Move to PyBind11

[henryiii]

You could simplify the code significantly if you moved to using pybind11. This would remove the need to import the numpy headers in setup.py completely. The binding code would be:

At the top of the file

#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>

namespace py = pybind11;
using namespace pybind11::literals;

And, after the C++ code is imported, or at the bottom of the file, in a file named landau.cpp:

PYBIND11_MODULE(landau, m) {
    m.def("landauPDF", py::vectorize(landauPDF),
            "x"_a, "x0"_a=0., "xi"_a=1.);

    m.def("gaussPDF", py::vectorize(gaussPDF),
            "x"_a, "mu"_a=0., "sigma"_a=1.);

    m.def("landauGaussPDF", py::vectorize(landauGaussPDF),
            "x"_a, "mu"_a=0., "eta"_a=1., "sigma"_a=1.);
}

This does automatic vectorization, as well, so x can be a vector or a single value. Here's a file example: https://gist.github.com/henryiii/9419a0b803664d8ff2a159c1fcfd644e

Compile by hand if you want:

c++ -O3 -Wall -shared -std=c++11 -fPIC `python -m pybind11 --includes` landau.cpp -o landau`python-config --extension-suffix`

To update pylandau, you'd need a little wrapper to add your new functions in python (probably, c++ is easy too as you can see), and a setup.py.

PS: The other option would be to move to using Cython's buffer interface, but I think the above method is simpler.

[DavidLP]

Thank you very much for your suggestion and nice write up. I was not aware of pybind.