bpo-42536: GC track recycled tuples (GH-23623)

Several built-in and standard library types now ensure that their internal result tuples are always tracked by the garbage collector:

- collections.OrderedDict.items
- dict.items
- enumerate
- functools.reduce
- itertools.combinations
- itertools.combinations_with_replacement
- itertools.permutations
- itertools.product
- itertools.zip_longest
- zip

Previously, they could have become untracked by a prior garbage collection.

diff --git a/Lib/test/test_builtin.py b/Lib/test/test_builtin.py
index edb4ec092e3586cbe94ac533f7aee1b37b97fc62..8c9573731ae2eda23b9a97b4819506324f93ceb3 100644 (file)
--- a/Lib/test/test_builtin.py
+++ b/Lib/test/test_builtin.py
@@ -6,6 +6,7 @@ import builtins
 import collections
 import decimal
 import fractions
+import gc
 import io
 import locale
 import os
@@ -1756,6 +1757,18 @@ class BuiltinTest(unittest.TestCase):
         l8 = self.iter_error(zip(Iter(3), "AB", strict=True), ValueError)
         self.assertEqual(l8, [(2, "A"), (1, "B")])
 
+    @support.cpython_only
+    def test_zip_result_gc(self):
+        # bpo-42536: zip's tuple-reuse speed trick breaks the GC's assumptions
+        # about what can be untracked. Make sure we re-track result tuples
+        # whenever we reuse them.
+        it = zip([[]])
+        gc.collect()
+        # That GC collection probably untracked the recycled internal result
+        # tuple, which is initialized to (None,). Make sure it's re-tracked when
+        # it's mutated and returned from __next__:
+        self.assertTrue(gc.is_tracked(next(it)))
+
     def test_format(self):
         # Test the basic machinery of the format() builtin.  Don't test
         #  the specifics of the various formatters

diff --git a/Lib/test/test_dict.py b/Lib/test/test_dict.py
index 9ff8b7d501aad63e59da742facd033e41c7ab1ba..4b31cdc79415fff8a6a90319442da5288814908e 100644 (file)
--- a/Lib/test/test_dict.py
+++ b/Lib/test/test_dict.py
@@ -1452,6 +1452,25 @@ class DictTest(unittest.TestCase):
         d = CustomReversedDict(pairs)
         self.assertEqual(pairs[::-1], list(dict(d).items()))
 
+    @support.cpython_only
+    def test_dict_items_result_gc(self):
+        # bpo-42536: dict.items's tuple-reuse speed trick breaks the GC's
+        # assumptions about what can be untracked. Make sure we re-track result
+        # tuples whenever we reuse them.
+        it = iter({None: []}.items())
+        gc.collect()
+        # That GC collection probably untracked the recycled internal result
+        # tuple, which is initialized to (None, None). Make sure it's re-tracked
+        # when it's mutated and returned from __next__:
+        self.assertTrue(gc.is_tracked(next(it)))
+
+    @support.cpython_only
+    def test_dict_items_result_gc(self):
+        # Same as test_dict_items_result_gc above, but reversed.
+        it = reversed({None: []}.items())
+        gc.collect()
+        self.assertTrue(gc.is_tracked(next(it)))
+
 
 class CAPITest(unittest.TestCase):
 

diff --git a/Lib/test/test_enumerate.py b/Lib/test/test_enumerate.py
index 5785cb46492ef7c65a7281f24065ccae0454ae6e..906bfc21a26aede0a6a9d9014e95c385b43c53b2 100644 (file)
--- a/Lib/test/test_enumerate.py
+++ b/Lib/test/test_enumerate.py
@@ -2,6 +2,7 @@ import unittest
 import operator
 import sys
 import pickle
+import gc
 
 from test import support
 
@@ -134,6 +135,18 @@ class EnumerateTestCase(unittest.TestCase, PickleTest):
         self.assertEqual(len(set(map(id, list(enumerate(self.seq))))), len(self.seq))
         self.assertEqual(len(set(map(id, enumerate(self.seq)))), min(1,len(self.seq)))
 
+    @support.cpython_only
+    def test_enumerate_result_gc(self):
+        # bpo-42536: enumerate's tuple-reuse speed trick breaks the GC's
+        # assumptions about what can be untracked. Make sure we re-track result
+        # tuples whenever we reuse them.
+        it = self.enum([[]])
+        gc.collect()
+        # That GC collection probably untracked the recycled internal result
+        # tuple, which is initialized to (None, None). Make sure it's re-tracked
+        # when it's mutated and returned from __next__:
+        self.assertTrue(gc.is_tracked(next(it)))
+
 class MyEnum(enumerate):
     pass
 

diff --git a/Lib/test/test_itertools.py b/Lib/test/test_itertools.py
index df2997e87d494cf10084f30664732f7056f13472..a99b5e2bb71db9b2d8cf39da1078cdefc5246f07 100644 (file)
--- a/Lib/test/test_itertools.py
+++ b/Lib/test/test_itertools.py
@@ -12,6 +12,8 @@ from functools import reduce
 import sys
 import struct
 import threading
+import gc
+
 maxsize = support.MAX_Py_ssize_t
 minsize = -maxsize-1
 
@@ -1573,6 +1575,51 @@ class TestBasicOps(unittest.TestCase):
             self.assertRaises(StopIteration, next, f(lambda x:x, []))
             self.assertRaises(StopIteration, next, f(lambda x:x, StopNow()))
 
+    @support.cpython_only
+    def test_combinations_result_gc(self):
+        # bpo-42536: combinations's tuple-reuse speed trick breaks the GC's
+        # assumptions about what can be untracked. Make sure we re-track result
+        # tuples whenever we reuse them.
+        it = combinations([None, []], 1)
+        next(it)
+        gc.collect()
+        # That GC collection probably untracked the recycled internal result
+        # tuple, which has the value (None,). Make sure it's re-tracked when
+        # it's mutated and returned from __next__:
+        self.assertTrue(gc.is_tracked(next(it)))
+
+    @support.cpython_only
+    def test_combinations_with_replacement_result_gc(self):
+        # Ditto for combinations_with_replacement.
+        it = combinations_with_replacement([None, []], 1)
+        next(it)
+        gc.collect()
+        self.assertTrue(gc.is_tracked(next(it)))
+
+    @support.cpython_only
+    def test_permutations_result_gc(self):
+        # Ditto for permutations.
+        it = permutations([None, []], 1)
+        next(it)
+        gc.collect()
+        self.assertTrue(gc.is_tracked(next(it)))
+
+    @support.cpython_only
+    def test_product_result_gc(self):
+        # Ditto for product.
+        it = product([None, []])
+        next(it)
+        gc.collect()
+        self.assertTrue(gc.is_tracked(next(it)))
+
+    @support.cpython_only
+    def test_zip_longest_result_gc(self):
+        # Ditto for zip_longest.
+        it = zip_longest([[]])
+        gc.collect()
+        self.assertTrue(gc.is_tracked(next(it)))
+
+
 class TestExamples(unittest.TestCase):
 
     def test_accumulate(self):

diff --git a/Lib/test/test_ordered_dict.py b/Lib/test/test_ordered_dict.py
index 31759f20d28349d3c668d00bd2e8e83afd0b067e..eb404463e92550daef2fef6183030d17ed93ce44 100644 (file)
--- a/Lib/test/test_ordered_dict.py
+++ b/Lib/test/test_ordered_dict.py
@@ -700,6 +700,17 @@ class OrderedDictTests:
         with self.assertRaises(ValueError):
             a |= "BAD"
 
+    @support.cpython_only
+    def test_ordered_dict_items_result_gc(self):
+        # bpo-42536: OrderedDict.items's tuple-reuse speed trick breaks the GC's
+        # assumptions about what can be untracked. Make sure we re-track result
+        # tuples whenever we reuse them.
+        it = iter(self.OrderedDict({None: []}).items())
+        gc.collect()
+        # That GC collection probably untracked the recycled internal result
+        # tuple, which is initialized to (None, None). Make sure it's re-tracked
+        # when it's mutated and returned from __next__:
+        self.assertTrue(gc.is_tracked(next(it)))
 
 class PurePythonOrderedDictTests(OrderedDictTests, unittest.TestCase):
 

diff --git a/Misc/NEWS.d/next/Core and Builtins/2020-12-02-20-23-31.bpo-42536.Kx3ZOu.rst b/Misc/NEWS.d/next/Core and Builtins/2020-12-02-20-23-31.bpo-42536.Kx3ZOu.rst
new file mode 100644 (file)
index 0000000..6ccacab
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2020-12-02-20-23-31.bpo-42536.Kx3ZOu.rst
@@ -0,0 +1,26 @@
+Several built-in and standard library types now ensure that their internal
+result tuples are always tracked by the :term:`garbage collector
+<garbage collection>`:
+
+- :meth:`collections.OrderedDict.items() <collections.OrderedDict>`
+
+- :meth:`dict.items`
+
+- :func:`enumerate`
+
+- :func:`functools.reduce`
+
+- :func:`itertools.combinations`
+
+- :func:`itertools.combinations_with_replacement`
+
+- :func:`itertools.permutations`
+
+- :func:`itertools.product`
+
+- :func:`itertools.zip_longest`
+
+- :func:`zip`
+
+Previously, they could have become untracked by a prior garbage collection.
+Patch by Brandt Bucher.

diff --git a/Modules/_functoolsmodule.c b/Modules/_functoolsmodule.c
index ff03c334766b899b2d2fd6c05c50dc3f5a34e714..621b721d011df73db56994bd9be890732141a069 100644 (file)
--- a/Modules/_functoolsmodule.c
+++ b/Modules/_functoolsmodule.c
@@ -1,5 +1,6 @@
 #include "Python.h"
 #include "pycore_long.h"          // _PyLong_GetZero()
+#include "pycore_object.h"        // _PyObject_GC_TRACK
 #include "pycore_pystate.h"       // _PyThreadState_GET()
 #include "pycore_tuple.h"         // _PyTuple_ITEMS()
 #include "structmember.h"         // PyMemberDef
@@ -673,6 +674,11 @@ functools_reduce(PyObject *self, PyObject *args)
             if ((result = PyObject_Call(func, args, NULL)) == NULL) {
                 goto Fail;
             }
+            // bpo-42536: The GC may have untracked this args tuple. Since we're
+            // recycling it, make sure it's tracked again:
+            if (!_PyObject_GC_IS_TRACKED(args)) {
+                _PyObject_GC_TRACK(args);
+            }
         }
     }
 

diff --git a/Modules/itertoolsmodule.c b/Modules/itertoolsmodule.c
index 7144856c352f81f6d5f60488d6de2c240b590cfc..293735a886428b5ab8ffdace03a4797089120ae9 100644 (file)
--- a/Modules/itertoolsmodule.c
+++ b/Modules/itertoolsmodule.c
@@ -3,6 +3,7 @@
 #define PY_SSIZE_T_CLEAN
 #include "Python.h"
 #include "pycore_long.h"          // _PyLong_GetZero()
+#include "pycore_object.h"        // _PyObject_GC_TRACK()
 #include "pycore_tuple.h"         // _PyTuple_ITEMS()
 #include <stddef.h>               // offsetof()
 
@@ -2378,6 +2379,11 @@ product_next(productobject *lz)
             lz->result = result;
             Py_DECREF(old_result);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        else if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         /* Now, we've got the only copy so we can update it in-place */
         assert (npools==0 || Py_REFCNT(result) == 1);
 
@@ -2701,6 +2707,11 @@ combinations_next(combinationsobject *co)
             co->result = result;
             Py_DECREF(old_result);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        else if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         /* Now, we've got the only copy so we can update it in-place
          * CPython's empty tuple is a singleton and cached in
          * PyTuple's freelist.
@@ -3035,6 +3046,11 @@ cwr_next(cwrobject *co)
             co->result = result;
             Py_DECREF(old_result);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        else if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         /* Now, we've got the only copy so we can update it in-place CPython's
            empty tuple is a singleton and cached in PyTuple's freelist. */
         assert(r == 0 || Py_REFCNT(result) == 1);
@@ -3379,6 +3395,11 @@ permutations_next(permutationsobject *po)
             po->result = result;
             Py_DECREF(old_result);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        else if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         /* Now, we've got the only copy so we can update it in-place */
         assert(r == 0 || Py_REFCNT(result) == 1);
 
@@ -4649,6 +4670,11 @@ zip_longest_next(ziplongestobject *lz)
             PyTuple_SET_ITEM(result, i, item);
             Py_DECREF(olditem);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
     } else {
         result = PyTuple_New(tuplesize);
         if (result == NULL)

diff --git a/Objects/dictobject.c b/Objects/dictobject.c
index 7a37313df8a6b1c08ea595c8e07e56f24a9e1363..35e881fe2723084fba7f546e9e3d9cdf83d36446 100644 (file)
--- a/Objects/dictobject.c
+++ b/Objects/dictobject.c
@@ -3989,6 +3989,11 @@ dictiter_iternextitem(dictiterobject *di)
         Py_INCREF(result);
         Py_DECREF(oldkey);
         Py_DECREF(oldvalue);
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
     }
     else {
         result = PyTuple_New(2);
@@ -4104,6 +4109,11 @@ dictreviter_iternext(dictiterobject *di)
             Py_INCREF(result);
             Py_DECREF(oldkey);
             Py_DECREF(oldvalue);
+            // bpo-42536: The GC may have untracked this result tuple. Since
+            // we're recycling it, make sure it's tracked again:
+            if (!_PyObject_GC_IS_TRACKED(result)) {
+                _PyObject_GC_TRACK(result);
+            }
         }
         else {
             result = PyTuple_New(2);

diff --git a/Objects/enumobject.c b/Objects/enumobject.c
index 8b5e7d3a3c6dd578535df264c1cfcd05229c715b..98ece3f13fc6fbe512b0f7b420aeab3e8024637d 100644 (file)
--- a/Objects/enumobject.c
+++ b/Objects/enumobject.c
@@ -2,6 +2,7 @@
 
 #include "Python.h"
 #include "pycore_long.h"          // _PyLong_GetOne()
+#include "pycore_object.h"        // _PyObject_GC_TRACK()
 
 #include "clinic/enumobject.c.h"
 
@@ -131,6 +132,11 @@ enum_next_long(enumobject *en, PyObject* next_item)
         PyTuple_SET_ITEM(result, 1, next_item);
         Py_DECREF(old_index);
         Py_DECREF(old_item);
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         return result;
     }
     result = PyTuple_New(2);
@@ -176,6 +182,11 @@ enum_next(enumobject *en)
         PyTuple_SET_ITEM(result, 1, next_item);
         Py_DECREF(old_index);
         Py_DECREF(old_item);
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
         return result;
     }
     result = PyTuple_New(2);

diff --git a/Objects/odictobject.c b/Objects/odictobject.c
index 4eb15f999bd1e42f9776e8644b57a8ced35cab8f..6c7f1175cd652ccb65bd2ab522ee0bf7887a95b8 100644 (file)
--- a/Objects/odictobject.c
+++ b/Objects/odictobject.c
@@ -1814,6 +1814,11 @@ odictiter_iternext(odictiterobject *di)
         Py_INCREF(result);
         Py_DECREF(PyTuple_GET_ITEM(result, 0));  /* borrowed */
         Py_DECREF(PyTuple_GET_ITEM(result, 1));  /* borrowed */
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
     }
     else {
         result = PyTuple_New(2);

diff --git a/Python/bltinmodule.c b/Python/bltinmodule.c
index a73b8cb320e975aeff3d3668660be1b4e7bf495d..352fb83d55e05393415dd59716ced6bbc13dc513 100644 (file)
--- a/Python/bltinmodule.c
+++ b/Python/bltinmodule.c
@@ -2636,6 +2636,11 @@ zip_next(zipobject *lz)
             PyTuple_SET_ITEM(result, i, item);
             Py_DECREF(olditem);
         }
+        // bpo-42536: The GC may have untracked this result tuple. Since we're
+        // recycling it, make sure it's tracked again:
+        if (!_PyObject_GC_IS_TRACKED(result)) {
+            _PyObject_GC_TRACK(result);
+        }
     } else {
         result = PyTuple_New(tuplesize);
         if (result == NULL)