bpo-35900: Add a state_setter arg to save_reduce (GH-12588)

Allow reduction methods to return a 6-item tuple where the 6th item specifies a
custom state-setting method that's called instead of the regular
``__setstate__`` method.

5 files changed, 114 insertions, 15 deletions

diff --git a/Doc/library/pickle.rst b/Doc/library/pickle.rst
index 53eb5d3..3d89536 100644
--- a/Doc/library/pickle.rst
+++ b/Doc/library/pickle.rst
@@ -598,7 +598,7 @@ or both.
    module; the pickle module searches the module namespace to determine the
    object's module.  This behaviour is typically useful for singletons.
 
-   When a tuple is returned, it must be between two and five items long.
+   When a tuple is returned, it must be between two and six items long.
    Optional items can either be omitted, or ``None`` can be provided as their
    value.  The semantics of each item are in order:
 
@@ -629,6 +629,15 @@ or both.
      value``.  This is primarily used for dictionary subclasses, but may be used
      by other classes as long as they implement :meth:`__setitem__`.
 
+   * Optionally, a callable with a ``(obj, state)`` signature. This
+     callable allows the user to programatically control the state-updating
+     behavior of a specific object, instead of using ``obj``'s static
+     :meth:`__setstate__` method. If not ``None``, this callable will have
+     priority over ``obj``'s :meth:`__setstate__`.
+
+     .. versionadded:: 3.8
+        The optional sixth tuple item, ``(obj, state)``, was added.
+
 
 .. method:: object.__reduce_ex__(protocol)
 
diff --git a/Lib/pickle.py b/Lib/pickle.py
index d533e66..47f0d28 100644
--- a/Lib/pickle.py
+++ b/Lib/pickle.py
@@ -537,9 +537,9 @@ class _Pickler:
 
         # Assert that it returned an appropriately sized tuple
         l = len(rv)
-        if not (2 <= l <= 5):
+        if not (2 <= l <= 6):
             raise PicklingError("Tuple returned by %s must have "
-                                "two to five elements" % reduce)
+                                "two to six elements" % reduce)
 
         # Save the reduce() output and finally memoize the object
         self.save_reduce(obj=obj, *rv)
@@ -561,7 +561,7 @@ class _Pickler:
                     "persistent IDs in protocol 0 must be ASCII strings")
 
     def save_reduce(self, func, args, state=None, listitems=None,
-                    dictitems=None, obj=None):
+                    dictitems=None, state_setter=None, obj=None):
         # This API is called by some subclasses
 
         if not isinstance(args, tuple):
@@ -655,8 +655,25 @@ class _Pickler:
             self._batch_setitems(dictitems)
 
         if state is not None:
-            save(state)
-            write(BUILD)
+            if state_setter is None:
+                save(state)
+                write(BUILD)
+            else:
+                # If a state_setter is specified, call it instead of load_build
+                # to update obj's with its previous state.
+                # First, push state_setter and its tuple of expected arguments
+                # (obj, state) onto the stack.
+                save(state_setter)
+                save(obj)  # simple BINGET opcode as obj is already memoized.
+                save(state)
+                write(TUPLE2)
+                # Trigger a state_setter(obj, state) function call.
+                write(REDUCE)
+                # The purpose of state_setter is to carry-out an
+                # inplace modification of obj. We do not care about what the
+                # method might return, so its output is eventually removed from
+                # the stack.
+                write(POP)
 
     # Methods below this point are dispatched through the dispatch table
 
diff --git a/Lib/test/pickletester.py b/Lib/test/pickletester.py
index bb8e6ce..19e8823 100644
--- a/Lib/test/pickletester.py
+++ b/Lib/test/pickletester.py
@@ -2992,7 +2992,26 @@ class AAA(object):
         return str, (REDUCE_A,)
 
 class BBB(object):
-    pass
+    def __init__(self):
+        # Add an instance attribute to enable state-saving routines at pickling
+        # time.
+        self.a = "some attribute"
+
+    def __setstate__(self, state):
+        self.a = "BBB.__setstate__"
+
+
+def setstate_bbb(obj, state):
+    """Custom state setter for BBB objects
+
+    Such callable may be created by other persons than the ones who created the
+    BBB class. If passed as the state_setter item of a custom reducer, this
+    allows for custom state setting behavior of BBB objects. One can think of
+    it as the analogous of list_setitems or dict_setitems but for foreign
+    classes/functions.
+    """
+    obj.a = "custom state_setter"
+
 
 class AbstractDispatchTableTests(unittest.TestCase):
 
@@ -3081,6 +3100,25 @@ class AbstractDispatchTableTests(unittest.TestCase):
         self.assertEqual(default_load_dump(a), REDUCE_A)
         self.assertIsInstance(default_load_dump(b), BBB)
 
+        # End-to-end testing of save_reduce with the state_setter keyword
+        # argument. This is a dispatch_table test as the primary goal of
+        # state_setter is to tweak objects reduction behavior.
+        # In particular, state_setter is useful when the default __setstate__
+        # behavior is not flexible enough.
+
+        # No custom reducer for b has been registered for now, so
+        # BBB.__setstate__ should be used at unpickling time
+        self.assertEqual(default_load_dump(b).a, "BBB.__setstate__")
+
+        def reduce_bbb(obj):
+            return BBB, (), obj.__dict__, None, None, setstate_bbb
+
+        dispatch_table[BBB] = reduce_bbb
+
+        # The custom reducer reduce_bbb includes a state setter, that should
+        # have priority over BBB.__setstate__
+        self.assertEqual(custom_load_dump(b).a, "custom state_setter")
+
 
 if __name__ == "__main__":
     # Print some stuff that can be used to rewrite DATA{0,1,2}
diff --git a/Misc/NEWS.d/next/Library/2019-03-27-15-09-00.bpo-35900.fh56UU.rst b/Misc/NEWS.d/next/Library/2019-03-27-15-09-00.bpo-35900.fh56UU.rst
new file mode 100644
index 0000000..7f3a067
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2019-03-27-15-09-00.bpo-35900.fh56UU.rst
@@ -0,0 +1,3 @@
+Allow reduction methods to return a 6-item tuple where the 6th item specifies a
+custom state-setting method that's called instead of the regular
+``__setstate__`` method.
diff --git a/Modules/_pickle.c b/Modules/_pickle.c
index 391ce5e..897bbe1 100644
--- a/Modules/_pickle.c
+++ b/Modules/_pickle.c
@@ -3662,6 +3662,7 @@ save_reduce(PicklerObject *self, PyObject *args, PyObject *obj)
     PyObject *state = NULL;
     PyObject *listitems = Py_None;
     PyObject *dictitems = Py_None;
+    PyObject *state_setter = Py_None;
     PickleState *st = _Pickle_GetGlobalState();
     Py_ssize_t size;
     int use_newobj = 0, use_newobj_ex = 0;
@@ -3672,14 +3673,15 @@ save_reduce(PicklerObject *self, PyObject *args, PyObject *obj)
     const char newobj_ex_op = NEWOBJ_EX;
 
     size = PyTuple_Size(args);
-    if (size < 2 || size > 5) {
+    if (size < 2 || size > 6) {
         PyErr_SetString(st->PicklingError, "tuple returned by "
-                        "__reduce__ must contain 2 through 5 elements");
+                        "__reduce__ must contain 2 through 6 elements");
         return -1;
     }
 
-    if (!PyArg_UnpackTuple(args, "save_reduce", 2, 5,
-                           &callable, &argtup, &state, &listitems, &dictitems))
+    if (!PyArg_UnpackTuple(args, "save_reduce", 2, 6,
+                           &callable, &argtup, &state, &listitems, &dictitems,
+                           &state_setter))
         return -1;
 
     if (!PyCallable_Check(callable)) {
@@ -3714,6 +3716,15 @@ save_reduce(PicklerObject *self, PyObject *args, PyObject *obj)
         return -1;
     }
 
+    if (state_setter == Py_None)
+        state_setter = NULL;
+    else if (!PyCallable_Check(state_setter)) {
+        PyErr_Format(st->PicklingError, "sixth element of the tuple "
+                     "returned by __reduce__ must be a function, not %s",
+                     Py_TYPE(state_setter)->tp_name);
+        return -1;
+    }
+
     if (self->proto >= 2) {
         PyObject *name;
         _Py_IDENTIFIER(__name__);
@@ -3933,11 +3944,32 @@ save_reduce(PicklerObject *self, PyObject *args, PyObject *obj)
         return -1;
 
     if (state) {
-        if (save(self, state, 0) < 0 ||
-            _Pickler_Write(self, &build_op, 1) < 0)
-            return -1;
-    }
+        if (state_setter == NULL) {
+            if (save(self, state, 0) < 0 ||
+                _Pickler_Write(self, &build_op, 1) < 0)
+                return -1;
+        }
+        else {
+
+            /* If a state_setter is specified, call it instead of load_build to
+             * update obj's with its previous state.
+             * The first 4 save/write instructions push state_setter and its
+             * tuple of expected arguments (obj, state) onto the stack. The
+             * REDUCE opcode triggers the state_setter(obj, state) function
+             * call. Finally, because state-updating routines only do in-place
+             * modification, the whole operation has to be stack-transparent.
+             * Thus, we finally pop the call's output from the stack.*/
 
+            const char tupletwo_op = TUPLE2;
+            const char pop_op = POP;
+            if (save(self, state_setter, 0) < 0 ||
+                save(self, obj, 0) < 0 || save(self, state, 0) < 0 ||
+                _Pickler_Write(self, &tupletwo_op, 1) < 0 ||
+                _Pickler_Write(self, &reduce_op, 1) < 0 ||
+                _Pickler_Write(self, &pop_op, 1) < 0)
+                return -1;
+        }
+    }
     return 0;
 }