Push the responsibility for handling ignore_killed_tuples down into

_bt_checkkeys(), instead of checking it in the top-level nbtree.c routines
as formerly.  This saves a little bit of loop overhead, but more importantly
it lets us skip performing the index key comparisons for dead tuples.

src/backend/access/nbtree/nbtree.c

@@ -12,7 +12,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.134 2005/11/22 18:17:06 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.135 2005/12/07 19:37:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -288,21 +288,6 @@ btgettuple(PG_FUNCTION_ARGS)
 	else
 		res = _bt_first(scan, dir);
 
-	/*
-	 * Skip killed tuples if asked to.
-	 */
-	if (scan->ignore_killed_tuples)
-	{
-		while (res)
-		{
-			offnum = ItemPointerGetOffsetNumber(&(scan->currentItemData));
-			page = BufferGetPage(so->btso_curbuf);
-			if (!ItemIdDeleted(PageGetItemId(page, offnum)))
-				break;
-			res = _bt_next(scan, dir);
-		}
-	}
-
 	/*
 	 * Save heap TID to use it in _bt_restscan.  Then release the read lock on
 	 * the buffer so that we aren't blocking other backends.
@@ -353,25 +338,6 @@ btgetmulti(PG_FUNCTION_ARGS)
 			res = _bt_next(scan, ForwardScanDirection);
 		else
 			res = _bt_first(scan, ForwardScanDirection);
-
-		/*
-		 * Skip killed tuples if asked to.
-		 */
-		if (scan->ignore_killed_tuples)
-		{
-			while (res)
-			{
-				Page		page;
-				OffsetNumber offnum;
-
-				offnum = ItemPointerGetOffsetNumber(&(scan->currentItemData));
-				page = BufferGetPage(so->btso_curbuf);
-				if (!ItemIdDeleted(PageGetItemId(page, offnum)))
-					break;
-				res = _bt_next(scan, ForwardScanDirection);
-			}
-		}
-
 		if (!res)
 			break;
 		/* Save tuple ID, and continue scanning */
@@ -385,9 +351,8 @@ btgetmulti(PG_FUNCTION_ARGS)
 	 */
 	if (res)
 	{
-		((BTScanOpaque) scan->opaque)->curHeapIptr = scan->xs_ctup.t_self;
+		so->curHeapIptr = scan->xs_ctup.t_self;
-		LockBuffer(((BTScanOpaque) scan->opaque)->btso_curbuf,
+		LockBuffer(so->btso_curbuf, BUFFER_LOCK_UNLOCK);
-				   BUFFER_LOCK_UNLOCK);
 	}
 
 	*returned_tids = ntids;

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 1.98 2005/12/07 18:03:48 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.99 2005/12/07 19:37:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -423,8 +423,6 @@ _bt_next(IndexScanDesc scan, ScanDirection dir)
 	Page		page;
 	OffsetNumber offnum;
 	ItemPointer current;
-	BTItem		btitem;
-	IndexTuple	itup;
 	BTScanOpaque so;
 	bool		continuescan;
 
@@ -445,13 +443,10 @@ _bt_next(IndexScanDesc scan, ScanDirection dir)
 		/* current is the next candidate tuple to return */
 		offnum = ItemPointerGetOffsetNumber(current);
 		page = BufferGetPage(buf);
-		btitem = (BTItem) PageGetItem(page, PageGetItemId(page, offnum));
-		itup = &btitem->bti_itup;
 
-		if (_bt_checkkeys(scan, itup, dir, &continuescan))
+		if (_bt_checkkeys(scan, page, offnum, dir, &continuescan))
 		{
 			/* tuple passes all scan key conditions, so return it */
-			scan->xs_ctup.t_self = itup->t_tid;
 			return true;
 		}
 
@@ -485,8 +480,6 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	Page		page;
 	BTStack		stack;
 	OffsetNumber offnum;
-	BTItem		btitem;
-	IndexTuple	itup;
 	ItemPointer current;
 	BlockNumber blkno;
 	StrategyNumber strat;
@@ -848,14 +841,11 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	/* okay, current item pointer for the scan is right */
 	offnum = ItemPointerGetOffsetNumber(current);
 	page = BufferGetPage(buf);
-	btitem = (BTItem) PageGetItem(page, PageGetItemId(page, offnum));
-	itup = &btitem->bti_itup;
 
 	/* is the first item actually acceptable? */
-	if (_bt_checkkeys(scan, itup, dir, &continuescan))
+	if (_bt_checkkeys(scan, page, offnum, dir, &continuescan))
 	{
 		/* yes, return it */
-		scan->xs_ctup.t_self = itup->t_tid;
 		res = true;
 	}
 	else if (continuescan)
@@ -1215,8 +1205,6 @@ _bt_endpoint(IndexScanDesc scan, ScanDirection dir)
 	OffsetNumber maxoff;
 	OffsetNumber start;
 	BlockNumber blkno;
-	BTItem		btitem;
-	IndexTuple	itup;
 	BTScanOpaque so;
 	bool		res;
 	bool		continuescan;
@@ -1284,16 +1272,12 @@ _bt_endpoint(IndexScanDesc scan, ScanDirection dir)
 		page = BufferGetPage(buf);
 	}
 
-	btitem = (BTItem) PageGetItem(page, PageGetItemId(page, start));
-	itup = &(btitem->bti_itup);
-
 	/*
 	 * Okay, we are on the first or last tuple.  Does it pass all the quals?
 	 */
-	if (_bt_checkkeys(scan, itup, dir, &continuescan))
+	if (_bt_checkkeys(scan, page, start, dir, &continuescan))
 	{
 		/* yes, return it */
-		scan->xs_ctup.t_self = itup->t_tid;
 		res = true;
 	}
 	else if (continuescan)

src/backend/access/nbtree/nbtutils.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtutils.c,v 1.66 2005/11/22 18:17:06 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtutils.c,v 1.67 2005/12/07 19:37:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -477,30 +477,77 @@ _bt_preprocess_keys(IndexScanDesc scan)
 /*
  * Test whether an indextuple satisfies all the scankey conditions.
  *
+ * If so, copy its TID into scan->xs_ctup.t_self, and return TRUE.
+ * If not, return FALSE (xs_ctup is not changed).
+ *
  * If the tuple fails to pass the qual, we also determine whether there's
  * any need to continue the scan beyond this tuple, and set *continuescan
  * accordingly.  See comments for _bt_preprocess_keys(), above, about how
  * this is done.
+ *
+ * scan: index scan descriptor
+ * page: buffer page containing index tuple
+ * offnum: offset number of index tuple (must be a valid item!)
+ * dir: direction we are scanning in
+ * continuescan: output parameter (will be set correctly in all cases)
  */
 bool
-_bt_checkkeys(IndexScanDesc scan, IndexTuple tuple,
+_bt_checkkeys(IndexScanDesc scan,
+			  Page page, OffsetNumber offnum,
 			  ScanDirection dir, bool *continuescan)
 {
-	BTScanOpaque so = (BTScanOpaque) scan->opaque;
+	ItemId		iid = PageGetItemId(page, offnum);
-	int			keysz = so->numberOfKeys;
+	bool		tuple_valid;
-	int			ikey;
+	BTItem		btitem;
+	IndexTuple	tuple;
 	TupleDesc	tupdesc;
+	BTScanOpaque so;
+	int			keysz;
+	int			ikey;
 	ScanKey		key;
 
-	*continuescan = true;
+	*continuescan = true;		/* default assumption */
 
-	/* If no keys, always scan the whole index */
+	/*
-	if (keysz == 0)
+	 * If the scan specifies not to return killed tuples, then we treat
-		return true;
+	 * a killed tuple as not passing the qual.  Most of the time, it's a
+	 * win to not bother examining the tuple's index keys, but just return
+	 * immediately with continuescan = true to proceed to the next tuple.
+	 * However, if this is the last tuple on the page, we should check
+	 * the index keys to prevent uselessly advancing to the next page.
+	 */
+	if (scan->ignore_killed_tuples && ItemIdDeleted(iid))
+	{
+		/* return immediately if there are more tuples on the page */
+		if (ScanDirectionIsForward(dir))
+		{
+			if (offnum < PageGetMaxOffsetNumber(page))
+				return false;
+		}
+		else
+		{
+			BTPageOpaque opaque = (BTPageOpaque) PageGetSpecialPointer(page);
+
+			if (offnum > P_FIRSTDATAKEY(opaque))
+				return false;
+		}
+		/*
+		 * OK, we want to check the keys, but we'll return FALSE even
+		 * if the tuple passes the key tests.
+		 */
+		tuple_valid = false;
+	}
+	else
+		tuple_valid = true;
+
+	btitem = (BTItem) PageGetItem(page, iid);
+	tuple = &btitem->bti_itup;
 
 	IncrIndexProcessed();
 
 	tupdesc = RelationGetDescr(scan->indexRelation);
+	so = (BTScanOpaque) scan->opaque;
+	keysz = so->numberOfKeys;
 
 	for (key = so->keyData, ikey = 0; ikey < keysz; key++, ikey++)
 	{
@@ -592,6 +639,9 @@ _bt_checkkeys(IndexScanDesc scan, IndexTuple tuple,
 		}
 	}
 
-	/* If we get here, the tuple passes all quals. */
+	/* If we get here, the tuple passes all index quals. */
-	return true;
+	if (tuple_valid)
+		scan->xs_ctup.t_self = tuple->t_tid;
+
+	return tuple_valid;
 }

src/include/access/nbtree.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/nbtree.h,v 1.88 2005/11/06 19:29:01 tgl Exp $
+ * $PostgreSQL: pgsql/src/include/access/nbtree.h,v 1.89 2005/12/07 19:37:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -467,8 +467,9 @@ extern ScanKey _bt_mkscankey_nodata(Relation rel);
 extern void _bt_freeskey(ScanKey skey);
 extern void _bt_freestack(BTStack stack);
 extern void _bt_preprocess_keys(IndexScanDesc scan);
-extern bool _bt_checkkeys(IndexScanDesc scan, IndexTuple tuple,
+extern bool _bt_checkkeys(IndexScanDesc scan,
-			  ScanDirection dir, bool *continuescan);
+						  Page page, OffsetNumber offnum,
+						  ScanDirection dir, bool *continuescan);
 extern BTItem _bt_formitem(IndexTuple itup);
 
 /*