This is a minimalist working example of building a python module to wrap a custom C++ library which uses the STL (vector) and template classes/functions using PyBind11.

Feel free to fork this project to start a new one or add any of the files to your project.

/
|- pybind/                      # The Pybind library (as git submodule)
|- src/
    |- mylib.*                  # The custom library
    |- pybind_def.cpp           # Pybind glue code
    |- test.cpp                 # simple C++ test of the library
|- CMakeLists.txt               # CMake configuration file
|- setup.py                     # Python install instructions
|- README.md                    # This file

In order to build your python module you need the pybind11 folder (from the PyBind github directly). Pybind is included here as a git submodule, therefore you have 2 options for cloning this repo (and get pybind11):

1. use directly: git clone --recursive https://github.com/mpascucci/pybind_stl_vector_example.git the first time you clone this repo.

2. clone normally and then run git submodule init and git submodule update to get pybind11.

Next, you have to do is edit the following files for your project's needs.

• CMakeLists.txt

• setup.py

• src/pybind_def.cpp

If you want to compile the python module for a particular python interpter installed on your system, you can specify it by editing this line in the CmakeLists.txt file: set(PYTHON_EXECUTABLE "~/miniconda3/bin/python3")

This project is based on this Pybind-Cmake example.

Go to the source folder and run the command

$ pip install -e .

This will compile and generate a .so importable python module in the project folder.

If you want to install the module in your system instead, run:

$ pip install .

Run the test.py python script (using the same interpreter if you specified one).

Alternatively tun a python session in the project root:

$ python

>>> import lib_example
>>> import numpy as np
>>> v = np.arange(10)
>>> print(lib_example.sum_vector(v))
45