Skip to content

PyRefTracer_DESTROY is not called for objects deferred by the trashcan mechanism #156371

Description

@corona10

Bug report

This works correctly until Python 3.13, but #132280 introduces a regression.
A simple PR will fix this issue, and I have already prepared a fix for this issue.

import os
import subprocess
import sys
import sysconfig
import tempfile

CSRC = r"""
#include <Python.h>

static Py_ssize_t list_destroys = 0;

static int
tracer(PyObject *obj, PyRefTracerEvent event, void *data)
{
    if (event == PyRefTracer_DESTROY && PyList_CheckExact(obj)) {
        list_destroys++;
    }
    return 0;
}

static PyObject *
start(PyObject *self, PyObject *args)
{
    list_destroys = 0;
    if (PyRefTracer_SetTracer(tracer, NULL) < 0) {
        return NULL;
    }
    Py_RETURN_NONE;
}

static PyObject *
stop(PyObject *self, PyObject *args)
{
    if (PyRefTracer_SetTracer(NULL, NULL) < 0) {
        return NULL;
    }
    return PyLong_FromSsize_t(list_destroys);
}

static PyMethodDef methods[] = {
    {"start", start, METH_NOARGS, NULL},
    {"stop", stop, METH_NOARGS, NULL},
    {NULL, NULL, 0, NULL},
};

static PyModuleDef_Slot slots[] = {
#ifdef Py_GIL_DISABLED
    {Py_mod_gil, Py_MOD_GIL_NOT_USED},
#endif
    {0, NULL},
};

static struct PyModuleDef module = {
    PyModuleDef_HEAD_INIT,
    .m_name = "reftrace_repro",
    .m_size = 0,
    .m_methods = methods,
    .m_slots = slots,
};

PyMODINIT_FUNC
PyInit_reftrace_repro(void)
{
    return PyModuleDef_Init(&module);
}
"""


def build_module():
    tmpdir = tempfile.mkdtemp(prefix="reftrace_repro_")
    src = os.path.join(tmpdir, "reftrace_repro.c")
    with open(src, "w") as f:
        f.write(CSRC)
    ext = os.path.join(tmpdir, "reftrace_repro.so")
    cmd = ["cc", "-O2", "-shared", "-fPIC",
           "-I", sysconfig.get_path("include"),
           "-I", sysconfig.get_path("platinclude"),
           "-o", ext, src]
    if sys.platform == "darwin":
        cmd.insert(3, "-undefined")
        cmd.insert(4, "dynamic_lookup")
    subprocess.run(cmd, check=True)
    sys.path.insert(0, tmpdir)


def run(depth):
    import reftrace_repro

    chain = None
    for _ in range(depth):
        chain = [chain]
    reftrace_repro.start()
    del chain
    destroys = reftrace_repro.stop()
    missing = depth - destroys
    print(f"depth={depth:>7}  DESTROY events={destroys:>7}  "
          f"missing={missing:>5}  [{'BUG' if missing else 'OK'}]")


if __name__ == "__main__":
    print(sys.version)
    build_module()
    run(100)        
    run(200_000)

Status

$ python3.13 repro.py
3.13.11 (main, Dec 17 2025, 20:55:16) [Clang 21.1.4 ]
depth=    100  DESTROY events=    100  missing=    0  [OK]
depth= 200000  DESTROY events= 200000  missing=    0  [OK]

$ python3.14 repro.py
3.14.3 (main, Feb  3 2026, 15:32:20) [Clang 17.0.0 (clang-1700.6.3.2)]
depth=    100  DESTROY events=    100  missing=    0  [OK]
depth= 200000  DESTROY events= 199923  missing=   77  [BUG]

$ python3.15 repro.py
3.15.0a3 (main, Dec 17 2025, 21:18:28) [Clang 21.1.4 ]
depth=    100  DESTROY events=    100  missing=    0  [OK]
depth= 200000  DESTROY events= 199570  missing=  430  [BUG]

$ ./python.exe repro.py   # main (unpatched)
3.16.0a0 (heads/main:e4b22adf249, Aug 26 2026, 01:02:19) [Clang 21.0.0]
depth=    100  DESTROY events=    100  missing=    0  [OK]
depth= 200000  DESTROY events= 199923  missing=   77  [BUG]

Linked PRs

Activity

  1. self-assigned this
    on Aug 25, 2026
  2. added
    type-bugAn unexpected behavior, bug, or error
    3.14bugs and security fixes
    3.15bugs and security fixes
    3.16new features, bugs and security fixes
    on Aug 25, 2026
  3. markshannon commented on Aug 26, 2026

    @markshannon
    Member

    OOI, how did you find this issue?

  4. corona10 commented on Aug 26, 2026

    @corona10
    MemberAuthor

    OOI, how did you find this issue?

    While investigating other issues with Claude Code, I came across this issue. :)
    To clarify, I independently validated the regression and the analysis, and I also directed the proposed solution myself, so this is not AI-generated slop.

  5. added 5 commits that reference this issue on Aug 26, 2026
  6. added a commit that references this issue on Sep 12, 2026
  7. added a commit that references this issue on Sep 30, 2026
  8. added a commit that references this issue on Oct 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

3.14bugs and security fixes3.15bugs and security fixes3.16new features, bugs and security fixestype-bugAn unexpected behavior, bug, or error

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions