bpo-42739: Don't use sentinels to mark end of line table. (GH-25657)

* Add length parameter to PyLineTable_InitAddressRange and doen't use sentinel values at end of table. Makes the line number table more robust.

* Update PyCodeAddressRange to match PEP 626.

Include/cpython/code.h

@@ -135,12 +135,17 @@ PyCode_NewEmpty(const char *filename, const char *funcname, int firstlineno);
 PyAPI_FUNC(int) PyCode_Addr2Line(PyCodeObject *, int);
 
 /* for internal use only */
+struct _opaque {
+    int computed_line;
+    char *lo_next;
+    char *limit;
+};
+
 typedef struct _line_offsets {
     int ar_start;
     int ar_end;
     int ar_line;
-    int ar_computed_line;
+    struct _opaque opaque;
-    char *lo_next;
 } PyCodeAddressRange;
 
 /* Update *bounds to describe the first and one-past-the-last instructions in the
@@ -170,7 +175,7 @@ PyAPI_FUNC(int) _PyCode_SetExtra(PyObject *code, Py_ssize_t index,
 int _PyCode_InitAddressRange(PyCodeObject* co, PyCodeAddressRange *bounds);
 
 /** Out of process API for initializing the line number table. */
-void PyLineTable_InitAddressRange(char *linetable, int firstlineno, PyCodeAddressRange *range);
+void PyLineTable_InitAddressRange(char *linetable, Py_ssize_t length, int firstlineno, PyCodeAddressRange *range);
 
 /** API for traversing the line number table. */
 int PyLineTable_NextAddressRange(PyCodeAddressRange *range);

Lib/ctypes/test/test_values.py

@@ -80,9 +80,9 @@ class struct_frozen(Structure):
                 continue
             items.append((entry.name.decode("ascii"), entry.size))
 
-        expected = [("__hello__", 139),
+        expected = [("__hello__", 137),
-                    ("__phello__", -139),
+                    ("__phello__", -137),
-                    ("__phello__.spam", 139),
+                    ("__phello__.spam", 137),
                     ]
         self.assertEqual(items, expected, "PyImport_FrozenModules example "
             "in Doc/library/ctypes.rst may be out of date")

Lib/importlib/_bootstrap_external.py

@@ -350,6 +350,7 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.10a7 3435 Use instruction offsets (as opposed to byte offsets).
 #     Python 3.10b1 3436 (Add GEN_START bytecode #43683)
 #     Python 3.10b1 3437 (Undo making 'annotations' future by default - We like to dance among core devs!)
+#     Python 3.10b1 3438 Safer line number table handling.
 
 #
 # MAGIC must change whenever the bytecode emitted by the compiler may no
@@ -359,7 +360,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3437).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3438).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'

Lib/test/test_code.py

@@ -264,6 +264,12 @@ def func2():
                 new_code = code.replace(**{attr: value})
                 self.assertEqual(getattr(new_code, attr), value)
 
+    def test_empty_linetable(self):
+        def func():
+            pass
+        new_code = code = func.__code__.replace(co_linetable=b'')
+        self.assertEqual(list(new_code.co_lines()), [])
+
 
 def isinterned(s):
     return s is sys.intern(('_' + s + '_')[1:-1])

Lib/test/test_dis.py

@@ -172,7 +172,7 @@ def bug42562():
 
 
 # Set line number for 'pass' to None
-bug42562.__code__ = bug42562.__code__.replace(co_linetable=b'\x04\x80\xff\x80')
+bug42562.__code__ = bug42562.__code__.replace(co_linetable=b'\x04\x80')
 
 
 dis_bug42562 = """\

Misc/NEWS.d/next/Core and Builtins/2021-04-27-10-59-10.bpo-42739.PrVkKM.rst

@@ -0,0 +1,4 @@
+The internal representation of line number tables is changed to not use
+sentinels, and an explicit length parameter is added to the out of process
+API function ``PyLineTable_InitAddressRange``. This makes the handling of
+line number tables more robust in some circumstances.

Objects/codeobject.c

@@ -456,15 +456,15 @@ code_getlnotab(PyCodeObject *code, void *closure)
     }
     _PyCode_InitAddressRange(code, &bounds);
     while (PyLineTable_NextAddressRange(&bounds)) {
-        if (bounds.ar_computed_line != line) {
+        if (bounds.opaque.computed_line != line) {
             int bdelta = bounds.ar_start - code_offset;
-            int ldelta = bounds.ar_computed_line - line;
+            int ldelta = bounds.opaque.computed_line - line;
             if (!emit_delta(&bytes, bdelta, ldelta, &table_offset)) {
                 Py_DECREF(bytes);
                 return NULL;
             }
             code_offset = bounds.ar_start;
-            line = bounds.ar_computed_line;
+            line = bounds.opaque.computed_line;
         }
     }
     _PyBytes_Resize(&bytes, table_offset);
@@ -1120,20 +1120,20 @@ code_linesiterator(PyCodeObject *code, PyObject *Py_UNUSED(args))
 static void
 retreat(PyCodeAddressRange *bounds)
 {
-    int ldelta = ((signed char *)bounds->lo_next)[-1];
+    int ldelta = ((signed char *)bounds->opaque.lo_next)[-1];
     if (ldelta == -128) {
         ldelta = 0;
     }
-    bounds->ar_computed_line -= ldelta;
+    bounds->opaque.computed_line -= ldelta;
-    bounds->lo_next -= 2;
+    bounds->opaque.lo_next -= 2;
     bounds->ar_end = bounds->ar_start;
-    bounds->ar_start -= ((unsigned char *)bounds->lo_next)[-2];
+    bounds->ar_start -= ((unsigned char *)bounds->opaque.lo_next)[-2];
-    ldelta = ((signed char *)bounds->lo_next)[-1];
+    ldelta = ((signed char *)bounds->opaque.lo_next)[-1];
     if (ldelta == -128) {
         bounds->ar_line = -1;
     }
     else {
-        bounds->ar_line = bounds->ar_computed_line;
+        bounds->ar_line = bounds->opaque.computed_line;
     }
 }
 
@@ -1141,23 +1141,22 @@ static void
 advance(PyCodeAddressRange *bounds)
 {
     bounds->ar_start = bounds->ar_end;
-    int delta = ((unsigned char *)bounds->lo_next)[0];
+    int delta = ((unsigned char *)bounds->opaque.lo_next)[0];
-    assert (delta < 255);
     bounds->ar_end += delta;
-    int ldelta = ((signed char *)bounds->lo_next)[1];
+    int ldelta = ((signed char *)bounds->opaque.lo_next)[1];
-    bounds->lo_next += 2;
+    bounds->opaque.lo_next += 2;
     if (ldelta == -128) {
         bounds->ar_line = -1;
     }
     else {
-        bounds->ar_computed_line += ldelta;
+        bounds->opaque.computed_line += ldelta;
-        bounds->ar_line = bounds->ar_computed_line;
+        bounds->ar_line = bounds->opaque.computed_line;
     }
 }
 
 static inline int
 at_end(PyCodeAddressRange *bounds) {
-    return ((unsigned char *)bounds->lo_next)[0] == 255;
+    return bounds->opaque.lo_next >= bounds->opaque.limit;
 }
 
 int
@@ -1256,12 +1255,13 @@ PyCode_Addr2Line(PyCodeObject *co, int addrq)
 }
 
 void
-PyLineTable_InitAddressRange(char *linetable, int firstlineno, PyCodeAddressRange *range)
+PyLineTable_InitAddressRange(char *linetable, Py_ssize_t length, int firstlineno, PyCodeAddressRange *range)
 {
-    range->lo_next = linetable;
+    range->opaque.lo_next = linetable;
+    range->opaque.limit = range->opaque.lo_next + length;
     range->ar_start = -1;
     range->ar_end = 0;
-    range->ar_computed_line = firstlineno;
+    range->opaque.computed_line = firstlineno;
     range->ar_line = -1;
 }
 
@@ -1269,7 +1269,8 @@ int
 _PyCode_InitAddressRange(PyCodeObject* co, PyCodeAddressRange *bounds)
 {
     char *linetable = PyBytes_AS_STRING(co->co_linetable);
-    PyLineTable_InitAddressRange(linetable, co->co_firstlineno, bounds);
+    Py_ssize_t length = PyBytes_GET_SIZE(co->co_linetable);
+    PyLineTable_InitAddressRange(linetable, length, co->co_firstlineno, bounds);
     return bounds->ar_line;
 }
 

Objects/lnotab_notes.txt

@@ -39,7 +39,6 @@ Note that the end - start value is always positive.
 Finally, in order to fit into a single byte we need to convert start deltas to the range 0 <= delta <= 254,
 and line deltas to the range -127  <= delta <= 127.
 A line delta of -128 is used to indicate no line number.
-A start delta of 255 is used as a sentinel to mark the end of the table.
 Also note that a delta of zero indicates that there are no bytecodes in the given range,
 which means we can use an invalid line number for that range.
 
@@ -54,7 +53,6 @@ Final form:
     16              +1
     0               +127 (line 135, but the range is empty as no bytecodes are at line 135)
     4               +73
-    255 (end mark)  ---
 
 Iterating over the table.
 -------------------------
@@ -68,8 +66,6 @@ def co_lines(code):
     end = 0
     table_iter = iter(code.internal_line_table):
     for sdelta, ldelta in table_iter:
-        if sdelta == 255:
-            break
         if ldelta == 0: # No change to line number, just accumulate changes to end
             end += odelta
             continue

Python/compile.c

@@ -6959,10 +6959,6 @@ assemble(struct compiler *c, int addNone)
     if (!assemble_line_range(&a)) {
         return 0;
     }
-    /* Emit sentinel at end of line number table */
-    if (!assemble_emit_linetable_pair(&a, 255, -128)) {
-        goto error;
-    }
 
     if (_PyBytes_Resize(&a.a_lnotab, a.a_lnotab_off) < 0) {
         goto error;

Python/frozen_hello.h

@@ -8,5 +8,5 @@ const unsigned char _Py_M__hello[] = {
     5,112,114,105,110,116,169,0,114,1,0,0,0,114,1,0,
     0,0,122,14,60,102,114,111,122,101,110,32,104,101,108,108,
     111,62,218,8,60,109,111,100,117,108,101,62,1,0,0,0,
-    115,6,0,0,0,4,0,12,1,255,128,
+    115,4,0,0,0,4,0,12,1,
 };