1. Vacuum is updated for MVCC.

2. Much faster btree tuples deletion in the case when first on page
   index tuple is deleted (no movement to the left page(s)).
3. Remember blkno of new root page in BTPageOpaque of
   left/right siblings when root page is splitted.

src/backend/access/heap/heapam.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/heap/heapam.c,v 1.41 1999/02/13 23:14:22 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/heap/heapam.c,v 1.42 1999/03/28 20:31:56 vadim Exp $
  *
  *
  * INTERFACE ROUTINES
@@ -1270,7 +1270,7 @@ l2:
 	newtup->t_data->t_cmin = GetCurrentCommandId();
 	StoreInvalidTransactionId(&(newtup->t_data->t_xmax));
 	newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
-	newtup->t_data->t_infomask |= HEAP_XMAX_INVALID;
+	newtup->t_data->t_infomask |= (HEAP_XMAX_INVALID | HEAP_UPDATED);
 
 	/* logically delete old item */
 	TransactionIdStore(GetCurrentTransactionId(), &(oldtup.t_data->t_xmax));

src/backend/access/nbtree/nbtinsert.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.35 1999/02/13 23:14:34 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.36 1999/03/28 20:31:56 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -853,6 +853,8 @@ _bt_split(Relation rel, Buffer buf, OffsetNumber firstright)
 	lopaque->btpo_next = BufferGetBlockNumber(rbuf);
 	ropaque->btpo_next = oopaque->btpo_next;
 
+	lopaque->btpo_parent = ropaque->btpo_parent = oopaque->btpo_parent;
+
 	/*
 	 * If the page we're splitting is not the rightmost page at its level
 	 * in the tree, then the first (0) entry on the page is the high key
@@ -1103,6 +1105,7 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	/* get a new root page */
 	rootbuf = _bt_getbuf(rel, P_NEW, BT_WRITE);
 	rootpage = BufferGetPage(rootbuf);
+	rootbknum = BufferGetBlockNumber(rootbuf);
 	_bt_pageinit(rootpage, BufferGetPageSize(rootbuf));
 
 	/* set btree special data */
@@ -1119,6 +1122,10 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	lpage = BufferGetPage(lbuf);
 	rpage = BufferGetPage(rbuf);
 
+	((BTPageOpaque) PageGetSpecialPointer(lpage))->btpo_parent = 
+	((BTPageOpaque) PageGetSpecialPointer(rpage))->btpo_parent = 
+		rootbknum;
+
 	/*
 	 * step over the high key on the left page while building the left
 	 * page pointer.
@@ -1156,11 +1163,13 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	pfree(new_item);
 
 	/* write and let go of the root buffer */
-	rootbknum = BufferGetBlockNumber(rootbuf);
 	_bt_wrtbuf(rel, rootbuf);
 
 	/* update metadata page with new root block number */
 	_bt_metaproot(rel, rootbknum, 0);
+
+	WriteNoReleaseBuffer(lbuf);
+	WriteNoReleaseBuffer(rbuf);
 }
 
 /*
@@ -1559,6 +1568,7 @@ _bt_shift(Relation rel, Buffer buf, BTStack stack, int keysz,
 		pageop->btpo_flags |= BTP_CHAIN;
 	pageop->btpo_prev = npageop->btpo_prev;		/* restore prev */
 	pageop->btpo_next = nbknum; /* next points to the new page */
+	pageop->btpo_parent = npageop->btpo_parent;
 
 	/* init shifted page opaque */
 	npageop->btpo_prev = bknum = BufferGetBlockNumber(buf);

src/backend/access/nbtree/nbtpage.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtpage.c,v 1.18 1999/02/13 23:14:35 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtpage.c,v 1.19 1999/03/28 20:31:57 vadim Exp $
  *
  *	NOTES
  *	   Postgres btree pages look like ordinary relation pages.	The opaque
@@ -421,6 +421,8 @@ _bt_pageinit(Page page, Size size)
 	MemSet(page, 0, size);
 
 	PageInit(page, size, sizeof(BTPageOpaqueData));
+	((BTPageOpaque) PageGetSpecialPointer(page))->btpo_parent = 
+		InvalidBlockNumber;
 }
 
 /*

src/backend/access/nbtree/nbtree.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.36 1999/02/21 03:48:27 scrappy Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.37 1999/03/28 20:31:58 vadim Exp $
  *
  * NOTES
  *	  This file contains only the public interface routines.
@@ -372,11 +372,6 @@ btinsert(Relation rel, Datum *datum, char *nulls, ItemPointer ht_ctid, Relation
 	pfree(btitem);
 	pfree(itup);
 
-#ifdef NOT_USED
-	/* adjust any active scans that will be affected by this insertion */
-	_bt_adjscans(rel, &(res->pointerData), BT_INSERT);
-#endif
-
 	return res;
 }
 
@@ -396,15 +391,9 @@ btgettuple(IndexScanDesc scan, ScanDirection dir)
 
 	if (ItemPointerIsValid(&(scan->currentItemData)))
 	{
-
 		/*
-		 * Now we don't adjust scans on insertion (comments in
-		 * nbtscan.c:_bt_scandel()) and I hope that we will unlock current
-		 * index page before leaving index in LLL: this means that current
-		 * index tuple could be moved right before we get here and we have
-		 * to restore our scan position. We save heap TID pointed by
-		 * current index tuple and use it. This will work untill we start
-		 * to re-use (move heap tuples) without vacuum... - vadim 07/29/98
+		 * Restore scan position using heap TID returned 
+		 * by previous call to btgettuple().
 		 */
 		_bt_restscan(scan);
 		res = _bt_next(scan, dir);
@@ -612,16 +601,12 @@ void
 btdelete(Relation rel, ItemPointer tid)
 {
 	/* adjust any active scans that will be affected by this deletion */
-	_bt_adjscans(rel, tid, BT_DELETE);
+	_bt_adjscans(rel, tid);
 
 	/* delete the data from the page */
 	_bt_pagedel(rel, tid);
 }
 
-/*
- * Reasons are in btgettuple... We have to find index item that
- * points to heap tuple returned by previous call to btgettuple().
- */
 static void
 _bt_restscan(IndexScanDesc scan)
 {
@@ -637,6 +622,20 @@ _bt_restscan(IndexScanDesc scan)
 	BTItem		item;
 	BlockNumber blkno;
 
+	/*
+	 * We use this as flag when first index tuple on page
+	 * is deleted but we do not move left (this would
+	 * slowdown vacuum) - so we set current->ip_posid
+	 * before first index tuple on the current page
+	 * (_bt_step will move it right)...
+	 */
+	if (!ItemPointerIsValid(&target))
+	{
+		ItemPointerSetOffsetNumber(&(scan->currentItemData), 
+			OffsetNumberPrev(P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY));
+		return;
+	}
+
 	if (maxoff >= offnum)
 	{
 

src/backend/access/nbtree/nbtscan.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/Attic/nbtscan.c,v 1.19 1999/02/13 23:14:36 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/Attic/nbtscan.c,v 1.20 1999/03/28 20:31:58 vadim Exp $
  *
  *
  * NOTES
@@ -43,8 +43,7 @@ typedef BTScanListData *BTScanList;
 
 static BTScanList BTScans = (BTScanList) NULL;
 
-static void _bt_scandel(IndexScanDesc scan, int op, BlockNumber blkno, OffsetNumber offno);
-static bool _bt_scantouched(IndexScanDesc scan, BlockNumber blkno, OffsetNumber offno);
+static void _bt_scandel(IndexScanDesc scan, BlockNumber blkno, OffsetNumber offno);
 
 /*
  *	_bt_regscan() -- register a new scan.
@@ -91,7 +90,7 @@ _bt_dropscan(IndexScanDesc scan)
  *					  for a given deletion or insertion
  */
 void
-_bt_adjscans(Relation rel, ItemPointer tid, int op)
+_bt_adjscans(Relation rel, ItemPointer tid)
 {
 	BTScanList	l;
 	Oid			relid;
@@ -100,41 +99,25 @@ _bt_adjscans(Relation rel, ItemPointer tid, int op)
 	for (l = BTScans; l != (BTScanList) NULL; l = l->btsl_next)
 	{
 		if (relid == RelationGetRelid(l->btsl_scan->relation))
-			_bt_scandel(l->btsl_scan, op,
+			_bt_scandel(l->btsl_scan,
 						ItemPointerGetBlockNumber(tid),
 						ItemPointerGetOffsetNumber(tid));
 	}
 }
 
 /*
- *	_bt_scandel() -- adjust a single scan
+ *	_bt_scandel() -- adjust a single scan on deletion
  *
- * because each index page is always maintained as an ordered array of
- * index tuples, the index tuples on a given page shift beneath any
- * given scan.	an index modification "behind" a scan position (i.e.,
- * same page, lower or equal offset number) will therefore force us to
- * adjust the scan in the following ways:
- *
- * - on insertion, we shift the scan forward by one item.
- * - on deletion, we shift the scan backward by one item.
- *
- * note that:
- *
- * - we need not worry about the actual ScanDirection of the scan
- * itself, since the problem is that the "current" scan position has
- * shifted.
- * - modifications "ahead" of our scan position do not change the
- * array index of the current scan position and so can be ignored.
  */
 static void
-_bt_scandel(IndexScanDesc scan, int op, BlockNumber blkno, OffsetNumber offno)
+_bt_scandel(IndexScanDesc scan, BlockNumber blkno, OffsetNumber offno)
 {
 	ItemPointer		current;
 	Buffer			buf;
 	BTScanOpaque	so;
-
-	if (!_bt_scantouched(scan, blkno, offno))
-		return;
+	OffsetNumber	start;
+	Page			page;
+	BTPageOpaque	opaque;
 
 	so = (BTScanOpaque) scan->opaque;
 	buf = so->btso_curbuf;
@@ -144,40 +127,39 @@ _bt_scandel(IndexScanDesc scan, int op, BlockNumber blkno, OffsetNumber offno)
 		&& ItemPointerGetBlockNumber(current) == blkno
 		&& ItemPointerGetOffsetNumber(current) >= offno)
 	{
-		switch (op)
+		page = BufferGetPage(buf);
+		opaque = (BTPageOpaque) PageGetSpecialPointer(page);
+		start = P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY;
+		if (ItemPointerGetOffsetNumber(current) == start)
+			ItemPointerSetInvalid(&(so->curHeapIptr));
+		else
 		{
-/*
- * Problems occure when current scan page is splitted!
- * We saw "Non-functional updates" (ie index tuples were read twice)
- * and partial updates ("good" tuples were not read at all) - due to
- * losing scan position here. Look @ nbtree.c:btgettuple()
- * what we do now...		- vadim 07/29/98
-			case BT_INSERT:
-				_bt_step(scan, &buf, ForwardScanDirection);
-				break;
- */
-			case BT_DELETE:
 			_bt_step(scan, &buf, BackwardScanDirection);
-				break;
-			default:
-				elog(ERROR, "_bt_scandel: bad operation '%d'", op);
-				/* NOTREACHED */
-		}
 			so->btso_curbuf = buf;
 			if (ItemPointerIsValid(current))
 			{
-			Page		page = BufferGetPage(buf);
-			BTItem		btitem = (BTItem) PageGetItem(page,
-			   PageGetItemId(page, ItemPointerGetOffsetNumber(current)));
+				Page		pg = BufferGetPage(buf);
+				BTItem		btitem = (BTItem) PageGetItem(pg,
+				   PageGetItemId(pg, ItemPointerGetOffsetNumber(current)));
 
 				so->curHeapIptr = btitem->bti_itup.t_tid;
 			}
 		}
+	}
 
 	current = &(scan->currentMarkData);
 	if (ItemPointerIsValid(current)
 		&& ItemPointerGetBlockNumber(current) == blkno
 		&& ItemPointerGetOffsetNumber(current) >= offno)
+	{
+
+		page = BufferGetPage(so->btso_mrkbuf);
+		opaque = (BTPageOpaque) PageGetSpecialPointer(page);
+		start = P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY;
+
+		if (ItemPointerGetOffsetNumber(current) == start)
+			ItemPointerSetInvalid(&(so->mrkHeapIptr));
+		else
 		{
 			ItemPointerData tmp;
 
@@ -187,21 +169,9 @@ _bt_scandel(IndexScanDesc scan, int op, BlockNumber blkno, OffsetNumber offno)
 			so->btso_curbuf = so->btso_mrkbuf;
 			so->btso_mrkbuf = buf;
 			buf = so->btso_curbuf;
-		switch (op)
-		{
-/*
- * ...comments are above...
-			case BT_INSERT:
-				_bt_step(scan, &buf, ForwardScanDirection);
-				break;
- */
-			case BT_DELETE:
+
 			_bt_step(scan, &buf, BackwardScanDirection);
-				break;
-			default:
-				elog(ERROR, "_bt_scandel: bad operation '%d'", op);
-				/* NOTREACHED */
-		}
+
 			so->btso_curbuf = so->btso_mrkbuf;
 			so->btso_mrkbuf = buf;
 			tmp = *current;
@@ -209,35 +179,12 @@ _bt_scandel(IndexScanDesc scan, int op, BlockNumber blkno, OffsetNumber offno)
 			scan->currentItemData = tmp;
 			if (ItemPointerIsValid(current))
 			{
-			Page		page = BufferGetPage(buf);
-			BTItem		btitem = (BTItem) PageGetItem(page,
-			   PageGetItemId(page, ItemPointerGetOffsetNumber(current)));
+				Page		pg = BufferGetPage(buf);
+				BTItem		btitem = (BTItem) PageGetItem(pg,
+				   PageGetItemId(pg, ItemPointerGetOffsetNumber(current)));
 
 				so->mrkHeapIptr = btitem->bti_itup.t_tid;
 			}
 		}
-}
-
-/*
- *	_bt_scantouched() -- check to see if a scan is affected by a given
- *						 change to the index
- */
-static bool
-_bt_scantouched(IndexScanDesc scan, BlockNumber blkno, OffsetNumber offno)
-{
-	ItemPointer current;
-
-	current = &(scan->currentItemData);
-	if (ItemPointerIsValid(current)
-		&& ItemPointerGetBlockNumber(current) == blkno
-		&& ItemPointerGetOffsetNumber(current) >= offno)
-		return true;
-
-	current = &(scan->currentMarkData);
-	if (ItemPointerIsValid(current)
-		&& ItemPointerGetBlockNumber(current) == blkno
-		&& ItemPointerGetOffsetNumber(current) >= offno)
-		return true;
-
-	return false;
+	}
 }

src/backend/access/nbtree/nbtsearch.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.41 1999/02/21 03:48:27 scrappy Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.42 1999/03/28 20:31:58 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -706,15 +706,7 @@ _bt_next(IndexScanDesc scan, ScanDirection dir)
 	so = (BTScanOpaque) scan->opaque;
 	current = &(scan->currentItemData);
 
-	/*
-	 * XXX 10 may 91:  somewhere there's a bug in our management of the
-	 * cached buffer for this scan.  wei discovered it.  the following is
-	 * a workaround so he can work until i figure out what's going on.
-	 */
-
-	if (!BufferIsValid(so->btso_curbuf))
-		so->btso_curbuf = _bt_getbuf(rel, ItemPointerGetBlockNumber(current),
-									 BT_READ);
+	Assert (BufferIsValid(so->btso_curbuf));
 
 	/* we still have the buffer pinned and locked */
 	buf = so->btso_curbuf;
@@ -1069,7 +1061,11 @@ _bt_step(IndexScanDesc scan, Buffer *bufP, ScanDirection dir)
 
 	rel = scan->relation;
 	current = &(scan->currentItemData);
-	offnum = ItemPointerGetOffsetNumber(current);
+	/*
+	 * Don't use ItemPointerGetOffsetNumber or you risk to get
+	 * assertion due to ability of ip_posid to be equal 0.
+	 */
+	offnum = current->ip_posid;
 	page = BufferGetPage(*bufP);
 	opaque = (BTPageOpaque) PageGetSpecialPointer(page);
 	so = (BTScanOpaque) scan->opaque;

src/backend/access/transam/xact.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.32 1999/02/13 23:14:49 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.33 1999/03/28 20:31:59 vadim Exp $
  *
  * NOTES
  *		Transaction aborts can now occur two ways:
@@ -933,7 +933,10 @@ CommitTransaction()
 	 * 11/26/96
 	 */
 	if (MyProc != (PROC *) NULL)
+	{
 		MyProc->xid = InvalidTransactionId;
+		MyProc->xmin = InvalidTransactionId;
+	}
 }
 
 /* --------------------------------
@@ -951,7 +954,10 @@ AbortTransaction()
 	 * 11/26/96
 	 */
 	if (MyProc != (PROC *) NULL)
+	{
 		MyProc->xid = InvalidTransactionId;
+		MyProc->xmin = InvalidTransactionId;
+	}
 
 	/* ----------------
 	 *	check the current transaction state

src/backend/storage/buffer/bufmgr.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.49 1999/02/21 03:49:21 scrappy Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.50 1999/03/28 20:32:17 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -95,7 +95,7 @@ static BufferDesc *BufferAlloc(Relation reln, BlockNumber blockNum,
 static int	FlushBuffer(Buffer buffer, bool release);
 static void BufferSync(void);
 static int	BufferReplace(BufferDesc *bufHdr, bool bufferLockHeld);
-static void PrintBufferDescs(void);
+void PrintBufferDescs(void);
 
 /* not static but used by vacuum only ... */
 int			BlowawayRelationBuffers(Relation rel, BlockNumber block);
@@ -1208,23 +1208,24 @@ int
 BufferPoolCheckLeak()
 {
 	int			i;
-	int			error = 0;
+	int			result = 0;
 
 	for (i = 1; i <= NBuffers; i++)
 	{
 		if (BufferIsValid(i))
 		{
+			BufferDesc *buf = &(BufferDescriptors[i - 1]);
+
 			elog(NOTICE,
-			"buffer leak [%d] detected in BufferPoolCheckLeak()", i - 1);
-			error = 1;
+			"Buffer Leak: [%03d] (freeNext=%d, freePrev=%d, \
+relname=%s, blockNum=%d, flags=0x%x, refcount=%d %d)",
+				 i - 1, buf->freeNext, buf->freePrev,
+				 buf->sb_relname, buf->tag.blockNum, buf->flags,
+				 buf->refcount, PrivateRefCount[i - 1]);
+			result = 1;
 		}
 	}
-	if (error)
-	{
-		PrintBufferDescs();
-		return 1;
-	}
-	return 0;
+	return (result);
 }
 
 /* ------------------------------------------------
@@ -1465,7 +1466,7 @@ DropBuffers(Oid dbid)
  *		use only.
  * -----------------------------------------------------------------
  */
-static void
+void
 PrintBufferDescs()
 {
 	int			i;
@@ -1474,16 +1475,14 @@ PrintBufferDescs()
 	if (IsUnderPostmaster)
 	{
 		SpinAcquire(BufMgrLock);
-#ifdef NOT_USED
 		for (i = 0; i < NBuffers; ++i, ++buf)
 		{
-			elog(NOTICE, "[%02d] (freeNext=%d, freePrev=%d, relname=%s, \
+			elog(DEBUG, "[%02d] (freeNext=%d, freePrev=%d, relname=%s, \
 blockNum=%d, flags=0x%x, refcount=%d %d)",
 				 i, buf->freeNext, buf->freePrev,
 				 buf->sb_relname, buf->tag.blockNum, buf->flags,
 				 buf->refcount, PrivateRefCount[i]);
 		}
-#endif
 		SpinRelease(BufMgrLock);
 	}
 	else

src/backend/storage/ipc/shmem.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/shmem.c,v 1.37 1999/02/22 06:16:48 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/shmem.c,v 1.38 1999/03/28 20:32:22 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -640,6 +640,7 @@ GetSnapshotData(bool serializable)
 	ShmemIndexEnt  *result;
 	PROC		   *proc;
 	TransactionId	cid = GetCurrentTransactionId();
+	TransactionId	xid;
 	uint32			count = 0;
 	uint32			have = 32;
 
@@ -669,19 +670,20 @@ GetSnapshotData(bool serializable)
 			strncmp(result->key, "PID ", 4) != 0)
 			continue;
 		proc = (PROC *) MAKE_PTR(result->location);
-		if (proc == MyProc || proc->xid < FirstTransactionId)
+		xid = proc->xid;	/* we don't use spin-locking in xact.c ! */
+		if (proc == MyProc || xid < FirstTransactionId)
 			continue;
-		if (proc->xid < snapshot->xmin)
-			snapshot->xmin = proc->xid;
-		else if (proc->xid > snapshot->xmax)
-			snapshot->xmax = proc->xid;
+		if (xid < snapshot->xmin)
+			snapshot->xmin = xid;
+		else if (xid > snapshot->xmax)
+			snapshot->xmax = xid;
 		if (have == 0)
 		{
 			snapshot->xip = (TransactionId *) realloc(snapshot->xip,
 								   (count + 32) * sizeof(TransactionId));
 			have = 32;
 		}
-		snapshot->xip[count] = proc->xid;
+		snapshot->xip[count] = xid;
 		have--;
 		count++;
 	}
@@ -692,3 +694,48 @@ GetSnapshotData(bool serializable)
 	elog(ERROR, "GetSnapshotData: ShmemIndex corrupted");
 	return NULL;
 }
+
+/*
+ * GetXmaxRecent -- returns oldest transaction that was running
+ * 					when all current transaction was started.
+ * 					It's used by vacuum to decide what deleted
+ * 					tuples must be preserved in a table.
+ *
+ * And yet another strange func for this place...	- vadim 03/18/99
+ */
+void
+GetXmaxRecent(TransactionId *XmaxRecent)
+{
+	ShmemIndexEnt  *result;
+	PROC		   *proc;
+	TransactionId	xmin;
+
+	Assert(ShmemIndex);
+
+	ReadNewTransactionId(XmaxRecent);
+
+	SpinAcquire(ShmemIndexLock);
+
+	hash_seq((HTAB *) NULL);
+	while ((result = (ShmemIndexEnt *) hash_seq(ShmemIndex)) != NULL)
+	{
+		if (result == (ShmemIndexEnt *) TRUE)
+		{
+			SpinRelease(ShmemIndexLock);
+			return;
+		}
+		if (result->location == INVALID_OFFSET ||
+			strncmp(result->key, "PID ", 4) != 0)
+			continue;
+		proc = (PROC *) MAKE_PTR(result->location);
+		xmin = proc->xmin;	/* we don't use spin-locking in xact.c ! */
+		if (proc == MyProc || xmin < FirstTransactionId)
+			continue;
+		if (xmin < *XmaxRecent)
+			*XmaxRecent = xmin;
+	}
+
+	SpinRelease(ShmemIndexLock);
+	elog(ERROR, "GetXmaxRecent: ShmemIndex corrupted");
+	return NULL;
+}

src/backend/storage/lmgr/multi.c

@@ -12,7 +12,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/Attic/multi.c,v 1.27 1999/02/13 23:18:27 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/Attic/multi.c,v 1.28 1999/03/28 20:32:25 vadim Exp $
  *
  * NOTES:
  *	 (1) The lock.c module assumes that the caller here is doing
@@ -34,55 +34,6 @@ static bool MultiAcquire(LOCKMETHOD lockmethod, LOCKTAG *tag,
 static bool MultiRelease(LOCKMETHOD lockmethod, LOCKTAG *tag,
 			 LOCKMODE lockmode, PG_LOCK_LEVEL level);
 
-#ifdef LowLevelLocking
-
-static MASK MultiConflicts[] = {
-	(int) NULL,
-
-/* RowShareLock */
-	(1 << ExclusiveLock),
-
-/* RowExclusiveLock */
-	(1 << ExclusiveLock) | (1 << ShareRowExclusiveLock) | (1 << ShareLock),
-
-/* ShareLock */
-	(1 << ExclusiveLock) | (1 << ShareRowExclusiveLock) |
-	(1 << RowExclusiveLock),
-
-/* ShareRowExclusiveLock */
-	(1 << ExclusiveLock) | (1 << ShareRowExclusiveLock) |
-	(1 << ShareLock) | (1 << RowExclusiveLock),
-
-/* ExclusiveLock */
-	(1 << ExclusiveLock) | (1 << ShareRowExclusiveLock) | (1 << ShareLock) |
-	(1 << RowExclusiveLock) | (1 << RowShareLock),
-
-/* ObjShareLock */
-	(1 << ObjExclusiveLock),
-
-/* ObjExclusiveLock */
-	(1 << ObjExclusiveLock) | (1 << ObjShareLock),
-
-/* ExtendLock */
-	(1 << ExtendLock)
-
-};
-
-/*
- * write locks have higher priority than read locks and extend locks.  May
- * want to treat INTENT locks differently.
- */
-static int	MultiPrios[] = {
-	(int) NULL,
-	2,
-	1,
-	2,
-	1,
-	1
-};
-
-#else
-
 /*
  * INTENT indicates to higher level that a lower level lock has been
  * set.  For example, a write lock on a tuple conflicts with a write
@@ -121,8 +72,6 @@ static int	MultiPrios[] = {
 	1
 };
 
-#endif	 /* !LowLevelLocking */
-
 /*
  * Lock table identifier for this lock table.  The multi-level
  * lock table is ONE lock table, not three.

src/backend/storage/lmgr/proc.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.51 1999/02/21 01:41:45 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.52 1999/03/28 20:32:26 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -46,7 +46,7 @@
  *		This is so that we can support more backends. (system-wide semaphore
  *		sets run out pretty fast.)				  -ay 4/95
  *
- * $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.51 1999/02/21 01:41:45 tgl Exp $
+ * $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.52 1999/03/28 20:32:26 vadim Exp $
  */
 #include <sys/time.h>
 #include <unistd.h>
@@ -300,9 +300,7 @@ InitProcess(IPCKey key)
 
 	MyProc->pid = MyProcPid;
 	MyProc->xid = InvalidTransactionId;
-#ifdef LowLevelLocking
 	MyProc->xmin = InvalidTransactionId;
-#endif
 
 	/* ----------------
 	 * Start keeping spin lock stats from here on.	Any botch before

src/backend/utils/time/tqual.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/time/tqual.c,v 1.25 1999/02/13 23:20:19 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/time/tqual.c,v 1.26 1999/03/28 20:32:29 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -86,11 +86,26 @@ HeapTupleSatisfiesItself(HeapTupleHeader tuple)
 
 	if (!(tuple->t_infomask & HEAP_XMIN_COMMITTED))
 	{
-		if (tuple->t_infomask & HEAP_XMIN_INVALID)		/* xid invalid or
-														 * aborted */
+		if (tuple->t_infomask & HEAP_XMIN_INVALID)
 			return false;
 
-		if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
+		if (tuple->t_infomask & HEAP_MOVED_OFF)
+		{
+			if (TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (tuple->t_infomask & HEAP_MOVED_IN)
+		{
+			if (!TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
 		{
 			if (tuple->t_infomask & HEAP_XMAX_INVALID)	/* xid invalid */
 				return true;
@@ -98,14 +113,12 @@ HeapTupleSatisfiesItself(HeapTupleHeader tuple)
 				return true;
 			return false;
 		}
-
-		if (!TransactionIdDidCommit(tuple->t_xmin))
+		else if (!TransactionIdDidCommit(tuple->t_xmin))
 		{
 			if (TransactionIdDidAbort(tuple->t_xmin))
 				tuple->t_infomask |= HEAP_XMIN_INVALID; /* aborted */
 			return false;
 		}
-
 		tuple->t_infomask |= HEAP_XMIN_COMMITTED;
 	}
 	/* the tuple was inserted validly */
@@ -200,11 +213,26 @@ HeapTupleSatisfiesNow(HeapTupleHeader tuple)
 
 	if (!(tuple->t_infomask & HEAP_XMIN_COMMITTED))
 	{
-		if (tuple->t_infomask & HEAP_XMIN_INVALID)		/* xid invalid or
-														 * aborted */
+		if (tuple->t_infomask & HEAP_XMIN_INVALID)
 			return false;
 
-		if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
+		if (tuple->t_infomask & HEAP_MOVED_OFF)
+		{
+			if (TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (tuple->t_infomask & HEAP_MOVED_IN)
+		{
+			if (!TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
 		{
 			if (CommandIdGEScanCommandId(tuple->t_cmin))
 				return false;	/* inserted after scan started */
@@ -222,18 +250,12 @@ HeapTupleSatisfiesNow(HeapTupleHeader tuple)
 			else
 				return false;	/* deleted before scan started */
 		}
-
-		/*
-		 * this call is VERY expensive - requires a log table lookup.
-		 */
-
-		if (!TransactionIdDidCommit(tuple->t_xmin))
+		else if (!TransactionIdDidCommit(tuple->t_xmin))
 		{
 			if (TransactionIdDidAbort(tuple->t_xmin))
 				tuple->t_infomask |= HEAP_XMIN_INVALID; /* aborted */
 			return false;
 		}
-
 		tuple->t_infomask |= HEAP_XMIN_COMMITTED;
 	}
 
@@ -288,7 +310,23 @@ HeapTupleSatisfiesUpdate(HeapTuple tuple)
 		if (th->t_infomask & HEAP_XMIN_INVALID)	/* xid invalid or aborted */
 			return HeapTupleInvisible;
 
-		if (TransactionIdIsCurrentTransactionId(th->t_xmin))
+		if (th->t_infomask & HEAP_MOVED_OFF)
+		{
+			if (TransactionIdDidCommit((TransactionId)th->t_cmin))
+			{
+				th->t_infomask |= HEAP_XMIN_INVALID;
+				return HeapTupleInvisible;
+			}
+		}
+		else if (th->t_infomask & HEAP_MOVED_IN)
+		{
+			if (!TransactionIdDidCommit((TransactionId)th->t_cmin))
+			{
+				th->t_infomask |= HEAP_XMIN_INVALID;
+				return HeapTupleInvisible;
+			}
+		}
+		else if (TransactionIdIsCurrentTransactionId(th->t_xmin))
 		{
 			if (CommandIdGEScanCommandId(th->t_cmin) && !heapisoverride())
 				return HeapTupleInvisible;	/* inserted after scan started */
@@ -306,19 +344,12 @@ HeapTupleSatisfiesUpdate(HeapTuple tuple)
 			else
 				return HeapTupleInvisible;	/* updated before scan started */
 		}
-
-		/*
-		 * This call is VERY expensive - requires a log table lookup.
-		 * Actually, this should be done by query before...
-		 */
-
-		if (!TransactionIdDidCommit(th->t_xmin))
+		else if (!TransactionIdDidCommit(th->t_xmin))
 		{
 			if (TransactionIdDidAbort(th->t_xmin))
 				th->t_infomask |= HEAP_XMIN_INVALID; /* aborted */
 			return HeapTupleInvisible;
 		}
-
 		th->t_infomask |= HEAP_XMIN_COMMITTED;
 	}
 
@@ -375,10 +406,26 @@ HeapTupleSatisfiesDirty(HeapTupleHeader tuple)
 
 	if (!(tuple->t_infomask & HEAP_XMIN_COMMITTED))
 	{
-		if (tuple->t_infomask & HEAP_XMIN_INVALID)	/* xid invalid or aborted */
+		if (tuple->t_infomask & HEAP_XMIN_INVALID)
 			return false;
 
-		if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
+		if (tuple->t_infomask & HEAP_MOVED_OFF)
+		{
+			if (TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (tuple->t_infomask & HEAP_MOVED_IN)
+		{
+			if (!TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
 		{
 			if (tuple->t_infomask & HEAP_XMAX_INVALID)	/* xid invalid */
 				return true;
@@ -390,8 +437,7 @@ HeapTupleSatisfiesDirty(HeapTupleHeader tuple)
 
 			return false;
 		}
-
-		if (!TransactionIdDidCommit(tuple->t_xmin))
+		else if (!TransactionIdDidCommit(tuple->t_xmin))
 		{
 			if (TransactionIdDidAbort(tuple->t_xmin))
 			{
@@ -401,7 +447,6 @@ HeapTupleSatisfiesDirty(HeapTupleHeader tuple)
 			SnapshotDirty->xmin = tuple->t_xmin;
 			return true;						/* in insertion by other */
 		}
-
 		tuple->t_infomask |= HEAP_XMIN_COMMITTED;
 	}
 
@@ -451,11 +496,26 @@ HeapTupleSatisfiesSnapshot(HeapTupleHeader tuple, Snapshot snapshot)
 
 	if (!(tuple->t_infomask & HEAP_XMIN_COMMITTED))
 	{
-		if (tuple->t_infomask & HEAP_XMIN_INVALID)		/* xid invalid or
-														 * aborted */
+		if (tuple->t_infomask & HEAP_XMIN_INVALID)
 			return false;
 
-		if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
+		if (tuple->t_infomask & HEAP_MOVED_OFF)
+		{
+			if (TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (tuple->t_infomask & HEAP_MOVED_IN)
+		{
+			if (!TransactionIdDidCommit((TransactionId)tuple->t_cmin))
+			{
+				tuple->t_infomask |= HEAP_XMIN_INVALID;
+				return false;
+			}
+		}
+		else if (TransactionIdIsCurrentTransactionId(tuple->t_xmin))
 		{
 			if (CommandIdGEScanCommandId(tuple->t_cmin))
 				return false;	/* inserted after scan started */
@@ -473,18 +533,12 @@ HeapTupleSatisfiesSnapshot(HeapTupleHeader tuple, Snapshot snapshot)
 			else
 				return false;	/* deleted before scan started */
 		}
-
-		/*
-		 * this call is VERY expensive - requires a log table lookup.
-		 */
-
-		if (!TransactionIdDidCommit(tuple->t_xmin))
+		else if (!TransactionIdDidCommit(tuple->t_xmin))
 		{
 			if (TransactionIdDidAbort(tuple->t_xmin))
 				tuple->t_infomask |= HEAP_XMIN_INVALID; /* aborted */
 			return false;
 		}
-
 		tuple->t_infomask |= HEAP_XMIN_COMMITTED;
 	}
 

src/include/access/htup.h

@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: htup.h,v 1.13 1999/02/13 23:20:54 momjian Exp $
+ * $Id: htup.h,v 1.14 1999/03/28 20:32:30 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -117,8 +117,11 @@ typedef HeapTupleData *HeapTuple;
 #define HEAP_XMAX_COMMITTED		0x0400	/* t_xmax committed */
 #define HEAP_XMAX_INVALID		0x0800	/* t_xmax invalid/aborted */
 #define HEAP_MARKED_FOR_UPDATE	0x1000	/* marked for UPDATE */
+#define HEAP_UPDATED			0x2000	/* this is UPDATEd version of row */
+#define HEAP_MOVED_OFF			0x4000	/* removed or moved to another place by vacuum */
+#define HEAP_MOVED_IN			0x8000	/* moved from another place by vacuum */
 
-#define HEAP_XACT_MASK			0x0F00	/* */
+#define HEAP_XACT_MASK			0xFF00	/* */
 
 #define HeapTupleNoNulls(tuple) \
 		(!(((HeapTuple) (tuple))->t_data->t_infomask & HEAP_HASNULL))

src/include/access/nbtree.h

@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: nbtree.h,v 1.23 1999/02/13 23:20:55 momjian Exp $
+ * $Id: nbtree.h,v 1.24 1999/03/28 20:32:34 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -42,6 +42,7 @@ typedef struct BTPageOpaqueData
 {
 	BlockNumber btpo_prev;
 	BlockNumber btpo_next;
+	BlockNumber	btpo_parent;
 	uint16		btpo_flags;
 
 #define BTP_LEAF		(1 << 0)
@@ -176,13 +177,6 @@ typedef struct BTPageState
 #define BT_INSERTION	0
 #define BT_DESCENT		1
 
-/*
- *	We must classify index modification types for the benefit of
- *	_bt_adjscans.
- */
-#define BT_INSERT		0
-#define BT_DELETE		1
-
 /*
  *	In general, the btree code tries to localize its knowledge about
  *	page layout to a couple of routines.  However, we need a special
@@ -268,7 +262,7 @@ extern void btdelete(Relation rel, ItemPointer tid);
  */
 extern void _bt_regscan(IndexScanDesc scan);
 extern void _bt_dropscan(IndexScanDesc scan);
-extern void _bt_adjscans(Relation rel, ItemPointer tid, int op);
+extern void _bt_adjscans(Relation rel, ItemPointer tid);
 
 /*
  * prototypes for functions in nbtsearch.c

src/include/commands/vacuum.h

@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: vacuum.h,v 1.17 1999/02/13 23:21:20 momjian Exp $
+ * $Id: vacuum.h,v 1.18 1999/03/28 20:32:38 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -95,6 +95,23 @@ typedef struct VRelListData
 
 typedef VRelListData *VRelList;
 
+typedef struct VTupleLinkData
+{
+	ItemPointerData		new_tid;
+	ItemPointerData		this_tid;
+} VTupleLinkData;
+
+typedef VTupleLinkData *VTupleLink;
+
+typedef struct VTupleMoveData
+{
+	ItemPointerData		tid;		/* tuple ID */
+	VPageDescr			vpd;		/* where to move */
+	bool				cleanVpd;	/* clean vpd before using */
+} VTupleMoveData;
+
+typedef VTupleMoveData *VTupleMove;
+
 typedef struct VRelStats
 {
 	Oid				relid;
@@ -105,6 +122,8 @@ typedef struct VRelStats
 	bool			hasindex;
 	int				va_natts;		/* number of attrs being analyzed */
 	VacAttrStats   *vacattrstats;
+	int				num_vtlinks;
+	VTupleLink		vtlinks;
 } VRelStats;
 
 extern bool VacuumRunning;