Skip to content

GH-127010: Don't lazily track and untrack dicts #127027

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Nov 20, 2024
Merged

GH-127010: Don't lazily track and untrack dicts #127027

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
498979f
Don't untrack dicts
markshannon Nov 19, 2024
3c11dba
Add news
markshannon Nov 19, 2024
2859bec
Update gc.is_tracked() docs
markshannon Nov 20, 2024
31c2839
Tidy docs about tuple tracking
markshannon Nov 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions Doc/library/gc.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,8 +204,6 @@ The :mod:`gc` module provides the following functions:
>>> gc.is_tracked({})
False
>>> gc.is_tracked({"a": 1})
False
>>> gc.is_tracked({"a": []})
True

.. versionadded:: 3.1
Expand Down
2 changes: 0 additions & 2 deletions Include/internal/pycore_dict.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,8 +43,6 @@ extern int _PyDict_Next(

extern int _PyDict_HasOnlyStringKeys(PyObject *mp);

extern void _PyDict_MaybeUntrack(PyObject *mp);

// Export for '_ctypes' shared extension
PyAPI_FUNC(Py_ssize_t) _PyDict_SizeOf(PyDictObject *);

Expand Down
16 changes: 4 additions & 12 deletions InternalDocs/garbage_collector.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -532,8 +532,8 @@ of `PyGC_Head` discussed in the `Memory layout and object structure`_ section:
currently in. Instead, when that's needed, ad hoc tricks (like the
`NEXT_MASK_UNREACHABLE` flag) are employed.

Optimization: delay tracking containers
=======================================
Optimization: delayed untracking containers
===========================================

Certain types of containers cannot participate in a reference cycle, and so do
not need to be tracked by the garbage collector. Untracking these objects
Expand All @@ -548,8 +548,8 @@ a container:
As a general rule, instances of atomic types aren't tracked and instances of
non-atomic types (containers, user-defined objects...) are. However, some
type-specific optimizations can be present in order to suppress the garbage
collector footprint of simple instances. Some examples of native types that
benefit from delayed tracking:
collector footprint of simple instances. Historically, both dictionaries and
tuples were untracked during garbage collection. Now it is only tuples:

- Tuples containing only immutable objects (integers, strings etc,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is now a single-item bullet list.

and recursively, tuples of immutable objects) do not need to be tracked. The
Expand All @@ -561,12 +561,6 @@ benefit from delayed tracking:
already not tracked. Tuples are examined for untracking in all garbage collection
cycles. It may take more than one cycle to untrack a tuple.

- Dictionaries containing only immutable objects also do not need to be tracked.
Dictionaries are untracked when created. If a tracked item is inserted into a
dictionary (either as a key or value), the dictionary becomes tracked. During a
full garbage collection (all generations), the collector will untrack any dictionaries
whose contents are not tracked.

The garbage collector module provides the Python function `is_tracked(obj)`, which returns
the current tracking status of the object. Subsequent garbage collections may change the
tracking status of the object.
Expand All @@ -581,8 +575,6 @@ tracking status of the object.
>>> gc.is_tracked({})
False
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is empty dict still untracked?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No it is tracked. I'll update the docs.

>>> gc.is_tracked({"a": 1})
False
>>> gc.is_tracked({"a": []})
True
```

Expand Down
109 changes: 0 additions & 109 deletions Lib/test/test_dict.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -880,115 +880,6 @@ class C(object):
gc.collect()
self.assertIs(ref(), None, "Cycle was not collected")

def _not_tracked(self, t):
# Nested containers can take several collections to untrack
gc.collect()
gc.collect()
self.assertFalse(gc.is_tracked(t), t)

def _tracked(self, t):
self.assertTrue(gc.is_tracked(t), t)
gc.collect()
gc.collect()
self.assertTrue(gc.is_tracked(t), t)

def test_string_keys_can_track_values(self):
# Test that this doesn't leak.
for i in range(10):
d = {}
for j in range(10):
d[str(j)] = j
d["foo"] = d

@support.cpython_only
def test_track_literals(self):
# Test GC-optimization of dict literals
x, y, z, w = 1.5, "a", (1, None), []

self._not_tracked({})
self._not_tracked({x:(), y:x, z:1})
self._not_tracked({1: "a", "b": 2})
self._not_tracked({1: 2, (None, True, False, ()): int})
self._not_tracked({1: object()})

# Dicts with mutable elements are always tracked, even if those
# elements are not tracked right now.
self._tracked({1: []})
self._tracked({1: ([],)})
self._tracked({1: {}})
self._tracked({1: set()})

@support.cpython_only
def test_track_dynamic(self):
# Test GC-optimization of dynamically-created dicts
class MyObject(object):
pass
x, y, z, w, o = 1.5, "a", (1, object()), [], MyObject()

d = dict()
self._not_tracked(d)
d[1] = "a"
self._not_tracked(d)
d[y] = 2
self._not_tracked(d)
d[z] = 3
self._not_tracked(d)
self._not_tracked(d.copy())
d[4] = w
self._tracked(d)
self._tracked(d.copy())
d[4] = None
self._not_tracked(d)
self._not_tracked(d.copy())

# dd isn't tracked right now, but it may mutate and therefore d
# which contains it must be tracked.
d = dict()
dd = dict()
d[1] = dd
self._not_tracked(dd)
self._tracked(d)
dd[1] = d
self._tracked(dd)

d = dict.fromkeys([x, y, z])
self._not_tracked(d)
dd = dict()
dd.update(d)
self._not_tracked(dd)
d = dict.fromkeys([x, y, z, o])
self._tracked(d)
dd = dict()
dd.update(d)
self._tracked(dd)

d = dict(x=x, y=y, z=z)
self._not_tracked(d)
d = dict(x=x, y=y, z=z, w=w)
self._tracked(d)
d = dict()
d.update(x=x, y=y, z=z)
self._not_tracked(d)
d.update(w=w)
self._tracked(d)

d = dict([(x, y), (z, 1)])
self._not_tracked(d)
d = dict([(x, y), (z, w)])
self._tracked(d)
d = dict()
d.update([(x, y), (z, 1)])
self._not_tracked(d)
d.update([(x, y), (z, w)])
self._tracked(d)

@support.cpython_only
def test_track_subtypes(self):
# Dict subtypes are always tracked
class MyDict(dict):
pass
self._tracked(MyDict())

def make_shared_key_dict(self, n):
class C:
pass
Expand Down
4 changes: 4 additions & 0 deletions Misc/NEWS.d/next/Core_and_Builtins/2024-11-19-17-17-32.gh-issue-127010.9Cl4bb.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
Simplify GC tracking of dictionaries. All dictionaries are tracked when
created, rather than being lazily tracked when a trackable object was added
to them. This simplifies the code considerably and results in a slight
speedup.
111 changes: 19 additions & 92 deletions Objects/dictobject.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -883,6 +883,7 @@ new_dict(PyInterpreterState *interp,
mp->ma_used = used;
mp->_ma_watcher_tag = 0;
ASSERT_CONSISTENT(mp);
_PyObject_GC_TRACK(mp);
return (PyObject *)mp;
}

Expand Down Expand Up @@ -1578,64 +1579,6 @@ _PyDict_HasOnlyStringKeys(PyObject *dict)
return 1;
}

#define MAINTAIN_TRACKING(mp, key, value) \
do { \
if (!_PyObject_GC_IS_TRACKED(mp)) { \
if (_PyObject_GC_MAY_BE_TRACKED(key) || \
_PyObject_GC_MAY_BE_TRACKED(value)) { \
_PyObject_GC_TRACK(mp); \
} \
} \
} while(0)

void
_PyDict_MaybeUntrack(PyObject *op)
{
PyDictObject *mp;
PyObject *value;
Py_ssize_t i, numentries;

ASSERT_WORLD_STOPPED_OR_DICT_LOCKED(op);

if (!PyDict_CheckExact(op) || !_PyObject_GC_IS_TRACKED(op))
return;

mp = (PyDictObject *) op;
ASSERT_CONSISTENT(mp);
numentries = mp->ma_keys->dk_nentries;
if (_PyDict_HasSplitTable(mp)) {
for (i = 0; i < numentries; i++) {
if ((value = mp->ma_values->values[i]) == NULL)
continue;
if (_PyObject_GC_MAY_BE_TRACKED(value)) {
return;
}
}
}
else {
if (DK_IS_UNICODE(mp->ma_keys)) {
PyDictUnicodeEntry *ep0 = DK_UNICODE_ENTRIES(mp->ma_keys);
for (i = 0; i < numentries; i++) {
if ((value = ep0[i].me_value) == NULL)
continue;
if (_PyObject_GC_MAY_BE_TRACKED(value))
return;
}
}
else {
PyDictKeyEntry *ep0 = DK_ENTRIES(mp->ma_keys);
for (i = 0; i < numentries; i++) {
if ((value = ep0[i].me_value) == NULL)
continue;
if (_PyObject_GC_MAY_BE_TRACKED(value) ||
_PyObject_GC_MAY_BE_TRACKED(ep0[i].me_key))
return;
}
}
}
_PyObject_GC_UNTRACK(op);
}

void
_PyDict_EnablePerThreadRefcounting(PyObject *op)
{
Expand Down Expand Up @@ -1761,7 +1704,6 @@ insert_split_value(PyInterpreterState *interp, PyDictObject *mp, PyObject *key,
{
assert(PyUnicode_CheckExact(key));
ASSERT_DICT_LOCKED(mp);
MAINTAIN_TRACKING(mp, key, value);
PyObject *old_value = mp->ma_values->values[ix];
if (old_value == NULL) {
_PyDict_NotifyEvent(interp, PyDict_EVENT_ADDED, mp, key, value);
Expand Down Expand Up @@ -1818,8 +1760,6 @@ insertdict(PyInterpreterState *interp, PyDictObject *mp,
if (ix == DKIX_ERROR)
goto Fail;

MAINTAIN_TRACKING(mp, key, value);

if (ix == DKIX_EMPTY) {
assert(!_PyDict_HasSplitTable(mp));
/* Insert into new slot. */
Expand Down Expand Up @@ -1878,8 +1818,6 @@ insert_to_emptydict(PyInterpreterState *interp, PyDictObject *mp,
/* We don't decref Py_EMPTY_KEYS here because it is immortal. */
assert(mp->ma_values == NULL);

MAINTAIN_TRACKING(mp, key, value);

size_t hashpos = (size_t)hash & (PyDict_MINSIZE-1);
dictkeys_set_index(newkeys, hashpos, 0);
if (unicode) {
Expand Down Expand Up @@ -4024,8 +3962,7 @@ copy_lock_held(PyObject *o)
split_copy->ma_used = mp->ma_used;
split_copy->_ma_watcher_tag = 0;
dictkeys_incref(mp->ma_keys);
if (_PyObject_GC_IS_TRACKED(mp))
_PyObject_GC_TRACK(split_copy);
_PyObject_GC_TRACK(split_copy);
return (PyObject *)split_copy;
}

Expand Down Expand Up @@ -4060,11 +3997,6 @@ copy_lock_held(PyObject *o)

new->ma_used = mp->ma_used;
ASSERT_CONSISTENT(new);
if (_PyObject_GC_IS_TRACKED(mp)) {
/* Maintain tracking. */
_PyObject_GC_TRACK(new);
}

return (PyObject *)new;
}

Expand Down Expand Up @@ -4351,7 +4283,6 @@ dict_setdefault_ref_lock_held(PyObject *d, PyObject *key, PyObject *default_valu
}
}

MAINTAIN_TRACKING(mp, key, value);
STORE_USED(mp, mp->ma_used + 1);
assert(mp->ma_keys->dk_usable >= 0);
ASSERT_CONSISTENT(mp);
Expand Down Expand Up @@ -4800,16 +4731,8 @@ dict_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
d->ma_keys = Py_EMPTY_KEYS;
d->ma_values = NULL;
ASSERT_CONSISTENT(d);

if (type != &PyDict_Type) {
// Don't track if a subclass tp_alloc is PyType_GenericAlloc()
if (!_PyObject_GC_IS_TRACKED(d)) {
_PyObject_GC_TRACK(d);
}
}
else {
// _PyType_AllocNoTrack() does not track the created object
assert(!_PyObject_GC_IS_TRACKED(d));
if (!_PyObject_GC_IS_TRACKED(d)) {
_PyObject_GC_TRACK(d);
}
return self;
}
Expand Down Expand Up @@ -6746,19 +6669,14 @@ make_dict_from_instance_attributes(PyInterpreterState *interp,
{
dictkeys_incref(keys);
Py_ssize_t used = 0;
Py_ssize_t track = 0;
size_t size = shared_keys_usable_size(keys);
for (size_t i = 0; i < size; i++) {
PyObject *val = values->values[i];
if (val != NULL) {
used += 1;
track += _PyObject_GC_MAY_BE_TRACKED(val);
}
}
PyDictObject *res = (PyDictObject *)new_dict(interp, keys, values, used, 0);
if (track && res) {
_PyObject_GC_TRACK(res);
}
return res;
}

Expand Down Expand Up @@ -7204,6 +7122,7 @@ _PyObject_SetManagedDict(PyObject *obj, PyObject *new_dict)
// since we locked it.
dict = _PyObject_ManagedDictPointer(obj)->dict;
err = _PyDict_DetachFromObject(dict, obj);
assert(err == 0 || new_dict == NULL);
if (err == 0) {
FT_ATOMIC_STORE_PTR(_PyObject_ManagedDictPointer(obj)->dict,
(PyDictObject *)Py_XNewRef(new_dict));
Expand Down Expand Up @@ -7236,7 +7155,21 @@ void
PyObject_ClearManagedDict(PyObject *obj)
{
if (_PyObject_SetManagedDict(obj, NULL) < 0) {
/* Must be out of memory */
assert(PyErr_Occurred() == PyExc_MemoryError);
PyErr_WriteUnraisable(NULL);
/* Clear the dict */
PyDictObject *dict = _PyObject_GetManagedDict(obj);
Py_BEGIN_CRITICAL_SECTION2(dict, obj);
dict = _PyObject_ManagedDictPointer(obj)->dict;
PyInterpreterState *interp = _PyInterpreterState_GET();
PyDictKeysObject *oldkeys = dict->ma_keys;
set_keys(dict, Py_EMPTY_KEYS);
dict->ma_values = NULL;
dictkeys_decref(interp, oldkeys, IS_DICT_SHARED(dict));
STORE_USED(dict, 0);
set_dict_inline_values(obj, NULL);
Py_END_CRITICAL_SECTION2();
}
}

Expand All @@ -7261,12 +7194,6 @@ _PyDict_DetachFromObject(PyDictObject *mp, PyObject *obj)
PyDictValues *values = copy_values(mp->ma_values);

if (values == NULL) {
/* Out of memory. Clear the dict */
PyInterpreterState *interp = _PyInterpreterState_GET();
PyDictKeysObject *oldkeys = mp->ma_keys;
set_keys(mp, Py_EMPTY_KEYS);
dictkeys_decref(interp, oldkeys, IS_DICT_SHARED(mp));
STORE_USED(mp, 0);
PyErr_NoMemory();
return -1;
}
Expand Down
Loading
Loading