[3.14] GH-139951: Fix major GC performance regression (GH-140262)

[miss-islington]

• Count number of actually tracked objects, instead of trackable objects. This ensures that untracking tuples has the desired effect of reducing GC overhead

• Do not track most untrackable tuples during creation. This prevents large numbers of small tuples causing execessive GCs.
  (cherry picked from commit 0c01090)

Co-authored-by: Mark Shannon mark@hotpy.org

• Issue: Potentially serious regression in garbage collector performance in Python 3.14 #139951

[sergey-miryanov]

The following change fixes free-threading tests (mainly -1386,10 +1385,6 fixes tests, other is a cleanup):

diff --git a/Lib/test/test_gc.py b/Lib/test/test_gc.py
index db643330894..d7a4c066437 100644
--- a/Lib/test/test_gc.py+++ b/Lib/test/test_gc.py@@ -3,8 +3,7 @@
from test import support
from test.support import (verbose, refcount_test,
cpython_only, requires_subprocess,
- requires_gil_enabled,- Py_GIL_DISABLED)+ requires_gil_enabled)
from test.support.import_helper import import_module
from test.support.os_helper import temp_dir, TESTFN, unlink
from test.support.script_helper import assert_python_ok, make_script, run_test_script
@@ -1386,10 +1385,6 @@ def callback(ignored):
junk = []
i = 0
detector = GC_Detector()
- if Py_GIL_DISABLED:- # The free-threaded build doesn't have multiple generations, so- # just trigger a GC manually.- gc.collect()
assert not detector.gc_happened
while not detector.gc_happened:
i += 1
@@ -1459,10 +1454,6 @@ def __del__(self):
detector = GC_Detector()
junk = []
i = 0
- if Py_GIL_DISABLED:- # The free-threaded build doesn't have multiple generations, so- # just trigger a GC manually.- gc.collect()
while not detector.gc_happened:
i += 1
if i > 50000:

Is it fine if I add such commit to this PR? Should we apply this change to the main?

[efimov-mikhail]

Сс @colesbury

[colesbury]

The test has a @unittest.skipIf(Py_GIL_DISABLED, "requires GC generations or increments") on main, it should have the same on 3.14

[markshannon]

Closing in favor of #140447