[Numpy-discussion] Altering/initializing NumPy array in C

[Bart Baker]

Hello,

I'm having issues with performing operations on an array in C and
passing it back to Python. The array values seem to become unitialized
upon being passed back to Python. My first attempt involved initializing
the array in C as so:

double a_fin[max_mth];

where max_mth is an int. I fill in the values of a_fin and then, before
returning back to Python, I create a NumPy array in C and fill it in
using a pointer to the a_fin array:

npy_intp a_dims[2] = {max_mth, 1};
a_fin_array = (PyArrayObject *) PyArray_SimpleNewFromData(2, a_dims,
                                                          NPY_DOUBLE,
                                                          a_fin);
I update the flags as so:

PyArray_UpdateFlags(a_fin_array, NPY_OWNDATA);

and return using:

PyObject *Result = Py_BuildValue("OO", a_fin_array, b_fin_array);
Py_DECREF(a_fin_array);

(there is another array, b_bin_array that I create in this way and it
suffers from the same issues).

Immediately upon returning to Python, all of a_fin_array appears
unitilized. This only happens in certain situations and sometime only
part of the arrary will be unitilized. I check the values of a_dim and
a_fin_array in C using gdb and they appear as expected, but are
over-written with unitialized values upon returning to Python.

I've tried initializing in Python and then passing the NumPy array in
instead of initializing in C, but the effects of the calculations in
C are still not kept.

My feeling is that, with the Numpy C-API, this should be a simple
process, but I'm having a lot of trouble with it.

Any help would be much appreciated. I didn't want to give too much
information in the post, but please let me know what other information
would be useful.

-Bart