Embedded Python/NumPy and reference counts

[Nigel Wade]

Hi,

I'm in the process of embedding Python in an application of mine, 
and want to be able to transfer Numerical Python arrays from the
application to the interpreter.

In essence what I need to do is initialize the interpreter, 
create a local dictionary, then loop repeatedly passing an array from 
C to the interpreter in the same item in the local dictionary.

It looks something like this:

int dims[1];
PyObject *locals, *numpy_array;
float *ptr;

Py_Initialize();
PyImport_ImportModule("Numeric");
locals = Py_DictNew()


loop {
   ptr = pointer to a C data array

   dims[0] = length of C data array;   
   numpy_array = 
      PyArray_FromDimsAndData(1, dims, PyArray_CFLOAT, (char *)ptr)

   PyDict_SetItemString(locals, "d_data", numpy_array)

   PyRun_String("print d_data", Py_single_input, NULL, locals)
}

Now, what I am not sure about is what to do about numpy_array and
the d_data item in the locals dictionary each time around the loop.

After I have called PyDict_SetItemString has the reference count on 
the object numpy_array been incremented, so I can call 
Py_DECREF(numpy_array) without losing d_data in the locals dictionary?

After I have called PyRun_String if I call 

PyDict_DelItemString(locals,"d_data")
 
will this remove all remaining references to d_data? Also, what 
happens to the data to which the numpy_array points? Is it
free'd, or is it up to me to free it?

For my purposes it would be much more efficient if I created 
numpy_array and d_data once and then manipulated the internal
values of dimensions and the data pointer for numpy_array.
But I don't really like doing this because messing around in the 
internal structure seems like a good way to cause future maintenance
headaches. Is there any defined interface for doing something
like this? I can't find anything in the Numerical Python manual.


-- 
-----------------------------------------------------------
Nigel Wade, System Administrator, Space Plasma Physics Group,
            University of Leicester, Leicester, LE1 7RH, UK 
E-mail :    nmw at ion.le.ac.uk 
Phone :     +44 (0)116 2523568, Fax : +44 (0)116 2523555