Confusion on C++ Python extensions. Things like getting C++ values for python values.

Question

I'm wanted to convert some of my python code to C++ for speed but it's not as easy as simply making a C++ function and making a few function calls. I have no idea how to get a C++ integer from a python integer object. I have an integer which is an attribute of an object that I want to use. I also have integers which are inside a list in the object which I need to use.

I wanted to test making a C++ extension with this function:

def setup_framebuffer(surface,flip=False):
   #Create texture if not done already
   if surface.texture is None:
      create_texture(surface)
   #Render child to parent
   if surface.frame_buffer is None:
      surface.frame_buffer = glGenFramebuffersEXT(1)
   glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, c_uint(int(surface.frame_buffer)))
   glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT, GL_TEXTURE_2D, surface.texture, 0)
   glPushAttrib(GL_VIEWPORT_BIT)
   glViewport(0,0,surface._scale[0],surface._scale[1])
   glMatrixMode(GL_PROJECTION)
   glLoadIdentity() #Load the projection matrix
   if flip:
      gluOrtho2D(0,surface._scale[0],surface._scale[1],0)
   else:
      gluOrtho2D(0,surface._scale[0],0,surface._scale[1])

That function calls create_texture, so I will have to pass that function to the C++ function which I will do with the third argument. This is what I have so far, while trying to follow information on the python documentation:

#include <Python.h>
#include <GL/gl.h>
static PyMethodDef SpamMethods[] = {
   ...
   {"setup_framebuffer",  setup_framebuffer, METH_VARARGS,"Loads a texture from a Surface object to the OpenGL framebuffer."},
   ...
   {NULL, NULL, 0, NULL}        /* Sentinel */
};
static PyObject * setup_framebuffer(PyObject *self, PyObject *args){
   bool flip;
   PyObject *create_texture, *arg_list,*pyflip,*frame_buffer_id;
   if (!PyArg_ParseTuple(args, "OOO", &surface,&pyflip,&create_texture)){
      return NULL;
   }
   if (PyObject_IsTrue(pyflip) == 1){
      flip = true;
   }else{
      flip = false;
   }
   Py_XINCREF(create_texture);
   //Create texture if not done already
   if(texture == NULL){
      arglist = Py_BuildValue("(O)", surface)
      result = PyEval_CallObject(create_texture, arglist);
      Py_DECREF(arglist);
      if (result == NULL){
         return NULL;
      }
      Py_DECREF(result);
   }
   Py_XDECREF(create_texture);
   //Render child to parent
   frame_buffer_id = PyObject_GetAttr(surface, Py_BuildValue("s","frame_buffer"))
   if(surface.frame_buffer == NULL){
      glGenFramebuffersEXT(1,frame_buffer_id);
   }
   glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, surface.frame_buffer));
   glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT, GL_TEXTURE_2D, surface.texture, 0);
   glPushAttrib(GL_VIEWPORT_BIT);
   glViewport(0,0,surface._scale[0],surface._scale[1]);
   glMatrixMode(GL_PROJECTION);
   glLoadIdentity(); //Load the projection matrix
   if (flip){
      gluOrtho2D(0,surface._scale[0],surface._scale[1],0);
   }else{
      gluOrtho2D(0,surface._scale[0],0,surface._scale[1]);
   }
   Py_INCREF(Py_None);
   return Py_None;
}
PyMODINIT_FUNC initcscalelib(void){
   PyObject *module;
   module = Py_InitModule("cscalelib", Methods);
   if (m == NULL){
      return;
   }
}
int main(int argc, char *argv[]){
   /* Pass argv[0] to the Python interpreter */
   Py_SetProgramName(argv[0]);
   /* Initialize the Python interpreter.  Required. */
   Py_Initialize();
   /* Add a static module */
   initscalelib();
}

Answer 1

The way to get a C int from a Python integer is with PyInt_AsLong() and a downcast (although you may want to use a C long instead.) To go the other way, you call PyInt_FromLong(). It's not obvious to me where in the code you want to do this, although I do have a couple of other comments about your code:

• Conceptually, PyObject_IsTrue() returns a boolean. Don't compare it to 1, just use it as a boolean value. However, you should check if it returned an error, which would be -1. (The usual check is < 0.) If you don't check for error returns you end up swallowing the exception, but leaving the exception object hanging around. This is bad.

• The Py_XINCREF() of create_texture is not necessary. You have a borrowed reference to the args tuple, which in turn owns a reference to the create_texture object. There is no way for create_texture to go away before your function returns. You only need to incref it if you are going to keep it around longer than this functioncall. If you did have to incref it, you wouldn't need to use Py_XINCREF() because it won't ever be NULL. And if you did have to incref it, you would need to remember to decref it in your error return case, as well.

• Instead of creating an argument tuple just to call PyEval_CallObject(), just call PyObject_CallFunctionObjectArgs(create_texture, arglist, NULL) or PyObject_CallFunction(create_texture, "O", arglist).

• Py_BuildValue("s", "frame_buffer") is not really the right way to get a Python string for "frame_buffer". A better way is PyString_FromString(). However, both of those return new references, and PyObject_GetAttr() doesn't eat the reference to the attribute name, so you end up leaking that reference. You should use none of these, and instead use PyObject_GetAttrString(), which takes the attribute name as a const char*.

• Remember to check the return value of all functions that can return an error value (which is almost all Python API functions.) Besides PyTrue_IsTrue() you're also forgetting this for Py_BuildValue() and PyObject_GetAttr().

Answer 2

May I make a sincere suggestion here? Instead of throwing a hundred lines of code at yourself when trying something new like this, first try the simplest test case possible. Try getting a C++ extension to work that does nothing but pass back the value 123. (Picking 0 or 1 might accidentally work, so I chose an easy to recognize value that is not likely to be an accident.) Then get the extension to double a number you pass in, then add two numbers, etc.

Sneak up on the problem! :-)