On 03/18/2012 04:41 AM, Guannan Ren wrote:
>> + if (totalbool == Py_False) {
> Per other code in libvirt-override.c, you can't compare totalbool (type
> PyObject) with Py_False, at least not on all compilers. You need
> something like this instead:
>
> /* Hack - Python's definition of Py_True breaks strict
> * aliasing rules, so can't directly compare
> */
> if (PyBool_Check(value)) {
> PyObject *hacktrue = PyBool_FromLong(1);
> temp->value.b = hacktrue == value ? 1 : 0;
> Py_DECREF(hacktrue);
Yes, it did report warning in compiling as follows due to the case
from PyIntObject* to PyObject*
warning :dereferencing type-punned pointer might break
strict-aliasing rules [-Wstrict-aliasing]
And that would trip up a -Werror compilation, so I'm glad to see you
changed it in v3.
GCC command line to reproduce the error:
gcc -Wstrict-aliasing=1 -O2 cpythonexample.c
Actually PyObject_IsTrue() is a more light-weight approach to do
the checking instead of
creating a intermediate PyObject * for the compare.
Is PyObject_IsTrue() available in the version of python present on RHEL
5? If so, I'd be in favor of a followup cleanup patch that removes all
our hacks in favor of the python glue code that does the same thing.
And even if not, we should write a decent wrapper in our own
typewrappers.c, so that the rest of our code doesn't have to look so
ugly with so much copy-and-paste.
--
Eric Blake eblake(a)redhat.com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org