gh-125063: marshal: Add version 5, improve documentation (GH-126829)

* Document that slices can be marshalled
* Deduplicate and organize the list of supported types
  in docs
* Organize the type code list in marshal.c, to make
  it more obvious that this is a versioned format
* Back-fill some historical info

Co-authored-by: Michael Droettboom <[email protected]>

Author: encukou

Doc/c-api/marshal.rst

@@ -13,11 +13,12 @@ binary mode.
 
 Numeric values are stored with the least significant byte first.
 
-The module supports two versions of the data format: version 0 is the
-historical version, version 1 shares interned strings in the file, and upon
-unmarshalling.  Version 2 uses a binary format for floating-point numbers.
-``Py_MARSHAL_VERSION`` indicates the current file format (currently 2).
+The module supports several versions of the data format; see
+the :py:mod:`Python module documentation <marshal>` for details.
 
+.. c:macro:: Py_MARSHAL_VERSION
+
+   The current format version. See :py:data:`marshal.version`.
 
 .. c:function:: void PyMarshal_WriteLongToFile(long value, FILE *file, int version)
 

Doc/library/marshal.rst

@@ -38,23 +38,39 @@ supports a substantially wider range of objects than marshal.
    maliciously constructed data.  Never unmarshal data received from an
    untrusted or unauthenticated source.
 
+There are functions that read/write files as well as functions operating on
+bytes-like objects.
+
 .. index:: object; code, code object
 
 Not all Python object types are supported; in general, only objects whose value
 is independent from a particular invocation of Python can be written and read by
-this module.  The following types are supported: booleans, integers, floating-point
-numbers, complex numbers, strings, bytes, bytearrays, tuples, lists, sets,
-frozensets, dictionaries, and code objects (if *allow_code* is true),
-where it should be understood that
-tuples, lists, sets, frozensets and dictionaries are only supported as long as
-the values contained therein are themselves supported.  The
-singletons :const:`None`, :const:`Ellipsis` and :exc:`StopIteration` can also be
-marshalled and unmarshalled.
-For format *version* lower than 3, recursive lists, sets and dictionaries cannot
-be written (see below).
+this module.  The following types are supported:
+
+* Numeric types: :class:`int`, :class:`bool`, :class:`float`, :class:`complex`.
+* Strings (:class:`str`) and :class:`bytes`.
+  :term:`Bytes-like objects <bytes-like object>` like :class:`bytearray` are
+  marshalled as :class:`!bytes`.
+* Containers: :class:`tuple`, :class:`list`, :class:`set`, :class:`frozenset`,
+  and (since :data:`version` 5), :class:`slice`.
+  It should be understood that these are supported only if the values contained
+  therein are themselves supported.
+  Recursive containers are supported since :data:`version` 3.
+* The singletons :const:`None`, :const:`Ellipsis` and :exc:`StopIteration`.
+* :class:`code` objects, if *allow_code* is true. See note above about
+  version dependence.
+
+.. versionchanged:: 3.4
+
+   * Added format version 3, which supports marshalling recursive lists, sets
+     and dictionaries.
+   * Added format version 4, which supports efficient representations
+     of short strings.
+
+.. versionchanged:: next
+
+   Added format version 5, which allows marshalling slices.
 
-There are functions that read/write files as well as functions operating on
-bytes-like objects.
 
 The module defines these functions:
 
@@ -140,11 +156,24 @@ In addition, the following constants are defined:
 
 .. data:: version
 
-   Indicates the format that the module uses. Version 0 is the historical
-   format, version 1 shares interned strings and version 2 uses a binary format
-   for floating-point numbers.
-   Version 3 adds support for object instancing and recursion.
-   The current version is 4.
+   Indicates the format that the module uses.
+   Version 0 is the historical first version; subsequent versions
+   add new features.
+   Generally, a new version becomes the default when it is introduced.
+
+   ======= =============== ====================================================
+   Version Available since New features
+   ======= =============== ====================================================
+   1       Python 2.4      Sharing interned strings
+   ------- --------------- ----------------------------------------------------
+   2       Python 2.5      Binary representation of floats
+   ------- --------------- ----------------------------------------------------
+   3       Python 3.4      Support for object instancing and recursion
+   ------- --------------- ----------------------------------------------------
+   4       Python 3.4      Efficient representation of short strings
+   ------- --------------- ----------------------------------------------------
+   5       Python 3.14     Support for :class:`slice` objects
+   ======= =============== ====================================================
 
 
 .. rubric:: Footnotes
@@ -154,4 +183,3 @@ In addition, the following constants are defined:
    around in a self-contained form. Strictly speaking, "to marshal" means to
    convert some data from internal to external form (in an RPC buffer for instance)
    and "unmarshalling" for the reverse process.
-

Include/marshal.h

@@ -13,7 +13,7 @@ PyAPI_FUNC(PyObject *) PyMarshal_ReadObjectFromString(const char *,
                                                       Py_ssize_t);
 PyAPI_FUNC(PyObject *) PyMarshal_WriteObjectToString(PyObject *, int);
 
-#define Py_MARSHAL_VERSION 4
+#define Py_MARSHAL_VERSION 5
 
 PyAPI_FUNC(long) PyMarshal_ReadLongFromFile(FILE *);
 PyAPI_FUNC(int) PyMarshal_ReadShortFromFile(FILE *);

Lib/test/test_marshal.py

@@ -28,6 +28,13 @@ def helper(self, sample, *extra):
         finally:
             os_helper.unlink(os_helper.TESTFN)
 
+def omit_last_byte(data):
+    """return data[:-1]"""
+    # This file's code is used in CompatibilityTestCase,
+    # but slices need marshal version 5.
+    # Avoid the slice literal.
+    return data[slice(0, -1)]
+
 class IntTestCase(unittest.TestCase, HelperMixin):
     def test_ints(self):
         # Test a range of Python ints larger than the machine word size.
@@ -241,7 +248,8 @@ def test_bug_5888452(self):
     def test_patch_873224(self):
         self.assertRaises(Exception, marshal.loads, b'0')
         self.assertRaises(Exception, marshal.loads, b'f')
-        self.assertRaises(Exception, marshal.loads, marshal.dumps(2**65)[:-1])
+        self.assertRaises(Exception, marshal.loads,
+                          omit_last_byte(marshal.dumps(2**65)))
 
     def test_version_argument(self):
         # Python 2.4.0 crashes for any call to marshal.dumps(x, y)
@@ -594,6 +602,19 @@ def testNoIntern(self):
         s2 = sys.intern(s)
         self.assertNotEqual(id(s2), id(s))
 
+class SliceTestCase(unittest.TestCase, HelperMixin):
+    def test_slice(self):
+        for obj in (
+            slice(None), slice(1), slice(1, 2), slice(1, 2, 3),
+            slice({'set'}, ('tuple', {'with': 'dict'}, ), self.helper.__code__)
+        ):
+            with self.subTest(obj=str(obj)):
+                self.helper(obj)
+
+                for version in range(4):
+                    with self.assertRaises(ValueError):
+                        marshal.dumps(obj, version)
+
 @support.cpython_only
 @unittest.skipUnless(_testcapi, 'requires _testcapi')
 class CAPI_TestCase(unittest.TestCase, HelperMixin):
@@ -654,7 +675,7 @@ def test_read_last_object_from_file(self):
             self.assertEqual(r, obj)
 
             with open(os_helper.TESTFN, 'wb') as f:
-                f.write(data[:1])
+                f.write(omit_last_byte(data))
             with self.assertRaises(EOFError):
                 _testcapi.pymarshal_read_last_object_from_file(os_helper.TESTFN)
             os_helper.unlink(os_helper.TESTFN)
@@ -671,7 +692,7 @@ def test_read_object_from_file(self):
             self.assertEqual(p, len(data))
 
             with open(os_helper.TESTFN, 'wb') as f:
-                f.write(data[:1])
+                f.write(omit_last_byte(data))
             with self.assertRaises(EOFError):
                 _testcapi.pymarshal_read_object_from_file(os_helper.TESTFN)
             os_helper.unlink(os_helper.TESTFN)

Misc/NEWS.d/next/Library/2024-11-14-13-16-20.gh-issue-125063.kJ-WnH.rst

@@ -0,0 +1,2 @@
+:mod:`marshal` now supports :class:`slice` objects. The marshal format
+version was increased to 5.

Programs/_freeze_module.c

@@ -121,6 +121,7 @@ compile_and_marshal(const char *name, const char *text)
         return NULL;
     }
 
+    assert(Py_MARSHAL_VERSION >= 5);
     PyObject *marshalled = PyMarshal_WriteObjectToString(code, Py_MARSHAL_VERSION);
     Py_CLEAR(code);
     if (marshalled == NULL) {

Python/marshal.c

@@ -50,41 +50,52 @@ module marshal
 #  define MAX_MARSHAL_STACK_DEPTH 2000
 #endif
 
+/* Supported types */
 #define TYPE_NULL               '0'
 #define TYPE_NONE               'N'
 #define TYPE_FALSE              'F'
 #define TYPE_TRUE               'T'
 #define TYPE_STOPITER           'S'
 #define TYPE_ELLIPSIS           '.'
-#define TYPE_INT                'i'
-/* TYPE_INT64 is not generated anymore.
-   Supported for backward compatibility only. */
-#define TYPE_INT64              'I'
-#define TYPE_FLOAT              'f'
-#define TYPE_BINARY_FLOAT       'g'
-#define TYPE_COMPLEX            'x'
-#define TYPE_BINARY_COMPLEX     'y'
-#define TYPE_LONG               'l'
-#define TYPE_STRING             's'
-#define TYPE_INTERNED           't'
-#define TYPE_REF                'r'
-#define TYPE_TUPLE              '('
+#define TYPE_BINARY_FLOAT       'g'  // Version 0 uses TYPE_FLOAT instead.
+#define TYPE_BINARY_COMPLEX     'y'  // Version 0 uses TYPE_COMPLEX instead.
+#define TYPE_LONG               'l'  // See also TYPE_INT.
+#define TYPE_STRING             's'  // Bytes. (Name comes from Python 2.)
+#define TYPE_TUPLE              '('  // See also TYPE_SMALL_TUPLE.
 #define TYPE_LIST               '['
 #define TYPE_DICT               '{'
 #define TYPE_CODE               'c'
 #define TYPE_UNICODE            'u'
 #define TYPE_UNKNOWN            '?'
+// added in version 2:
 #define TYPE_SET                '<'
 #define TYPE_FROZENSET          '>'
+// added in version 5:
 #define TYPE_SLICE              ':'
-#define FLAG_REF                '\x80' /* with a type, add obj to index */
+// Remember to update the version and documentation when adding new types.
 
+/* Special cases for unicode strings (added in version 4) */
+#define TYPE_INTERNED           't' // Version 1+
 #define TYPE_ASCII              'a'
 #define TYPE_ASCII_INTERNED     'A'
-#define TYPE_SMALL_TUPLE        ')'
 #define TYPE_SHORT_ASCII        'z'
 #define TYPE_SHORT_ASCII_INTERNED 'Z'
 
+/* Special cases for small objects */
+#define TYPE_INT                'i'  // All versions. 32-bit encoding.
+#define TYPE_SMALL_TUPLE        ')'  // Version 4+
+
+/* Supported for backwards compatibility */
+#define TYPE_COMPLEX            'x'  // Generated for version 0 only.
+#define TYPE_FLOAT              'f'  // Generated for version 0 only.
+#define TYPE_INT64              'I'  // Not generated any more.
+
+/* References (added in version 3) */
+#define TYPE_REF                'r'
+#define FLAG_REF                '\x80' /* with a type, add obj to index */
+
+
+// Error codes:
 #define WFERR_OK 0
 #define WFERR_UNMARSHALLABLE 1
 #define WFERR_NESTEDTOODEEP 2
@@ -615,6 +626,11 @@ w_complex_object(PyObject *v, char flag, WFILE *p)
         PyBuffer_Release(&view);
     }
     else if (PySlice_Check(v)) {
+        if (p->version < 5) {
+            w_byte(TYPE_UNKNOWN, p);
+            p->error = WFERR_UNMARSHALLABLE;
+            return;
+        }
         PySliceObject *slice = (PySliceObject *)v;
         W_TYPE(TYPE_SLICE, p);
         w_object(slice->start, p);