How to transfer and parse snap graph from python to c++ - c++

Question

Stanford SNAP is a well-known package for graph mining, and has both Python implementation and C++ implementation.
I have some code in python to do graph mining using SNAP. I also have a C++ function process the snap graph. Now I need to write a wrapper so that this C++ function can be invoked from Python.
The problem is that I don't know how to parse/dereference the snap graph object from Python to C++.
The python code looks like: (More explanations come after the code examples)
import my_module;
import snap;
G = snap.GenRndGnm(snap.PUNGraph, 100, 1000);
print(type(G));
A = my_module.CppFunction(G); # customized function
print(A);
The CPP wrapper my_module_in_cpp.cpp looks like:
// module name: my_module, defined in the setup file
// function to be called from python: CppFunction
#include <Python.h>
//#include "my_file.h" // can be ignored in this minimal working example.
#include "Snap.h"
#include <iostream>
static PyObject *My_moduleError;
// module_name_function, THE CORE FUNCTION OF THIS QUESTION
static PyObject *my_module_CppFunction(PyObject *self, PyObject *args) {
PUNGraph G_py;
int parseOK = PyArg_ParseTuple(args, "O", &G_py);
if (!parseOK) return NULL;
std::cout << "N: " << G_py->GetNodes() << ", E: " << G_py->GetEdges() << std::endl;
fflush(stdout);
if ((G_py->GetNodes()!=100)||(G_py->GetEdges()!=1000)) {
PyErr_SetString(My_moduleError, "Graph reference incorrect.");
return NULL;
}
PyObject *PList = PyList_New(0);
PyList_Append(PList,Py_BuildValue("i",G_py->GetNodes()));
PyList_Append(PList,Py_BuildValue("i",G_py->GetEdges()));
return PList;
}
// To register the core function to python
static PyMethodDef CppFunctionMethod[] = {
{"CppFunction", my_module_CppFunction, METH_VARARGS, "To call CppFunction in C++"},
{NULL,NULL,0,NULL}
};
extern "C" PyMODINIT_FUNC initmy_module(void) {
PyObject *m = Py_InitModule("my_module",CppFunctionMethod);
if (m==NULL) return;
My_moduleError = PyErr_NewException("my_module.error", NULL, NULL);
Py_INCREF(My_moduleError);
PyModule_AddObject(m, "error", My_moduleError);
}
I'm using Ubuntu, python-2.7. In case someone may want to re-produce the problem, the setup.py file is also provided.
from distutils.core import setup, Extension
module1 = Extension('my_module',\
include_dirs = ['/usr/include/python2.7/','/users/<my_local>/Snap-3.0/','/users/<my_local>/Snap-3.0/snap-core','/users/<my_local>/Snap-3.0/glib-core'],
library_dirs = ['/users/<my_local>/Snap-3.0/snap-core/'],
extra_objects = ['/users/<my_local>/Snap-3.0/snap-core/Snap.o'],
extra_compile_args=['-fopenmp','-std=c++11'],
extra_link_args=['-lgomp'],
sources = ['my_module_in_cpp.cpp'])
setup (name = 'NoPackageName', version = '0.1',\
description = 'No description.', ext_modules = [module1])
Every time I run the python code above, the error message "Graph reference incorrect." is displayed.
Apparently G_py->GetNodes() and G_py->GetEdges() cause the problem. This must result from G_py not pointing to the right address/in right format. I tried using TUNGraph in the cpp code as well, it still does not point to the correct address. Is there any way that the pointer in C++ can point to the correct address of the original C++ object?
Although in general it is hard to dereference a PythonObject from C++, but in this case I think it is doable since Snap-Python is also implemented in C++. We just need to unwrap its python wrapper. And the snap authors also provided the SWIG files.
Of course we can write the graph file in the disk, and read from that, but this will result in I/O and incur extra time consumption. And the snap-user-group does not have as much user traffic as stackoverflow.
BTW, there are networkx and stanford-nlp tags, but no stanford-snap or similar tag referring to that tool. Can someone create such a tag?