Developer Guide

Design Philosophy

Here are some guiding principles we have used when building pyvale:

  • Scientists and engineers want a python interface (preferably a GUI), they will not want to build code from source and they want to work on the OS of their choice:

    • Installation should be as simple as pip install pyvale and work across all platforms.

    • Write computationally demanding algorithms in lower level languages like Cython, C/C++ but provide a python interface.

  • Users want to get started quickly but have the option to customise everything:

    • Provide a simple interface with as many defaults set as possible.

    • Provide examples that build in complexity while covering all possible features.

    • Good documentation is essential.

Examples for successful python packages that have achieved this include: numpy, matplotlib, scipy and scikit-learn. The documentation for matplotlib is a particularly good reference example for pyvale and the simple interface of scikit-learn into many machine learning models also serves as a good reference.

These are some resources we have drawn inspiration from when writing our code, especially the computationally intensive simulation modules:

Coding Languages

All user interfaces in pyvale should be written in Python to allow ease of use for the general engineering and scientific community. Code where performance is required (e.g. rendering engines, image analysis and digital image correlation analysis) or where you feel the need to write a for loop should preferably be written in a compiled language. A list of preferred coding languages for pyvale is given below:

  • Python

  • Cython

  • C/C++

  • Zig

The following can be used for linking compiled code to a python interface:

  • Cython

  • Pybind

For Zig we support a custom build process through our setup.py that pulls in the Zig compiler on pypi to build and dynamically link Zig libraries through the C ABI and Cython. Unfortunately, this requires a lot more boiler plate code (i.e. a C header and a .pxd cython header restating the same thing) than using Pybind with C/C++.

GPU programming should be vendor agnostic. The following can be used for GPU programming:

Further Information