Index: Include/objimpl.h
===================================================================
--- Include/objimpl.h	(revision 74649)
+++ Include/objimpl.h	(working copy)
@@ -108,6 +108,13 @@
 PyAPI_FUNC(void) _PyObject_DebugDumpAddress(const void *p);
 PyAPI_FUNC(void) _PyObject_DebugCheckAddress(const void *p);
 PyAPI_FUNC(void) _PyObject_DebugMallocStats(void);
+PyAPI_FUNC(void *) _PyObject_DebugMallocId(int id, size_t nbytes);
+PyAPI_FUNC(void *) _PyObject_DebugReallocId(int id, void *p, size_t nbytes);
+PyAPI_FUNC(void) _PyObject_DebugFreeId(int id, void *p);
+PyAPI_FUNC(void) _PyObject_DebugCheckAddressId(int id, const void *p);
+PyAPI_FUNC(void *) _PyMem_DebugMalloc(size_t nbytes);
+PyAPI_FUNC(void *) _PyMem_DebugRealloc(void *p, size_t nbytes);
+PyAPI_FUNC(void) _PyMem_DebugFree(void *p);
 #define PyObject_MALLOC		_PyObject_DebugMalloc
 #define PyObject_Malloc		_PyObject_DebugMalloc
 #define PyObject_REALLOC	_PyObject_DebugRealloc
Index: Include/pymem.h
===================================================================
--- Include/pymem.h	(revision 74649)
+++ Include/pymem.h	(working copy)
@@ -59,9 +59,9 @@
 /* Macros. */
 #ifdef PYMALLOC_DEBUG
 /* Redirect all memory operations to Python's debugging allocator. */
-#define PyMem_MALLOC		PyObject_MALLOC
-#define PyMem_REALLOC		PyObject_REALLOC
-#define PyMem_FREE		PyObject_FREE
+#define PyMem_MALLOC		_PyMem_DebugMalloc
+#define PyMem_REALLOC		_PyMem_DebugRealloc
+#define PyMem_FREE		_PyMem_DebugFree
 
 #else	/* ! PYMALLOC_DEBUG */
 
Index: Objects/obmalloc.c
===================================================================
--- Objects/obmalloc.c	(revision 74649)
+++ Objects/obmalloc.c	(working copy)
@@ -1254,6 +1254,26 @@
 
 #define SST SIZEOF_SIZE_T
 
+/* merge in a single byte id to the high byte of the serial no and return */
+static size_t
+idserialno(int id)
+{
+
+	int shift = (SST-1)*8;
+	size_t mask = ~(((size_t)0xff) << shift);
+	return (serialno & mask) | ((size_t)(id & 0xff) << shift);
+}
+static int
+idfromserialno(size_t *serial)
+{
+	size_t id;
+	int shift = (SST-1)*8;
+	size_t mask = ~(((size_t)0xff) << shift);
+	id = ((*serial) & (~mask))>>shift;
+	*serial &= mask;
+	return (int)id;
+}
+
 /* Read sizeof(size_t) bytes at p as a big-endian size_t. */
 static size_t
 read_size_t(const void *p)
@@ -1331,9 +1351,50 @@
     instant at which this block was passed out.
 */
 
+
+/* wrap PyMem api with id 0 */
 void *
+_PyMem_DebugMalloc(size_t nbytes)
+{
+	return _PyObject_DebugMallocId(0, nbytes);
+}
+void *
+_PyMem_DebugRealloc(void *p, size_t nbytes)
+{
+	return _PyObject_DebugReallocId(0, p, nbytes);
+}
+void
+_PyMem_DebugFree(void *p)
+{
+	_PyObject_DebugFreeId(0, p);
+}
+
+/* wrap PyObject api with id 1 */
+void *
 _PyObject_DebugMalloc(size_t nbytes)
 {
+	return _PyObject_DebugMallocId(1, nbytes);
+}
+void *
+_PyObject_DebugRealloc(void *p, size_t nbytes)
+{
+	return _PyObject_DebugReallocId(1, p, nbytes);
+}
+void
+_PyObject_DebugFree(void *p)
+{
+	_PyObject_DebugFreeId(1, p);
+}
+void
+_PyObject_DebugCheckAddress(void *p)
+{
+	_PyObject_DebugCheckAddressId(1, p);
+}
+
+
+void *
+_PyObject_DebugMallocId(int id, size_t nbytes)
+{
 	uchar *p;	/* base address of malloc'ed block */
 	uchar *tail;	/* p + 2*SST + nbytes == pointer to tail pad bytes */
 	size_t total;	/* nbytes + 4*SST */
@@ -1356,7 +1417,7 @@
 
 	tail = p + 2*SST + nbytes;
 	memset(tail, FORBIDDENBYTE, SST);
-	write_size_t(tail + SST, serialno);
+	write_size_t(tail + SST, idserialno(id));
 
 	return p + 2*SST;
 }
@@ -1367,14 +1428,14 @@
    Then calls the underlying free.
 */
 void
-_PyObject_DebugFree(void *p)
+_PyObject_DebugFreeId(int id, void *p)
 {
 	uchar *q = (uchar *)p - 2*SST;  /* address returned from malloc */
 	size_t nbytes;
 
 	if (p == NULL)
 		return;
-	_PyObject_DebugCheckAddress(p);
+	_PyObject_DebugCheckAddressId(id, p);
 	nbytes = read_size_t(q);
 	if (nbytes > 0)
 		memset(q, DEADBYTE, nbytes);
@@ -1382,7 +1443,7 @@
 }
 
 void *
-_PyObject_DebugRealloc(void *p, size_t nbytes)
+_PyObject_DebugReallocId(int id, void *p, size_t nbytes)
 {
 	uchar *q = (uchar *)p;
 	uchar *tail;
@@ -1391,9 +1452,9 @@
 	int i;
 
 	if (p == NULL)
-		return _PyObject_DebugMalloc(nbytes);
+		return _PyObject_DebugMallocId(id, nbytes);
 
-	_PyObject_DebugCheckAddress(p);
+	_PyObject_DebugCheckAddressId(id, p);
 	bumpserialno();
 	original_nbytes = read_size_t(q - 2*SST);
 	total = nbytes + 4*SST;
@@ -1417,7 +1478,7 @@
 	q += 2*SST;
 	tail = q + nbytes;
 	memset(tail, FORBIDDENBYTE, SST);
-	write_size_t(tail + SST, serialno);
+	write_size_t(tail + SST, idserialno(id));
 
 	if (nbytes > original_nbytes) {
 		/* growing:  mark new extra memory clean */
@@ -1433,13 +1494,16 @@
  * and call Py_FatalError to kill the program.
  */
  void
-_PyObject_DebugCheckAddress(const void *p)
+_PyObject_DebugCheckAddressId(int id, const void *p)
 {
 	const uchar *q = (const uchar *)p;
+	char msgbuf[64];
 	char *msg;
 	size_t nbytes;
 	const uchar *tail;
 	int i;
+	size_t serial;
+	int id0;
 
 	if (p == NULL) {
 		msg = "didn't expect a NULL pointer";
@@ -1466,6 +1530,15 @@
 		}
 	}
 
+	serial = read_size_t(tail+SST);
+	id0 = idfromserialno(&serial);
+	if (id0 != id) {
+		msg = msgbuf;
+		snprintf(msg, sizeof(msgbuf), "bad ID: Allocated using %d, verified using %d", id0, id);
+		msgbuf[sizeof(msgbuf)-1] = 0;
+		goto error;
+	}
+
 	return;
 
 error:
@@ -1482,6 +1555,7 @@
 	size_t nbytes, serial;
 	int i;
 	int ok;
+	int id;
 
 	fprintf(stderr, "Debug memory block at address p=%p:\n", p);
 	if (p == NULL)
@@ -1544,6 +1618,8 @@
 	}
 
 	serial = read_size_t(tail + SST);
+	id = idfromserialno(&serial);
+	fprintf(stderr, "the block was allocated using id %d.\n", id);
 	fprintf(stderr, "    The block was made by call #%" PY_FORMAT_SIZE_T
 			"u to debug malloc/realloc.\n", serial);