Bue Krogh Vedel-Larsen wrote:
> But if I call
>
> SA_PyVector_Type.tp_new = PyType_GenericNew;
> PyType_Ready( &SA_PyVector_Type );
>
> then, when Py_Finalize is called, PyObject_IS_GC(op) in visit_decref() in
> gcmodule.c causes an access violation. If I don't call PyType_Ready, then
> the access violation doesn't occure, but then the type can't be used...
>
> So, the question is, does anyone have any idea about what could be
> causing this?
Most likely some code that you haven't shown. Here is the expansion
of PyObject_IS_GC(op)
#define PyObject_IS_GC(o) (PyType_IS_GC((o)->ob_type) && \
((o)->ob_type->tp_is_gc == NULL || (o)->ob_type->tp_is_gc(o)))
so it becomes
PyType_IS_GC(op->type) && (op->ob_type->tp_is_gc==NULL ||
op->ob_type->tp_is_gc(op))
Then, PyType_IS_GC(op->type) becomes
PyType_HasFeature((op->type), Py_TPFLAGS_HAVE_GC)
which in turn becomes
(op->tp_flags & Py_TPFLAGS_HAVE_GC) != 0
So typically, PyObject_IS_GC goes to the type of the object,
which should never crash, and then looks into the flags of
the type, which cannot crash - unless somebody messed with
ob_type of the object, and unless this isn't a Python
object in the first place.
You did not say what kind of object op was in the crash - this
is something you should investigate further. Does it point to
a Python object? If so, what is the type of the Python object?
Regards,
Martin
--
http://mail.python.org/mailman/listinfo/python-list
