[Pyrex] Bad code generation in Pyrex 0.9.4.1

[Konrad Hinsen]

I think I found a bug in Pyrex; it generates bad C code for one of my  
modules. Reduced to the bare minimum, my Pyrex module becomes

	cdef class vector:
	    def __div__(vector self, double factor):
         	result = vector()
         	return result

When __div__ is called with an argument that is not a float, it  
should raise an exception, which is what it does on my Mac. However,  
under Linux it causes a segmentation fault. A look at the generated  
code reveals the reason:

static PyObject *__pyx_f_5debug_6vector___div__(PyObject  
*__pyx_v_self, PyObject *__pyx_arg_factor); /*proto*/
static PyObject *__pyx_f_5debug_6vector___div__(PyObject  
*__pyx_v_self, PyObject *__pyx_arg_factor) {
   double __pyx_v_factor;
   PyObject *__pyx_v_result;
   PyObject *__pyx_r;
   PyObject *__pyx_1 = 0;
   PyObject *__pyx_2 = 0;
   Py_INCREF(__pyx_v_self);
   __pyx_v_factor = PyFloat_AsDouble(__pyx_arg_factor); if  
(PyErr_Occurred()) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 3;  
goto __pyx_L1;}

/* some lines ommitted */

   __pyx_L1:;
   Py_XDECREF(__pyx_1);
   Py_XDECREF(__pyx_2);
   __Pyx_AddTraceback("debug.vector.__div__");
   __pyx_r = 0;
   __pyx_L0:;
   Py_DECREF(__pyx_v_result);
   Py_DECREF(__pyx_v_self);
   return __pyx_r;
}


If the conversion of "factor" to double fails, the variable  
__pyx_v_result remains unitialized but is passed to Py_DECREF. It  
should instead be treated just like the temporaries __pyx_1 and  
__pyx_2: it should be initialized to null and passed to Py_XDECREF.

Konrad.
--
---------------------------------------------------------------------
Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: hinsen at cnrs-orleans.fr
---------------------------------------------------------------------