mitogen: Raise TypeError on `mitogen.utils.cast(custom_str)` failures

If casting a string fails then raise a TypeError. This is potentially an API
breaking change; chosen as the lesser evil vs. allowing silent errors.

`cast()` relies on `bytes(obj)` & `str(obj)` returning the respective
supertype. That's no longer the case for `AnsibleUnsafeBytes` &
`AnsibleUnsafeText`; since fixes/mitigations for  CVE-2023-5764.

fixes #1046, refs #977

See also
- https://github.com/advisories/GHSA-7j69-qfc3-2fq9
- https://github.com/ansible/ansible/pull/82293

docs/changelog.rst

@@ -21,6 +21,9 @@ Unreleased
 ----------
 
 * :gh:issue:`974` Support Ansible 7
+* :gh:issue:`1046` Raise :py:exc:`TypeError` in :func:`<mitogen.util.cast()>`
+  when casting a string subtype to `bytes()` or `str()` fails. This is
+  potentially an API breaking change. Failures previously passed silently.
 
 
 v0.3.5 (2024-03-17)

mitogen/utils.py

@@ -190,10 +190,13 @@ PASSTHROUGH = (
 
 def cast(obj):
     """
+    Return obj (or a copy) with subtypes of builtins cast to their supertype.
+    Subtypes of those in :data:`PASSTHROUGH` are not modified.
+
     Many tools love to subclass built-in types in order to implement useful
     functionality, such as annotating the safety of a Unicode string, or adding
-    additional methods to a dict. However, cPickle loves to preserve those
+    additional methods to a dict. However :py:mod:`pickle` serializes these
-    subtypes during serialization, resulting in CallError during :meth:`call
+    exactly, leading to :exc:`mitogen.CallError` during :meth:`Context.call
     <mitogen.parent.Context.call>` in the target when it tries to deserialize
     the data.
 
@@ -201,6 +204,9 @@ def cast(obj):
     custom sub-types removed. The functionality is not default since the
     resulting walk may be computationally expensive given a large enough graph.
 
+    Raises :py:exc:`TypeError` if an unknown subtype is encountered, or
+    casting does not return the desired supertype.
+
     See :ref:`serialization-rules` for a list of supported types.
 
     :param obj:
@@ -215,8 +221,16 @@ def cast(obj):
     if isinstance(obj, PASSTHROUGH):
         return obj
     if isinstance(obj, mitogen.core.UnicodeType):
-        return mitogen.core.UnicodeType(obj)
+        return _cast(obj, mitogen.core.UnicodeType)
     if isinstance(obj, mitogen.core.BytesType):
-        return mitogen.core.BytesType(obj)
+        return _cast(obj, mitogen.core.BytesType)
 
     raise TypeError("Cannot serialize: %r: %r" % (type(obj), obj))
+
+
+def _cast(obj, desired_type):
+    result = desired_type(obj)
+    if type(result) is not desired_type:
+        raise TypeError("Cast of %r to %r failed, got %r"
+                        % (type(obj), desired_type, type(result)))
+    return result

tests/utils_test.py

@@ -44,6 +44,44 @@ class Unicode(mitogen.core.UnicodeType): pass
 class Bytes(mitogen.core.BytesType): pass
 
 
+class StubbornBytes(mitogen.core.BytesType):
+    """
+    A binary string type that persists through `bytes(...)`.
+
+    Stand-in for `AnsibleUnsafeBytes()` in Ansible 7-9 (core 2.14-2.16), after
+    fixes/mitigations for CVE-2023-5764.
+    """
+    if mitogen.core.PY3:
+        def __bytes__(self): return self
+        def __str__(self): return self.decode()
+    else:
+        def __str__(self): return self
+        def __unicode__(self): return self.decode()
+
+    def decode(self, encoding='utf-8', errors='strict'):
+        s = super(StubbornBytes).encode(encoding=encoding, errors=errors)
+        return StubbornText(s)
+
+
+class StubbornText(mitogen.core.UnicodeType):
+    """
+    A text string type that persists through `unicode(...)` or `str(...)`.
+
+    Stand-in for `AnsibleUnsafeText()` in Ansible 7-9 (core 2.14-2.16), after
+    following fixes/mitigations for CVE-2023-5764.
+    """
+    if mitogen.core.PY3:
+        def __bytes__(self): return self.encode()
+        def __str__(self): return self
+    else:
+        def __str__(self): return self.encode()
+        def __unicode__(self): return self
+
+    def encode(self, encoding='utf-8', errors='strict'):
+        s = super(StubbornText).encode(encoding=encoding, errors=errors)
+        return StubbornBytes(s)
+
+
 class CastTest(testlib.TestCase):
     def test_dict(self):
         self.assertEqual(type(mitogen.utils.cast({})), dict)
@@ -91,6 +129,15 @@ class CastTest(testlib.TestCase):
         self.assertEqual(type(mitogen.utils.cast(b(''))), mitogen.core.BytesType)
         self.assertEqual(type(mitogen.utils.cast(Bytes())), mitogen.core.BytesType)
 
+    def test_stubborn_types_raise(self):
+        stubborn_bytes = StubbornBytes(b('abc'))
+        self.assertIs(stubborn_bytes, mitogen.core.BytesType(stubborn_bytes))
+        self.assertRaises(TypeError, mitogen.utils.cast, stubborn_bytes)
+
+        stubborn_text = StubbornText(u'abc')
+        self.assertIs(stubborn_text, mitogen.core.UnicodeType(stubborn_text))
+        self.assertRaises(TypeError, mitogen.utils.cast, stubborn_text)
+
     def test_unknown(self):
         self.assertRaises(TypeError, mitogen.utils.cast, set())
         self.assertRaises(TypeError, mitogen.utils.cast, 4j)