Data race reading `UnicodeDecodeError.start`/`end` in `__str__` under free-threading

[Naserume]

Bug report

Bug description:

On a free-threaded build, UnicodeDecodeError.__str__ reads the start and end members with plain (non-atomic) loads, while assigning to exc.start / exc.end goes through PyMember_SetOne, which does an atomic store. So calling str(exc) on a shared UnicodeDecodeError concurrently with setting its start/end is a data race on those Py_ssize_t members.

start and end are exposed as Py_T_PYSSIZET members:

cpython/Objects/exceptions.c

Lines 3758 to 3761 in 22a6c51

	{"start", Py_T_PYSSIZET, offsetof(PyUnicodeErrorObject, start), 0,
	PyDoc_STR("exception start")},
	{"end", Py_T_PYSSIZET, offsetof(PyUnicodeErrorObject, end), 0,
	PyDoc_STR("exception end")},

Plain read in UnicodeDecodeError_str:

cpython/Objects/exceptions.c

Lines 3947 to 3966 in 22a6c51

	Py_ssize_t len = PyBytes_GET_SIZE(exc->object);
	Py_ssize_t start = exc->start, end = exc->end;
	if ((start >= 0 && start < len) && (end >= 0 && end <= len) && end == start + 1) {
	int badbyte = (int)(PyBytes_AS_STRING(exc->object)[start] & 0xff);
	result = PyUnicode_FromFormat(
	"'%U' codec can't decode byte 0x%02x in position %zd: %U",
	encoding_str,
	badbyte,
	start,
	reason_str);
	}
	else {
	result = PyUnicode_FromFormat(
	"'%U' codec can't decode bytes in position %zd-%zd: %U",
	encoding_str,
	start,
	end - 1,
	reason_str);
	}

Reproducer:

from threading import Thread

shared_exc = UnicodeDecodeError('utf-8', b'\xff\xfe\xfa', 1, 2, 'invalid start byte')

def chain1_thread():
    for i in range(20000):
        try:
            shared_exc.start = i % 3
        except Exception:
            pass

def chain2_thread():
    for _ in range(20000):
        try:
            str(shared_exc)
        except Exception:
            pass

N_C1 = 4
N_C2 = 8
threads  = [Thread(target=chain1_thread) for _ in range(N_C1)]
threads += [Thread(target=chain2_thread) for _ in range(N_C2)]
for t in threads: t.start()
for t in threads: t.join()

TSAN Report:

WARNING: ThreadSanitizer: data race (pid=656224)
  Read of size 8 at 0x7fffb63734e8 by thread T5:
    #0 UnicodeDecodeError_str /cpython/Objects/exceptions.c:3948:29 
    #1 PyObject_Str /cpython/Objects/object.c:826:11 
    #2 unicode_vectorcall /cpython/Objects/unicodeobject.c:14279:16 
    #3 _PyObject_VectorcallTstate /cpython/./Include/internal/pycore_call.h:144:11 
    #4 PyObject_Vectorcall /cpython/Objects/call.c:327:12 
    #5 _Py_VectorCallInstrumentation_StackRefSteal /cpython/Python/ceval.c:768:11 
    #6 _PyEval_EvalFrameDefault /cpython/Python/generated_cases.c.h:1906:35

  Previous atomic write of size 8 at 0x7fffb63734e8 by thread T3:
    #0 PyMember_SetOne /cpython/Python/structmember.c 
    #1 member_set /cpython/Objects/descrobject.c:239:12 
    #2 _PyObject_GenericSetAttrWithDict /cpython/Objects/object.c:2049:19 
    #3 PyObject_GenericSetAttr /cpython/Objects/object.c:2120:12 
    #4 PyObject_SetAttr /cpython/Objects/object.c:1533:15 
    #5 _PyEval_EvalFrameDefault /cpython/Python/generated_cases.c.h:12057:27

SUMMARY: ThreadSanitizer: data race/cpython/Objects/exceptions.c:3948:29 in UnicodeDecodeError_str

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

[picnixz]

When I implemented this I did not consider this kind of usage as I believe it was un-necessary and a GIGO case but we can make this safe.