Can others reproduce your results? Recipes for using uv and pixi to manage Python dependencies

Reproducible environments are crucial for ensuring consistent and reliable software development. They enable RSEs to recreate exact configurations, facilitating the long-term maintenance of projects that may include collaborators with varied levels of software engineering experience. But they're not always easy to achieve. In the worst case scenario, a complicated or difficult pathway during the setup of a project make the whole thing inaccessible to entry-level coders.

This walkthrough will demonstrate the common pitfalls experienced when using pip or conda alone (package conflicts, inconsistent state, and the inability to work across platforms and architectures), and use these to motivate a desire for something better. It will then explore how environment management tools like uv and pixi can help. It will include practical steps and examples of how an RSE, data scientist or scientific coder might incorporate them into a project. The sharing of advice and experiences of attendees would also be welcome!

This walkthrough will be of interest to people that use pip or conda/mamba to install packages for a Python project, or a more established tool like poetry, Pipenv or pip-tools. It will be especially useful to people that collaborate with researchers that have less software engineering experience, but still need to run and develop the project.

This work was developed with support and funding from the Jean Golding Institute at the University of Bristol. I am grateful to colleagues at the Institute and across the University for helpful conversations when preparing this material.