Thanks a lot for your guidance, knowing that there was no standard way really helped me. What I ended up doing was incorporating PyObjects ( which map directly to Python Lists on being returned for C functions through ctypes. You're also able to make PyObjects of PyObjects, so it was a simple matter of making a two dimensional PyObject that held both time and fluxes.

Nicholas Earl
Graduate Research Assistant
San Diego State University

On Sun, Sep 29, 2013 at 11:30 PM, Thor Andreas Tangen <> wrote:
Thats a challenge then, that the size is unknown. I dont have too much experience here, but you could allocate the memory persistently in you C-library using malloc, and in the destructor of the PhotoInfo class in Python, make sure to call some memory free function in you c-library to make sure the memory is freed correctly. You can use vectors to populate your array and then allocat using malloc since you now know the size and copy over the data from the vector to the new memory.

Someone else might have a more elegant solution to this, as I said I have not done this type of operation before.

Thor Andreas

On Mon, Sep 30, 2013 at 8:21 AM, Nicholas Earl <> wrote:
I see. So -- sorry to drag you through this -- if I create an instance of PhotoInfo, which is a subclass of ctypes.STRUCTURE, and then pass that in as a parameter to the C function, I could then have the C function assign the array object to the PhotoInfo variables? But the array objects can' t be pointers, right? The problem is, I don't know how big the array objects will be, so I've been using vectors and then retrieving their array pointers.

Thanks for your help,

Nicholas Earl
Graduate Research Assistant
San Diego State University

On Sun, Sep 29, 2013 at 11:17 PM, Thor Andreas Tangen <> wrote:
I would allocate the memory on the python side. Allocating persistent memory on the C-side makes it difficult to free, unless you do it specifically. If you allocate it on the python side the GC would take care of all memory management.

Thor Andreas

On Mon, Sep 30, 2013 at 8:10 AM, Nicholas Earl <> wrote:

Thanks for the response! Is there, then, a standard way of calling a C function, creating an object (struct) that can hold two arrays, and then return that data object to the python program?


Nicholas Earl
Graduate Research Assistant
San Diego State University

On Sun, Sep 29, 2013 at 10:11 PM, Thor Andreas Tangen <> wrote:

The vectors times and fluxes are destroyed when exiting the function start, the memory allocated and pointed to be times and fluxes are invalid upon exiting the function.

Thor Andreas

On Sun, Sep 29, 2013 at 12:53 AM, Nicholas Earl <> wrote:
I'm struggling with returning a struct value using ctypes. Mainly, it keeps segfaulting when I try and get the values from the returned object. I have my python script setup like this:

import ctypes
import numpy as np
from numpy.ctypeslib import ndpointer

class PhotoInfo(ctypes.Structure):
    _fields_ = [
        ('times', ctypes.POINTER(ctypes.c_double)),
        ('fluxes', ctypes.POINTER(ctypes.c_double)),

def main():
    lib = ctypes.cdll.LoadLibrary('')

    start = lib.start
    start.restype = ctypes.POINTER(PhotoInfo)
    start.argtypes = [

    N = 3
    mass = np.array([0.00020335520, 5.977884E-05, 9.320397E-08])

    phin = start(N, mass)

    print phin[0].times[0]

And in my c file:

#include <iostream>
#include <iomanip>
#include <fstream>
#include <sstream>
#include <string>
#include <vector>
#include <math.h>
#include <stdlib.h>
#include <stdio.h>

using namespace std;

typedef struct {
  double *times;
  double *fluxes;
} PhotoInfo;

extern "C" {
  PhotoInfo *start(int N, double *mass);

PhotoInfo *start(int N, double *mass) {
  PhotoInfo phin;
  vector<double> times;
  vector<double> fluxes;

// Here, I read in a file and fill times and fluxes with doubles using
// times.push_back(t) and fluxes.push_back(f)\

  phin.times = &times[0];
  phin.fluxes = &times[0];
  return &phin;

Any guidance on this would be greatly appreciated!

Nicholas Earl
Graduate Research Assistant
San Diego State University

October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
the latest Intel processors and coprocessors. See abstracts and register >
ctypes-users mailing list