python / python/cpython

Segfault & Debug abort when functools.partial/operator.methodcaller keywords contains non-string keys

オープン
#155,733 コメント 4 件 リアクション 0 件 担当者 0 名 GitHub で見る

まだ誰も着手していません。

3.14 3.15 3.16 extension-modules type-crash
主要言語
Python
スター
77.2k
フォーク
35.9k
PR マージ指標
PR 指標を取得中

説明

Crash report

What happened?
>>> import operator
>>> operator.methodcaller('x', **{1: 'x'})
fish: Job 1, 'python' terminated by signal SIGSEGV (Address boundary error)

and

>>> import functools
>>> functools.partial(print, **{1: "x"})()
Assertion failed: (PyUnicode_Check(kwname)), function find_keyword, file getargs.c, line 2271.

also:

import functools

p = functools.partial(print)
p.keywords[1] = 'x'
p()

type_call() does not validate kwargs contents, passing them through to tp_new/tp_init as-is. Most of the time, PyArg_ParseTupleAndKeywords gets run pretty soon, and this catches any non-str keywords, but if for any reason this doesn't happen, then we get non-str keywords in places that assume all keywords are str.

Partial Stack
    frame #4: 0x0000fffff7d413ac libc.so.6`__assert_fail + 204
    frame #5: 0x0000aaaaab398170 python3`find_keyword(kwnames=0x0000fffff7200190, kwstack=0x0000ffffffffbb70, key=0x0000aaaaab7d51b0) at getargs.c:2271:9
    frame #6: 0x0000aaaaab3995c0 python3`_PyArg_UnpackKeywords(args=0x0000ffffffffbb70, nargs=0, kwargs=0x0000000000000000, kwnames=<unavailable>, parser=<unavailable>, minpos=0, maxpos=0, minkw=0, varpos=1, buf=0x0000ffffffffb958) at getargs.c:2636:31
    frame #7: 0x0000aaaaab267120 python3`builtin_print(module=0x0000fffff7c60200, args=0x0000ffffffffbb70, nargs=0, kwnames=0x0000fffff7200190) at bltinmodule.c.h:1094:16
    frame #8: 0x0000aaaaaafe9850 python3`cfunction_vectorcall_FASTCALL_KEYWORDS(func=0x0000fffff7c56f30, args=0x0000ffffffffbb70, nargsf=0, kwnames=0x0000fffff7200190) at methodobject.c:465:24
    frame #9: 0x0000aaaaab6e4d28 python3`_PyObject_VectorcallTstate(tstate=0x0000aaaaab7f98e0, callable=0x0000fffff7c56f30, args=0x0000ffffffffbb70, nargsf=0, kwnames=0x0000fffff7200190) at pycore_call.h:144:11
    frame #10: 0x0000aaaaab6e42a8 python3`partial_vectorcall(self=0x0000fffff71b09d0, args=<unavailable>, nargsf=<unavailable>, kwnames=0x0000000000000000) at _functoolsmodule.c:555:14
    frame #11: 0x0000aaaaaaeb0b64 python3`_PyObject_VectorcallTstate(tstate=0x0000aaaaab7f98e0, callable=0x0000fffff71b09d0, args=0x0000ffffffffbcd0, nargsf=9223372036854775808, kwnames=0x0000000000000000) at pycore_call.h:144:11
    frame #12: 0x0000aaaaab2d7d64 python3`PyObject_Vectorcall(callable=<unavailable>, args=0x0000ffffffffbcd0, nargsf=<unavailable>, kwnames=0x0000000000000000) at call.c:327:12 [inlined]
    frame #13: 0x0000aaaaab2d7d54 python3`_Py_VectorCallInstrumentation_StackRefSteal(callable=<unavailable>, arguments=0x0000fffff7fe94a8, total_args=<unavailable>, kwnames=(bits = 1), call_instrumentation=false, frame=0x0000fffff7fe9448, this_instr=0x0000fffff7c5c378, tstate=0x0000aaaaab7f98e0) at ceval.c:768:11
    frame #14: 0x0000aaaaab28436c python3`_PyEval_EvalFrameDefault(tstate=<unavailable>, frame=<unavailable>, throwflag=<unavailable>) at generated_cases.c.h:1906:35
    frame #15: 0x0000aaaaab270f54 python3`_PyEval_EvalFrame(tstate=0x0000aaaaab7f98e0, frame=0x0000fffff7fe9448, throwflag=0) at pycore_ceval.h:122:16 [inlined]
    frame #16: 0x0000aaaaab270f2c python3`_PyEval_Vector(tstate=<unavailable>, func=<unavailable>, locals=0x0000fffff7acd610, args=<unavailable>, argcount=<unavailable>, kwnames=0x0000000000000000) at ceval.c:2172:12
    frame #17: 0x0000aaaaab2708f0 python3`PyEval_EvalCode(co=0x0000fffff7c5c280, globals=0x0000fffff7acd610, locals=0x0000fffff7acd610) at ceval.c:679:21
    frame #18: 0x0000aaaaab262c38 python3`builtin_exec_impl(module=<unavailable>, source=0x0000fffff7c5c280, globals=0x0000fffff7acd610, locals=0x0000fffff7acd610, closure=<unavailable>) at bltinmodule.c:1267:17
    frame #19: 0x0000aaaaab2629b0 python3`builtin_exec(module=<unavailable>, args=0x0000ffffffffcf98, nargs=<unavailable>, kwnames=0x0000000000000000) at bltinmodule.c.h:676:20
    frame #20: 0x0000aaaaab2da030 python3`_Py_BuiltinCallFastWithKeywords_StackRef(callable=(bits = 281474838652656), arguments=0x0000fffff7fe9438, total_args=2) at ceval.c:841:11
    frame #21: 0x0000aaaaab288600 python3`_PyEval_EvalFrameDefault(tstate=<unavailable>, frame=<unavailable>, throwflag=<unavailable>) at generated_cases.c.h:2603:35
    frame #22: 0x0000aaaaab270f54 python3`_PyEval_EvalFrame(tstate=0x0000aaaaab7f98e0, frame=0x0000fffff7fe9020, throwflag=0) at pycore_ceval.h:122:16 [inlined]
    frame #23: 0x0000aaaaab270f2c python3`_PyEval_Vector(tstate=<unavailable>, func=<unavailable>, locals=0x0000000000000000, args=<unavailable>, argcount=<unavailable>, kwnames=0x0000fffff77962d0) at ceval.c:2172:12
    frame #24: 0x0000aaaaaaeb31d0 python3`_PyVectorcall_Call(tstate=0x0000aaaaab7f98e0, func=<unavailable>, callable=0x0000fffff71cb890, tuple=0x0000aaaaab7da228, kwargs=0x0000fffff71c43b0) at call.c:285:24
    frame #25: 0x0000aaaaab526178 python3`PyObject_Call(callable=0x0000fffff71cb890, args=<unavailable>, kwargs=0x0000fffff71c43b0) at call.c:373:12 [inlined]

I wasn't sure wether to raise one or two issues here, as I think it's likely that there will have to be two fixes at the individual call sites, but they're the same underlying problem.

Additional context

As the functools.partial call actually calls the wrapped function with the bad argument names, there are a few functions that get tripped up here:

  • print as in the example above
  • sorted
  • enumerate
  • sqlite3.connect
    for example:
>>> functools.partial(enumerate, **{1: None})([])
Objects/unicodeobject.c:606: _PyUnicode_CheckConsistency: Assertion failed: PyType_HasFeature((_Py_TYPE_impl(((PyObject*)((op))))), ((1UL << 28)))
Enable tracemalloc to get the memory block allocation traceback

object address  : 0x100e94800
object refcount : 3221225472
object type     : 0x100e5ec80
object type name: int
object repr     : 1

Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed
Python runtime state: initialized

Current thread 0x00000002051c7b40 (most recent call first):
  File "<python-input-2>", line 1 in <module>
  File "/Users/sstagg/src/tp/cpython/Lib/_pyrepl/console.py", line 235 in runcode
  File "/Users/sstagg/src/tp/cpython/Lib/_pyrepl/console.py", line 292 in runsource
  File "/Users/sstagg/src/tp/cpython/Lib/code.py", line 324 in push
  File "/Users/sstagg/src/tp/cpython/Lib/_pyrepl/simple_interact.py", line 151 in run_multiline_interactive_console
  File "/Users/sstagg/src/tp/cpython/Lib/_pyrepl/main.py", line 58 in interactive_console
fish: Job 1, './python.exe' terminated by signal SIGABRT (Abort)
CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

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

Python 3.16.0a0 (heads/main:9f878002205, Aug 13 2026, 15:07:54) [Clang 21.0.0 (clang-2100.3.27.1)]

Linked PRs
  • gh-155779

コントリビューションガイド

コントリビューションガイドを開く

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

調査の方向性

_functoolsmodule.c の type_call() と partial_vectorcall パスから始め、次に getargs.c と operator.methodcaller パスに示されているキーワードに関する前提を確認します。CPython main 上で functools.partial と operator.methodcaller の例を再現します。文字列ではないキーワードキーが segfault や abort なしに拒否されれば完了です。

索引モデルが issue の本文から書いたものです。

評価

技術スタック
python
領域
backend
issue の種類
バグ
難易度
4/5
見積もり時間
3〜5日
活発さ
停滞
明瞭さ
おおむね明確
初心者へのやさしさ
25/100

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。