Hi,

the numpy.i interface provides us with most of the items I need for a project, however, I have a case whereby I build an array in C++ and need to pass it to numpy. The size of the array is unknown at the beginning, so I cannot use the numpy.i typedefs for ARGOUT_ARRAY1 (as these require you to supply the dimension of the array to be returned).

Below is my code, which runs into a segmentation fault upon calling test.func(). Gdb backtrace is provided as well.

======= test.cxx =========
#include "test.h"
#include <iostream>
#include <Python.h>
#include <numpy/npy_common.h>
#include <numpy/arrayobject.h>

using namespace std;
void func(double * points, int nbr_points, PyObject **py_A)
{
    npy_intp *dims = new npy_intp[1];
    dims[0] = 3; //makes the length along the first (and only, in this example) dimension equal to 3
    double * data = new double[3];
   
    data[0] = points[0];
    data[1] = (double) nbr_points;
    data[2] = 3.3;

    *py_A = PyArray_SimpleNew(1,dims,NPY_DOUBLE);
    double * t = (double*)PyArray_DATA((PyArrayObject*) (*py_A));
    t = data;
    delete[] dims;
    //delete[] data;*/
}

void func2(double * points, int nbr_points)
{
    cout << points[0] << " " << nbr_points << endl;
}


============ test.h ===========
#include <Python.h>
//#include <numpy/npy_common.h>
//#include <numpy/arrayobject.h>
 
 
void func(double * points, int nbr_points, PyObject **py_A);
void func2(double * points, int nbr_points);



========== test.i ===========
%module test
 
%{
#define SWIG_FILE_WITH_INIT
#include "test.h"
%}
 
%include "numpy.i"
 
 
%init %{
    import_array();
%}
 
%include "typemaps.i"
 
// make use of the typemap provided by numpi.i
%apply (double* IN_ARRAY1, int DIM1 ) {(double * points, int nbr_points)}
 
// this makes use of typemaps.i, as I see no way to do the above using numpy.i
%apply PyObject **OutValue {PyObject **py_A}
 
 
%include "test.h"
 
============== typemaps.i ==============
// Code originates from M. Fiers
/*
 * This tells SWIG to treat an double * argument with name 'OutValue' as
 * an output value.  We'll append the value to the current result which
 * is guaranteed to be a List object by SWIG.
 */
 
%typemap(argout) PyObject **OutValue {
    PyObject *o, *o2, *o3;
   
    o = *$1;
   
    if ((!$result) || ($result == Py_None)) {
        $result = o;
    } else {
        if (!PyTuple_Check($result)) {
            PyObject *o2 = $result;
            $result = PyTuple_New(1);
            PyTuple_SetItem($result,0,o2);
        }
        o3 = PyTuple_New(1);
        PyTuple_SetItem(o3,0,o);
        o2 = $result;
        $result = PySequence_Concat(o2,o3);
        Py_DECREF(o2);
        Py_DECREF(o3);
    }
}

// This is used because the arguments are not used as input arguments, but as output arguments.
%typemap(in,numinputs=0) PyObject **OutValue(PyObject *temp) {
    $1 = &temp;
}



==========compilation sequence ========
#!/bin/bash

if [ $# -ne 1 ]; then
    echo "please indicate the filename (without extensions) as the first CL option"
    exit
fi

file=$(basename "$1" )
echo "working on '$file'"
echo "swigging the main interface file..."
swig -c++ -python $file.i

echo "compiling the C/C++ file with python2.7 included..."
g++ -O0 -g -fPIC -I/usr/include/python2.7 -c ${file}.cxx ${file}_wrap.cxx

echo "linking object files..."
g++ -O0 -g -shared $file.o ${file}_wrap.o -o _$file.so

echo "done"

=========testrun.py=========
import numpy as np
import test

a=np.zeros(3)
b=np.array([1., 2. , 3.])

c=test.func2(b)
d=test.func(a)


================ gdb backtrace =========
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Python 2.7.3 (default, Apr 20 2012, 22:39:59)
[GCC 4.6.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import testrun
1 3

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff28d89d6 in func (points=0xecabb0, nbr_points=3, py_A=0x7fffffffa6e8) at test.cxx:18
18        *py_A = PyArray_SimpleNew(1,dims,NPY_DOUBLE);
(gdb) bt
#0  0x00007ffff28d89d6 in func (points=0xecabb0, nbr_points=3, py_A=0x7fffffffa6e8) at test.cxx:18
#1  0x00007ffff28dd83a in _wrap_func (args=0x7ffff7f55ed0) at test_wrap.cxx:3454
#2  0x000000000042a485 in PyEval_EvalFrameEx ()
#3  0x00000000004317f2 in PyEval_EvalCodeEx ()
#4  0x000000000054a078 in PyImport_ExecCodeModuleEx ()
#5  0x000000000050d091 in ?? ()
#6  0x000000000050da8b in ?? ()
#7  0x0000000000488793 in ?? ()
#8  0x000000000050e2e4 in ?? ()
#9  0x0000000000432f0b in ?? ()
#10 0x00000000004c7c76 in PyObject_Call ()
#11 0x00000000004c7d36 in PyEval_CallObjectWithKeywords ()
#12 0x000000000042c8a5 in PyEval_EvalFrameEx ()
#13 0x00000000004317f2 in PyEval_EvalCodeEx ()
#14 0x000000000054bd50 in PyRun_InteractiveOneFlags ()
#15 0x000000000054c045 in PyRun_InteractiveLoopFlags ()
#16 0x000000000054ce9f in Py_Main ()
#17 0x00007ffff68e576d in __libc_start_main () from /lib/x86_64-linux-gnu/libc.so.6
#18 0x000000000041b931 in _start ()
(gdb)


Does anyone have an idea why this is not working or perhaps a suggestion for an alternative? I would be grateful for either

- Oliver