Question

This should be a trivial question, but didn’t figure out as a beginner.

I have the following Python C-extenstion using numpy arrays:

#include <cmath>
#include <Python.h>
#include <iostream>
#include "numpy/arrayobject.h"

using namespace std;
PyObject *func_calc(PyObject *self, PyObject *args)
{
    PyObject * PO_clmn;
    PyArrayObject * py_clmn;
    if (!PyArg_ParseTuple(args, "O", &PO_clmn))
        return NULL;
    py_clmn = (PyArrayObject*)PyArray_ContiguousFromObject(PO_clmn,PyArray_DOUBLE,1,1);
    double *clmn = (double*)(py_clmn->data);

    int i;
    int N = py_clmn->dimensions[0];
    int flag_threadholds[N_threadholds];
    for (i=0; i<N; i++)
    {
        clmn[i]=1;
    }
    return Py_None;
}

static PyMethodDef exampleMethods[] = 
{
    { "calc", func_calc, METH_VARARGS },
    { NULL, NULL }
} ;

PyMODINIT_FUNC initcalc()
{
    import_array();
    Py_InitModule("calc", exampleMethods);
}

After compiling it as a shared library, I found the following call failed to modify the element clmn array to “1”:

import numpy
from calc import calc
clmn=numpy.zeros(10)
calc(clmn)
print clmn #[0,0...

Thanks in advance!

Answer 1

Depending on the data you pass in, the call to PyArray_ContiguousFromObject may return the original object, or it may return a copy of the object. If it returns a copy, then your code is modifying that copy, not the original object.