Cross-references between dynamically loaded modules under AIX

[Jakob Schiotz]

  I am tuning a C program into a python module.  The program contains
some very large arrays of doubles, that I would like to become NumPy
arrays in Python.  I wrote a small function to convert a C pointer to
a NumPy array (I am using SWIG), the program is appended below.  It
works fine on an alpha processor running OSF1 V4.0, but it coredumps
under AIX version 4.1.5.  The coredump occurs when PyArray_FromDims is
called.  It looks like the dynamical loader under AIX cannot resolve
symbols in one module that refers to another module.  It is probably
related to the primitive way dynamic loading is working under AIX.
The modules are created by the ld_so_aix, which is using a text file
with the symbols defined in the python executable (created when the
python executable was linked).  I assume that the linker cannot link
the references to symbols in the NumPy modules without similar
information.

  Does anyone have a solution for this problem?  Repeating the relevant
code from the NumPy source would not be very nice, as it is quite a
bit that would have to be repeated.


Best regards,

Jakob Schiotz


Here is the interface file for SWIG.  Vector is typedef'ed to
double[3] in main.h.
------------------------------------------------------------
/* Emacs: Use -*- C -*- mode for this stuff */
%module fast

%title "FAST - a molecular dynamics module",keep

%{
#include "main.h"
#include "arrayobject.h"
%}

%include fastcommon.i

/* Initialize FAST, reading input files */
CmFile *initfast(char *potfilename, char *infilename, int wantclass,
		 int dynamics, int frame, int hmode);


%init %{
    setdefaults();
    fprintf(stderr, "FAST: %s %s\n", __DATE__, __TIME__);
%}

/* Access functions for global arrays of Vector */
%{
PyObject *my_Vectors_As_NumPy(PyObject *self, PyObject *args)
    {
	char *vectorp;
	double *cdata;
	PyArrayObject *numpy;
	int dims[2];

	/* Read the argument, it should be a string encoding a pointer */
	if (!PyArg_ParseTuple(args, "s", &vectorp))
	    return NULL;

	/* Decode the pointer */
	if (SWIG_GetPtr(vectorp, (void **) &cdata, "_Vector_p"))
	{
	    PyErr_SetString(PyExc_TypeError, "Not a pointer to Vector");
	    return NULL;
	}

	/* Create an empty numpy array of dimension nAtoms * 3 */
	dims[0] = nAtoms;
	dims[1] = 3;
	numpy = (PyArrayObject *) PyArray_FromDims(2, dims, PyArray_DOUBLE);

	/* Copy the data to the numpy array */
	memcpy(numpy->data, cdata, dims[0]*dims[1]*sizeof(double));

	return (PyObject*) numpy;
    }
%}

%native(Vectors_As_NumPy) extern PyObject *my_Vectors_As_NumPy(PyObject *self,
PyObject *args);

--
Jakob Schiotz, CAMP and Department of Physics, Tech. Univ. of Denmark,
DK-2800 Lyngby, Denmark.  http://www.fysik.dtu.dk/~schiotz/
This email address is used for newsgroups and mailing lists
(spam protection).  Official email: schiotz @ fysik . dtu . dk

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own