Possible (benign) overflow for 'K' format code in do_mkvalue

[picnixz]

Bug report

Bug description:

This is probably not an issue but here are the known "temporary" overflows:

// format = 'K'
return PyLong_FromUnsignedLongLong((long long)va_arg(*p_va, unsigned long long));

Note that va_arg(*p_va, unsigned long long) will be converted into an unsigned long long, and then into a long long, and then back to an unsigned long long. So if we were to have an overflow here, it shouldn't really matter. Indeed, take v = (1ULL << 63) + 2. We have:

unsigned long long v2 = va_arg(*p_va, unsigned long long);
// v2 = 9223372036854775810
long long v3 = (long long)v2;
// v3 = -9223372036854775806
PyObject *v4 = PyLong_FromUnsignedLongLong(v3);
// v4 = 9223372036854775810 as PyObject as v3 was casted back to an unsigned long long

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-132909: handle overflow for 'K' format in do_mkvalue #132911
• [3.13] gh-132909: handle overflow for 'K' format in do_mkvalue (GH-132911) #132932

[serhiy-storchaka]

wchar_t is at least 2 bytes on platforms supported by Python. wcslen() never returns result larger than SSIZE_MAX = SIZE_MAX/2.