I'm curious what the most flexible, most efficient, and most seamless method is for getting C++ and Python to talk to each other.
The contenders seem to be Pybind11, Boost.Python, and neither (simply writing functions and wrappers as below).
using namespace boost::algorithm;
static PyObject* strtest(PyObject* self, PyObject* args)
{
std::string s = "Boost C++ Libraries";
to_upper(s);
PyObject * python_val = Py_BuildValue("s", s.c_str());
return python_val;
}
PyMODINIT_FUNC initmath_demo(void)
{
static PyMethodDef methods[] = {
"Test boost libraries" },
{ NULL, NULL, 0, NULL }
};
PyObject *m = Py_InitModule("math_demo", methods);
}
I would recommend PyBind11. I am using it for similar use-case where Python modules calls C++ to perform operations which are costlier and performance extensive. Boost Python is a richer library with size cost where as PyBind11 is header only and it supports STL which makes life easier to pass on basic data structure without writing any code! If you can wrap your calls to basic C function and primitive data types then Cpython would be the best!
We are using pybind11 in-house for something similar to what you describe (C++ module being accessed from Python). I am unwilling to pull in all the requirements of Boost, and pybind11 takes care of much of the "glue" normally necessary between C++ and Python. Things like converting exceptions are not readily thought of in many cases, but are handled nicely by pybind11.
In addition we write wrappers around our C++ internal classes, so that pybind11 can be pybind11 and our internal core engine can be C++. Translations are typically not that hard.
Personal experience with pybind11 is that "easy things are easy, hard things are still hard". Anything that is covered in the sample tutorial is quickly second nature, but things that aren't obvious usually take some digging to find a good way to do.
Still, all in all, very recommended. I would use it again for new projects in a heartbeat.
