Use an self-clearing subclass to store hash cache

Rather than attempting to remove the hash cache from the object state on
deserialization or serialization, instead we store the hash cache in an
object that reduces to None, thus clearing itself when pickled or
copied.

This fixes GH python-attrs#494 and python-attrs#613.

Co-authored-by: Matt Wozniski <[email protected]>

src/attr/_make.py

@@ -70,6 +70,30 @@ def __repr__(self):
 """
 
 
+class _CacheHashWrapper(int):
+    """An integer subclass that pickles / copies as None
+
+    This is used for non-slots classes with ``cache_hash=True``, to avoid
+    serializing a potentially (even likely) invalid hash value. Since ``None``
+    is the default value for uncalculated hashes, whenever this is copied,
+    the copy's value for hte hash should automatically reset.
+
+    See GH #613 for more details.
+    """
+
+    if PY2:
+        # For some reason `type(None)` isn't callable in Python 2, but we don't
+        # actually need a constructor for None objects, we just need any
+        # available function that returns None.
+        def __reduce__(self, _none_constructor=getattr, _args=(0, "", None)):
+            return _none_constructor, _args
+
+    else:
+
+        def __reduce__(self, _none_constructor=type(None), _args=()):
+            return _none_constructor, _args
+
+
 def attrib(
     default=NOTHING,
     validator=None,
@@ -523,34 +547,6 @@ def _patch_original_class(self):
         for name, value in self._cls_dict.items():
             setattr(cls, name, value)
 
-        # Attach __setstate__. This is necessary to clear the hash code
-        # cache on deserialization. See issue
-        # https://github.com/python-attrs/attrs/issues/482 .
-        # Note that this code only handles setstate for dict classes.
-        # For slotted classes, see similar code in _create_slots_class .
-        if self._cache_hash:
-            existing_set_state_method = getattr(cls, "__setstate__", None)
-            if existing_set_state_method:
-                raise NotImplementedError(
-                    "Currently you cannot use hash caching if "
-                    "you specify your own __setstate__ method."
-                    "See https://github.com/python-attrs/attrs/issues/494 ."
-                )
-
-            # Clears the cached hash state on serialization; for frozen
-            # classes we need to bypass the class's setattr method.
-            if self._frozen:
-
-                def cache_hash_set_state(chss_self, _):
-                    object.__setattr__(chss_self, _hash_cache_field, None)
-
-            else:
-
-                def cache_hash_set_state(chss_self, _):
-                    setattr(chss_self, _hash_cache_field, None)
-
-            cls.__setstate__ = cache_hash_set_state
-
         return cls
 
     def _create_slots_class(self):
@@ -1103,22 +1099,42 @@ def _make_hash(cls, attrs, frozen, cache_hash):
     unique_filename = _generate_unique_filename(cls, "hash")
     type_hash = hash(unique_filename)
 
-    method_lines = ["def __hash__(self):"]
+    hash_def = "def __hash__(self"
+    hash_func = "hash(("
+    closing_braces = "))"
+    if not cache_hash:
+        hash_def += "):"
+    else:
+        if not PY2:
+            hash_def += ", *"
+
+        hash_def += (
+            ", _cache_wrapper="
+            + "__import__('attr._make')._make._CacheHashWrapper):"
+        )
+        hash_func = "_cache_wrapper(" + hash_func
+        closing_braces += ")"
+
+    method_lines = [hash_def]
 
     def append_hash_computation_lines(prefix, indent):
         """
         Generate the code for actually computing the hash code.
         Below this will either be returned directly or used to compute
         a value which is then cached, depending on the value of cache_hash
         """
+
         method_lines.extend(
-            [indent + prefix + "hash((", indent + "        %d," % (type_hash,)]
+            [
+                indent + prefix + hash_func,
+                indent + "        %d," % (type_hash,),
+            ]
         )
 
         for a in attrs:
             method_lines.append(indent + "        self.%s," % a.name)
 
-        method_lines.append(indent + "    ))")
+        method_lines.append(indent + "    " + closing_braces)
 
     if cache_hash:
         method_lines.append(tab + "if self.%s is None:" % _hash_cache_field)