Improve size validation in _Py_DecodeUTF8Ex

[stoeckmann]

The size check in _Py_DecodeUTF8Ex can be improved to always check against a constant value without further arithmetic involved. This is already done at other places within the file, e.g. here.

I was curious if this could actually be triggered with a proof of concept by overflowing the check and eventually performing an out of boundary heap access. And in fact, with a very artificial setup, it is possible on a 32 bit system which tries to convert a 2 GB long string:

#include "Python.h"

#include <sys/mman.h>

#include <err.h>
#include <stdlib.h>

int
main(int argc, char *argv[])
{
        char *str;
        size_t wlen;
        wchar_t *program;

        // force UTF-8 mode
        Py_UTF8Mode = 1;
        if ((program = Py_DecodeLocale(argv[0], NULL)) == NULL)
                errx(1, "PyDecodeLocale");
        Py_SetProgramName(program);
        Py_Initialize();

        // try to convert a 2 GB long string
        if ((str = mmap(NULL, (size_t)INT_MAX + 1, PROT_READ | PROT_WRITE,
            MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)) == (void *)-1)
                err(1, "malloc");
        memset(str, 'a', INT_MAX);
        str[INT_MAX] = '\0';
        Py_DecodeLocale(str, &wlen);

        PyMem_RawFree(program);
        return 0;
}

I doubt that this is really reachable with actual code. But at least it is a good showcase that actual arithmetic is left over in the if-check. Let's remove it and save us this possible headache.

PS: Not sure if this is the correct way to create python issues with GitHub now. Let me know if something's missing or wrong!