I wrote a scientific C++ simulation that I would like to wrap in a Python package so that users can easily install it via pip and run it from within Python. I am new to packaging and I need advice on which approach to choose. The packaging best-practices seem to be subject to on-going change, and the added complexity that the C++ part should run on any architecture leaves me a bit overwhelmed. My current setup is a C++ compiled binary that is run from within Python via subprocess.Popen.On top of that, there are a few added Python files containing pure Python data processing utility. I understand that there are two possible approaches: Ship the package with compiled binaries. This can be achieved by using setuptools and cibuildwheels to create wheels for any system architecture. Have the installation process compile the C++ source (by adding a CMake file to the package) such that I don't need to bother with shipping system-specific files/binaries. As the package is still being developed, I think going with option 2) would be better for now. On a sidenote, rather than running a binary from within Python, I also tried using C++ bindings like PyBind11 and Nanobind, but I observed a significant efficiency drop compared to calling a binary via subprocess.Popen . Any comments on what the best approach is for this endeavour or helpful links to up-to-date guidance would be appreciated. Continue reading...
