GeezerVison is showing github.com — Segfault from `unicodedata.iter_graphemes` on a free threaded build · Issue #145202 · python/cpython
Segfault from unicodedata.iter_graphemes on a free threaded build #145202DescriptionCrash report
What happened?
A free-threaded build will segfault when calling unicodedata.iter_graphemes() and storing the result, even with PYTHON_GIL=0.
import unicodedata
res = unicodedata.iter_graphemes("a")Backtrace:
Program received signal SIGSEGV, Segmentation fault.
GBI_dealloc (self=0x20002ac2b60) at ./Modules/unicodedata.c:1992
1992 Py_DECREF(((GraphemeBreakIterator *)self)->iter.str);
#0 GBI_dealloc (self=0x20002ac2b60) at ./Modules/unicodedata.c:1992
#1 0x000055555573b0b7 in _Py_Dealloc (op=0x20002ac2b60) at Objects/object.c:3271
#2 0x000055555570cbce in Py_DECREF (lineno=513, op=0x20002ac2b60, filename=<optimized out>) at ./Include/refcount.h:356
#3 Py_XDECREF (op=0x20002ac2b60) at ./Include/refcount.h:513
#4 dictkeys_decref (dk=dk@entry=0x200020d4a10, use_qsbr=true) at Objects/dictobject.c:494
#5 0x000055555571132b in clear_lock_held (op=op@entry=0x2000270ccd0) at Objects/dictobject.c:2998
#6 0x0000555555711578 in PyDict_Clear (op=0x2000270ccd0) at Objects/dictobject.c:3027
#7 0x00005555557174a9 in dict_tp_clear (op=0x20002ac2b60) at Objects/dictobject.c:4843
#8 0x00005555558f5c61 in delete_garbage (state=0x7fffffffd5c0) at Python/gc_free_threading.c:1780
#9 gc_collect_internal (interp=0x555555d33880 <_PyRuntime+174016>, state=0x7fffffffd5c0, generation=2) at Python/gc_free_threading.c:2375
#10 gc_collect_main (tstate=0x555555d6bfd8 <_PyRuntime+405272>, generation=2, reason=_Py_GC_REASON_SHUTDOWN) at Python/gc_free_threading.c:2446
#11 0x000055555594203e in finalize_modules (tstate=tstate@entry=0x555555d6bfd8 <_PyRuntime+405272>) at Python/pylifecycle.c:1793
#12 0x000055555594081a in _Py_Finalize (runtime=<optimized out>) at Python/pylifecycle.c:2295
#13 0x00005555559aa459 in Py_RunMain () at Modules/main.c:774
#14 0x00005555559aafe1 in pymain_main (args=args@entry=0x7fffffffda40) at Modules/main.c:802
#15 0x00005555559ab042 in Py_BytesMain (argc=<optimized out>, argv=0xffffffffffffffe8) at Modules/main.c:826
#16 0x00007ffff7c2a575 in __libc_start_call_main (main=main@entry=0x5555555ed170 <main>, argc=argc@entry=2, argv=argv@entry=0x7fffffffdb98) at ../sysdeps/nptl/libc_start_call_main.h:58
#17 0x00007ffff7c2a628 in __libc_start_main_impl (main=0x5555555ed170 <main>, argc=2, argv=0x7fffffffdb98, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>,
stack_end=0x7fffffffdb88) at ../csu/libc-start.c:360
#18 0x00005555555ed0a5 in _start ()
Found using fusil by @vstinner.
CPython versions tested on:
CPython main branch
Operating systems tested on:
Linux
Output from running 'python -VV' on the command line:
Python 3.15.0a6+ free-threading build (heads/main:1ac9d138ae0, Feb 25 2026, 05:27:56) [Clang 21.1.2 (2ubuntu6)]
Linked PRs
Activityserhiy-storchaka commented on Feb 25, 2026 Good catch. I think that remove the tp_clear slots would be a better solution. These objects are virtually immutable and cannot create a loop (they can only participate in a loop created by a str subclass). added a commit that references this issue on Feb 26, 2026gh-145202: Fix crash in unicodedata's GraphemeBreakIterator and Segme…
MetadataMetadataDevelopmentNo branches or pull requestsIssue actions
|