On Wednesday 29 March 2017, Andi Vajda wrote:
> On Wed, 29 Mar 2017, Petrus Hyvönen wrote:
> > Hi,
> >
> > With the /DLL, sprintf(buffer, "%0*%jx", (int) hexdig, hash); and
> > Py_SIZE it compiles under windows (Windows 7, 64 bit)
> >
> > I haven't set up for building pylucene but has another library that
> > I build.
> >
> > For that I get a udf-8 error on:
> >
> > File
> > "C:\Users\phy\AppData\Local\Continuum\Anaconda3-430\conda-bld\oreki
> >t_1490824040916\_b_env\lib\site-packages\jcc\c pp.py", line 898, in
> > header
> > env.strhash(signature(constructor)))
> > UnicodeDecodeError: 'utf-8' codec can't decode byte 0xac in
> > position 9: invalid start byte
>
> That means that the "%0*%jx" change for _MSC_VER also needs a change
> for the hexdig sizeof computation. There is a mismatch and garbage is
> left in the buffer array.
Hm, actually "%jx" and PRIxMAX should be equivalent modifiers to print
uintmax_t (c99). According to the MSC documentation I thought that the
macro is even more safe to use.
The problem is that we do C++ here which does not need to support c99 at
all. Though MSC might be the only C++ compiler which ignores c99...
To fix it I would suggest to use non-c99 types here. Lets use "unsigned
long long" and "%llx". This may look a bit less rock-solid but good
enough for any existing systems:
static PyObject *t_jccenv_strhash(PyObject *self, PyObject *arg)
{
unsigned long long hash = (unsigned long long) PyObject_Hash(arg);
static const size_t hexdig = sizeof(hash) * 2;
char buffer[hexdig + 1];
sprintf(buffer, "%0*llx", (int) hexdig, hash);
return PyUnicode_FromStringAndSize(buffer, hexdig);
}
BTW this function should be also copied to the py2 directory where we
still use int allthough PyObject_Hash returns already long on python
>2.x.
cu,
Rudi
