Py_initialize / Py_Finalize not working twice with numpy

Question

On the second call of the following code, my app segfault, so I guess I am missing something :

Py_Initialize();
pName = PyString_FromString("comp_macbeth");
pModule = PyImport_Import(pName);
Py_DECREF(pName);

if(pModule == NULL) {
    PyErr_Print();
    Py_Finalize();
    return;
}

pFunc = PyObject_GetAttrString(pModule, "compute");
/* pFunc is a new reference */

if (!pFunc || !PyCallable_Check(pFunc) ) {
    PyErr_Print();
    Py_Finalize();
    return;
}

Py_Finalize();

The comp_macbeth.py is importing numpy. If I remove the numpy import, everything is fine. Is it a numpy bug, or am I missing something about imports ?

Answer 1

From the Py_Finalize docs:

Some extensions may not work properly if their initialization routine is called more than once; this can happen if an application calls Py_Initialize() and Py_Finalize() more than once.

Apparently Numpy is one of those. See also this message from Numpy-discussion.

Calling Py_Initialize() only once, and cleaning up at exit, is the way to go. (And it's should be faster, too!)

Answer 2

I have this in my module initialization part, but the URL does not exist anymore. In case it helps:

// http://numpy.scipy.org/numpydoc/numpy-13.html mentions this must be done in module init, otherwise we will crash
import_array();