons like (x, y) > (c1, c2), having the SQL-spec semantics
+ * "x > c1 OR (x = c1 AND y > c2)".  Note that this is currently only
+ * implemented for btree index searches, not for heapscans or any other index
+ * type.  A row comparison is represented by a "header" ScanKey entry plus
+ * a separate array of ScanKeys, one for each column of the row comparison.
+ * The header entry has these properties:
+ *		sk_flags = SK_ROW_HEADER
+ *		sk_attno = index column number for leading column of row comparison
+ *		sk_strategy = btree strategy code for semantics of row comparison
+ *				(ie, < <= > or >=)
+ *		sk_subtype, sk_func: not used
+ *		sk_argument: pointer to subsidiary ScanKey array
+ * If the header is part of a ScanKey array that's sorted by attno, it
+ * must be sorted according to the leading column number.
+ *
+ * The subsidiary ScanKey array appears in logical column order of the row
+ * comparison, which may be different from index column order.  The array
+ * elements are like a normal ScanKey array except that:
+ *		sk_flags must include SK_ROW_MEMBER, plus SK_ROW_END in the last
+ *				element (needed since row header does not include a count)
+ *		sk_func points to the btree comparison support function for the
+ *				opclass, NOT the operator's implementation function.
+ * sk_strategy must be the same in all elements of the subsidiary array,
+ * that is, the same as in the header entry.
+ */
+
+/*
  * ScanKeyData sk_flags
  *
  * sk_flags bits 0-15 are reserved for system-wide use (symbols for those
@@ -78,6 +108,9 @@ typedef ScanKeyData *ScanKey;
  */
 #define SK_ISNULL		0x0001	/* sk_argument is NULL */
 #define SK_UNARY		0x0002	/* unary operator (currently unsupported) */
+#define SK_ROW_HEADER	0x0004	/* row comparison header (see above) */
+#define SK_ROW_MEMBER	0x0008	/* row comparison member (see above) */
+#define SK_ROW_END		0x0010	/* last row comparison member (see above) */
 
 
 /*
diff --git a/src/include/executor/nodeIndexscan.h b/src/include/executor/nodeIndexscan.h
index 21bb254f63e..d36defaa016 100644
--- a/src/include/executor/nodeIndexscan.h
+++ b/src/include/executor/nodeIndexscan.h
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/executor/nodeIndexscan.h,v 1.25 2005/11/25 19:47:50 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/executor/nodeIndexscan.h,v 1.26 2006/01/25 20:29:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -25,8 +25,8 @@ extern void ExecIndexRestrPos(IndexScanState *node);
 extern void ExecIndexReScan(IndexScanState *node, ExprContext *exprCtxt);
 
 /* routines exported to share code with nodeBitmapIndexscan.c */
-extern void ExecIndexBuildScanKeys(PlanState *planstate, List *quals,
-					   List *strategies, List *subtypes,
+extern void ExecIndexBuildScanKeys(PlanState *planstate, Relation index,
+					   List *quals, List *strategies, List *subtypes,
 					   ScanKey *scanKeys, int *numScanKeys,
 					   IndexRuntimeKeyInfo **runtimeKeys, int *numRuntimeKeys,
 					   IndexArrayKeyInfo **arrayKeys, int *numArrayKeys);
diff --git a/src/include/optimizer/clauses.h b/src/include/optimizer/clauses.h
index 0d3770dc5c4..11eb6417fa7 100644
--- a/src/include/optimizer/clauses.h
+++ b/src/include/optimizer/clauses.h
@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/include/optimizer/clauses.h,v 1.81 2005/12/20 02:30:36 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/optimizer/clauses.h,v 1.82 2006/01/25 20:29:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -67,7 +67,9 @@ extern bool has_distinct_clause(Query *query);
 extern bool has_distinct_on_clause(Query *query);
 
 extern int	NumRelids(Node *clause);
-extern void CommuteClause(OpExpr *clause);
+
+extern void CommuteOpExpr(OpExpr *clause);
+extern void CommuteRowCompareExpr(RowCompareExpr *clause);
 
 extern Node *strip_implicit_coercions(Node *node);