`PyCode_GetCode` is not thread-safe and causes assertion fail with Python 3.13td

[XuehaiPan]

Crash report

What happened?

Race condition here:

cpython/Objects/codeobject.c

Lines 1663 to 1665 in 60403a5

	deopt_code(co, (_Py_CODEUNIT *)PyBytes_AS_STRING(code));
	assert(co->_co_cached->_co_code == NULL);
	co->_co_cached->_co_code = Py_NewRef(code);

Core dump and backtrace:

https://github.com/metaopt/optree/actions/runs/11913729282/job/33200071659#step:15:172

Core was generated by `python -X dev -m pytest --verbose --color=yes --durations=10 --showlocals --cov'.
Program terminated with signal SIGABRT, Aborted.
#0  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140335707584064) at ./nptl/pthread_kill.c:44
[Current thread is 1 (Thread 0x7fa273fff640 (LWP 2491))]
#0  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140335707584064) at ./nptl/pthread_kill.c:44
#1  __pthread_kill_internal (signo=6, threadid=140335707584064) at ./nptl/pthread_kill.c:78
#2  __GI___pthread_kill (threadid=140335707584064, signo=signo@entry=6) at ./nptl/pthread_kill.c:89
#3  0x00007fa27b642476 in __GI_raise (sig=6) at ../sysdeps/posix/raise.c:26
#4  0x00007fa27be3f0f0 in faulthandler_fatal_error (signum=6) at ./Modules/faulthandler.c:338
#5  <signal handler called>
#6  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140335707584064) at ./nptl/pthread_kill.c:44
#7  __pthread_kill_internal (signo=6, threadid=140335707584064) at ./nptl/pthread_kill.c:78
#8  __GI___pthread_kill (threadid=140335707584064, signo=signo@entry=6) at ./nptl/pthread_kill.c:89
#9  0x00007fa27b642476 in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26
#10 0x00007fa27b6287f3 in __GI_abort () at ./stdlib/abort.c:79
#11 0x00007fa27b62871b in __assert_fail_base (fmt=0x7fa27b7dd130 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", 
    assertion=0x7fa27bf0b730 "co->_co_cached->_co_code == NULL", file=0x7fa27bf0b174 "Objects/codeobject.c", line=1664, 
    function=<optimized out>) at ./assert/assert.c:92
#12 0x00007fa27b639e96 in __GI___assert_fail (assertion=0x7fa27bf0b730 "co->_co_cached->_co_code == NULL", 
    file=0x7fa27bf0b174 "Objects/codeobject.c", line=1664, function=0x7fa27bf0c0e0 <__PRETTY_FUNCTION__.13> "_PyCode_GetCode")
    at ./assert/assert.c:101
#13 0x00007fa27bb7f165 in _PyCode_GetCode (co=0x200023271d0) at Objects/codeobject.c:1664
#14 0x00007fa27bb7f1a5 in PyCode_GetCode (co=0x200023271d0) at Objects/codeobject.c:1672
#15 0x00007fa27b1b2547 in CTracer_handle_call (frame=0x20026051010, self=0x20026070110) at coverage/ctracer/tracer.c:557
#16 CTracer_trace (self=0x20026070110, frame=0x20026051010, what=0, arg_unused=<optimized out>)
    at coverage/ctracer/tracer.c:844
#17 0x00007fa27bdd738d in call_trace_func (self=0x20002152ea0, arg=0x7fa27c14fc60 <_Py_NoneStruct>)
    at Python/legacy_tracing.c:189
#18 0x00007fa27bdd75fa in sys_trace_start (self=0x20002152ea0, args=0x7fa273ffcdf8, nargsf=9223372036854775810, kwnames=0x0)
    at Python/legacy_tracing.c:229
#19 0x00007fa27bdcc0a5 in _PyObject_VectorcallTstate (tstate=0x5620713a56e0, callable=0x20002152ea0, args=0x7fa273ffcdf8, 
    nargsf=9223372036854775810, kwnames=0x0) at ./Include/internal/pycore_call.h:168
#20 0x00007fa27bdce182 in call_one_instrument (interp=0x7fa27c1975c0 <_PyRuntime+128640>, tstate=0x5620713a56e0, 
    args=0x7fa273ffcdf8, nargsf=9223372036854775810, tool=7 '\a', event=0) at Python/instrumentation.c:907
#21 0x00007fa27bdcea63 in call_instrumentation_vector (tstate=0x5620713a56e0, event=0, frame=0x7fa27b0d8350, 
    instr=0x200023272aa, nargs=2, args=0x7fa273ffcdf0) at Python/instrumentation.c:1095
#22 0x00007fa27bdcec5d in _Py_call_instrumentation (tstate=0x5620713a56e0, event=0, frame=0x7fa27b0d8350, instr=0x200023272aa)
    at Python/instrumentation.c:1132
#23 0x00007fa27bd46096 in _PyEval_EvalFrameDefault (tstate=0x5620713a56e0, frame=0x7fa27b0d8350, throwflag=0)
    at Python/generated_cases.c.h:3474
#24 0x00007fa27bd32d31 in _PyEval_EvalFrame (tstate=0x5620713a56e0, frame=0x7fa27b0d8020, throwflag=0)
    at ./Include/internal/pycore_ceval.h:119
#25 0x00007fa27bd56877 in _PyEval_Vector (tstate=0x5620713a56e0, func=0x200012a69d0, locals=0x0, args=0x7fa273ffec40, 
    argcount=1, kwnames=0x0) at Python/ceval.c:1806
#26 0x00007fa27bb7271e in _PyFunction_Vectorcall (func=0x200012a69d0, stack=0x7fa273ffec40, nargsf=1, kwnames=0x0)
    at Objects/call.c:413
#27 0x00007fa27bb765f3 in _PyObject_VectorcallTstate (tstate=0x5620713a56e0, callable=0x200012a69d0, args=0x7fa273ffec40, 
    nargsf=1, kwnames=0x0) at ./Include/internal/pycore_call.h:168
#28 0x00007fa27bb76c80 in method_vectorcall (method=0x200248b8890, args=0x7fa27c197548 <_PyRuntime+128520>, nargsf=0, 
    kwnames=0x0) at Objects/classobject.c:70
#29 0x00007fa27bb7208f in _PyVectorcall_Call (tstate=0x5620713a56e0, func=0x7fa27bb76a66 <method_vectorcall>, 
    callable=0x200248b8890, tuple=0x7fa27c197520 <_PyRuntime+128480>, kwargs=0x0) at Objects/call.c:273
#30 0x00007fa27bb7243c in _PyObject_Call (tstate=0x5620713a56e0, callable=0x200248b8890, 
    args=0x7fa27c197520 <_PyRuntime+128480>, kwargs=0x0) at Objects/call.c:348
#31 0x00007fa27bb72517 in PyObject_Call (callable=0x200248b8890, args=0x7fa27c197520 <_PyRuntime+128480>, kwargs=0x0)
    at Objects/call.c:373
#32 0x00007fa27bec8297 in thread_run (boot_raw=0x5620712f0d80) at ./Modules/_threadmodule.c:337
#33 0x00007fa27be210f7 in pythread_wrapper (arg=0x5620712f0f30) at Python/thread_pthread.h:243
#34 0x00007fa27b694ac3 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:442
#35 0x00007fa27b726850 in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81

CPython versions tested on:

3.13

Operating systems tested on:

Linux

Output from running 'python -VV' on the command line:

Python 3.13.0 experimental free-threading build

Linked PRs

• gh-127020: Make PyCode_GetCode thread-safe for free threading #127043
• [3.13] gh-127020: Make PyCode_GetCode thread-safe for free threading (GH-127043) #127107

[colesbury]

Thanks for the bug report @XuehaiPan. I think the linked PR should fix the crash. I verified it with the following command when running optree tests:

python -X dev -m pytest --verbose --color=yes --durations=10 --showlocals --cov="optree" --cov-config=.coveragerc --cov-report=xml --cov-report=term-missing --exitfirst --cov-report=xml:coverage-cp313td-Linux.xml --junit-xml=junit-cp313td-Linux.xml .

[XuehaiPan]

I think the linked PR should fix the crash. I verified it with the following command when running optree tests.

Thanks for the fix. I can confirm the unit test in #127043 fails on the main branch and passes with the patch in #127043.

$ python3 -VV
Python 3.14.0a2+ experimental free-threading build (heads/main:c9b399fbdb0, Nov 20 2024, 15:18:27) [Clang 16.0.0 (clang-1600.0.26.4)]
$ python3 test_code.py
Assertion failed: (co->_co_cached->_co_code == NULL), function _PyCode_GetCode, file codeobject.c, line 1686.
[1]    37607 abort      python3 test_code.py

$ python3 -VV
Python 3.14.0a2+ experimental free-threading build (heads/pr/colesbury/127043:9242a83444c, Nov 20 2024, 15:13:34) [Clang 16.0.0 (clang-1600.0.26.4)]
$ python3 test_code.py
.
----------------------------------------------------------------------
Ran 1 test in 0.019s

OK

[Eclips4]

I guess it can be closed now?