untidy blogsimon steele writes about stuff…

So far, instead of exposing functions to Python, we have (the reverse I guess) embedded the Python interpreter in our process using Boost::Python. So, running Python from within our process is all good and well but we want the Python code to interact with our C++ code as well.

We’ll add to the previous example:

const char* greet()
{
	return "Hello World from C++";
}

BOOST_PYTHON_MODULE(embed_test)
{
	def("greet", greet);
}

Here we have defined a python module embed_test which we will now expose to our embedded Python interpreter. The module has one function, “greet” in the Python module calls the C++ greet function.

Now we add to the main function:

if(PyImport_AppendInittab("embed_test", initembed_test) == -1)
		throw std::runtime_error("Failed to add test module to builtins");

The initembed_test reference is the function name generated by the BOOST_PYTHON_MODULE macro. Once we have made Python aware of this built-in module, we can modify the python code we run in the example to print out the result of the greet function:

handle<> ignored((PyRun_String(
	"import embed_test\n"
	"print embed_test.greet()"

	, Py_file_input
	, main_namespace.ptr()
	, main_namespace.ptr())
));

The techniques used here are all based around the basic functionality available in the Python C API. However, the Boost::Python library does make it easier to expose your own code to the Python interpreter. Next we’ll look at exposing classes and then move on to the use of SWIG.

I just created a “Hello World” application in Boost::Python. I’m comparing the Boost and SWIG options for embedding Python and making PN code available to it so this is the first step for the Boost implementation.

Here’s the code:

#include <boost /python.hpp>

int main(int argc, char* argv[])
{
  Py_Initialize();

  object main_module((
    handle<>(borrowed(PyImport_AddModule("__main__")))
  ));

  object main_namespace = main_module.attr("__dict__");

  // Run some python!

  handle<> ignored((PyRun_String(

    "print \"Hello World from Python!\""

    , Py_file_input
    , main_namespace.ptr()
    , main_namespace.ptr())
  ));

  Py_Finalize();

  return 0;
}

Not very much code at all to get going! boost_python.dll (required to use Boost::Python) comes in at 180kb. The initial test app (as coded above) in Release configuration takes 40kb.

The differences between this and the SWIG embedding version will be minimal, boost doesn’t hide much of the embedding code for you. The real differences will be in creating wrappers for functions and classes. With boost I have to write wrappers, SWIG will automatically generate them. The interesting bit will be the size of the code created.