`crashers/mutation_inside_cyclegc.py` hangs and sometimes segfaults on the free-threading build

[tomasr8]

Crash report

What happened?

I ran into this while working on #126360

Running ./python Lib/test/crashers/mutation_inside_cyclegc.py built with --with-pydebug --disable-gil
hangs and randomly segfaults. I managed to get the backtrace from gdb:

gdb output

(gdb) run ./Lib/test/crashers/mutation_inside_cyclegc.py 
Starting program: /home/tomas/dev/cpython/python ./Lib/test/crashers/mutation_inside_cyclegc.py
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
visit_decref (op=<unknown at remote 0x28080fbd450>, arg=0x0) at Python/gc_free_threading.c:442
442         if (_PyObject_GC_IS_TRACKED(op) && !_Py_IsImmortal(op)) {
(gdb) bt
#0  visit_decref (op=<unknown at remote 0x28080fbd450>, arg=0x0) at Python/gc_free_threading.c:442
#1  0x00005555556bfad9 in list_traverse (
    self=[<weakref.ReferenceType at remote 0x20000573e90>, ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], [...(truncated), visit=0x5555558a2b58 <visit_decref>, arg=0x0) at Objects/listobject.c:3286
#2  0x00005555558a5f2d in update_refs (heap=<optimized out>, area=<optimized out>, block=<optimized out>, block_size=<optimized out>, args=<optimized out>) at Python/gc_free_threading.c:509
#3  0x0000555555715d7a in _mi_heap_area_visit_blocks (area=area@entry=0x7fffffffcff0, page=0x200000018a8, visitor=0x5555558a5e62 <update_refs>, arg=0x7fffffffd200) at Objects/mimalloc/heap.c:566
#4  0x000055555571650d in mi_heap_area_visitor (heap=heap@entry=0x555555c95140 <_PyRuntime+366400>, xarea=xarea@entry=0x7fffffffcff0, arg=arg@entry=0x7fffffffd0d0) at Objects/mimalloc/heap.c:681
#5  0x0000555555715a93 in mi_heap_visit_areas_page (heap=0x555555c95140 <_PyRuntime+366400>, pq=<optimized out>, page=<optimized out>, vfun=0x5555557164c9 <mi_heap_area_visitor>, arg=0x7fffffffd0d0) at Objects/mimalloc/heap.c:661
#6  0x000055555570bd28 in mi_heap_visit_pages (heap=heap@entry=0x555555c95140 <_PyRuntime+366400>, fn=fn@entry=0x555555715a4a <mi_heap_visit_areas_page>, arg1=arg1@entry=0x5555557164c9 <mi_heap_area_visitor>, arg2=arg2@entry=0x7fffffffd0d0)
    at Objects/mimalloc/heap.c:46
#7  0x000055555570bd93 in mi_heap_visit_areas (heap=heap@entry=0x555555c95140 <_PyRuntime+366400>, visitor=visitor@entry=0x5555557164c9 <mi_heap_area_visitor>, arg=arg@entry=0x7fffffffd0d0) at Objects/mimalloc/heap.c:667
#8  0x000055555571be62 in mi_heap_visit_blocks (heap=heap@entry=0x555555c95140 <_PyRuntime+366400>, visit_blocks=visit_blocks@entry=true, visitor=visitor@entry=0x5555558a5e62 <update_refs>, arg=arg@entry=0x7fffffffd200) at Objects/mimalloc/heap.c:692
#9  0x00005555558a2d8e in gc_visit_heaps_lock_held (interp=interp@entry=0x555555c5b280 <_PyRuntime+129152>, visitor=visitor@entry=0x5555558a5e62 <update_refs>, arg=arg@entry=0x7fffffffd200) at Python/gc_free_threading.c:309
#10 0x00005555558a2e71 in gc_visit_heaps (interp=interp@entry=0x555555c5b280 <_PyRuntime+129152>, visitor=visitor@entry=0x5555558a5e62 <update_refs>, arg=arg@entry=0x7fffffffd200) at Python/gc_free_threading.c:348
#11 0x00005555558a36da in deduce_unreachable_heap (interp=interp@entry=0x555555c5b280 <_PyRuntime+129152>, state=state@entry=0x7fffffffd200) at Python/gc_free_threading.c:688
#12 0x00005555558a6c88 in gc_collect_internal (interp=0x555555c5b280 <_PyRuntime+129152>, state=state@entry=0x7fffffffd200, generation=generation@entry=0) at Python/gc_free_threading.c:1235
#13 0x00005555558a6e7a in gc_collect_main (tstate=tstate@entry=0x555555c937e0 <_PyRuntime+359904>, generation=generation@entry=0, reason=reason@entry=_Py_GC_REASON_HEAP) at Python/gc_free_threading.c:1349
#14 0x00005555558a7228 in _Py_RunGC (tstate=tstate@entry=0x555555c937e0 <_PyRuntime+359904>) at Python/gc_free_threading.c:1799
#15 0x00005555558b133e in _Py_HandlePending (tstate=tstate@entry=0x555555c937e0 <_PyRuntime+359904>) at Python/ceval_gil.c:1297
#16 0x000055555583e6cc in _PyEval_EvalFrameDefault (tstate=tstate@entry=0x555555c937e0 <_PyRuntime+359904>, frame=<optimized out>, throwflag=throwflag@entry=0) at Python/generated_cases.c.h:1003
#17 0x000055555586b756 in _PyEval_EvalFrame (throwflag=0, frame=<optimized out>, tstate=0x555555c937e0 <_PyRuntime+359904>) at ./Include/internal/pycore_ceval.h:116
#18 _PyEval_Vector (tstate=tstate@entry=0x555555c937e0 <_PyRuntime+359904>, func=func@entry=0x20000a52cb0, 
    locals=locals@entry={'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <SourceFileLoader(name='__main__', path='/home/tomas/dev/cpython/./Lib/test/crashers/mutation_inside_cyclegc.py') at remote 0x20000330a20>, '__spec__': None, '__builtins__': <module at remote 0x2000025c1e0>, '__file__': '/home/tomas/dev/cpython/./Lib/test/crashers/mutation_inside_cyclegc.py', '__cached__': None, 'weakref': <module at remote 0x20000798190>, 'A': <type at remote 0x20000c44b10>, 'callback': <function at remote 0x20000a5a030>, 'keepalive': [<weakref.ReferenceType at remote 0x20000573e90>, ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'], ['0'],

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

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

No response

[ZeroIntensity]

Well, isn't the point of anything inside crashers/ that it crashes? I find it odd that this doesn't crash on the GIL-ful builds.

[tomasr8]

I should clarify, this example does not crash anymore (and probably hasn't for a while) on the normal build. The crasher also predates free-threading and the crash seems to be related to the way GC works in free-threading, though I don't know enough about that topic and hence this issue 🙂

[ZeroIntensity]

I've been doing some work on the free-threaded GC recently, I can investigate. Though, we probably should add tests to make sure that the crashers actually crash.

[tomasr8]

That'd be great, thanks!

Though, we probably should add tests to make sure that the crashers actually crash.

See the original issue for more context: #121921 I started by moving non-crashers out and into unit tests to ensure we don't regress, but having tests for crashers might also make sense.

[zware]

There's some history around running the crashers that should be dug up and understood before re-enabling them in automated testing. I particularly don't want them running on buildbots unless we can guarantee (better than we used to :)) that they're not going to leave things in a mess, but they could possibly make sense in GHA.