[Numpy-discussion] C-extensions to numpy - a little help is needed

[Mads Ipsen]

Hey,

I am trying to understand how to write numpy extensions, in order to be
able to do various things fast, that I need in my own apps. I have looked
at the examples from the Numeric package, which I managed to get working
with numpy, so I guess my questions holds for this forum.

There's one detail, that I don't understand in the examples provided with
the Numeric documentation, namely the use of

    Py_DECREF(...)

which is used in the first example but not in the last. Let me be more
specific:

In the example, showing how to implement a trace() function, you
access the data using

    array = (PyArrayObject *)
        PyArray_ContiguousFromObject(input, PyArray_DOUBLE, 2, 2);

do some summing to find the trace and then decrement and return the
result

    Py_DECREF(array);
    return PyFloat_FromDouble(sum);

Now if you look at the other example for the matrix-vector multiply,
the matrix and vector data are accessed using a similar approach

    PyObject *input1, *input2;
    PyArrayObject *matrix, *vector, *result;
    ...
    if (!PyArg_ParseTuple(args, "dOO", factor, &input1, &input2))
        return NULL;
    matrix = (PyArrayObject *)
        PyArray_ContiguousFromObject(input1, PyArray_DOUBLE, 2, 2);
    ...
    vector = (PyArrayObject *)
        PyArray_ContiguousFromObject(input2, PyArray_DOUBLE, 1, 1);
    ...
    result = (PyArrayObject *)
        PyArray_FromDims(1, dimensions, PyArray_DOUBLE);

Then calculate the DGEMV storing the result in result->data and return
with

    return PyArray_Return(result);

Finally here's my question:
----------------------------
Since the two pointers *matrix and *vector are created just like
*array in the trace() example, how come they are not decremented with
two calls to

    Py_DECREF(matrix);
    Py_DECREF(vector);

Any help is appreciated.

// Mads