A deep dive into how Apache Airflow manages one of the largest open-source Python monorepos in the world — over 1.2 million lines of Python across 120+ distributions. The episode covers how uv workspaces enforce true dependency isolation between sub-packages, how prek (a Rust-based pre-commit replacement) enables per-module hooks with workspace awareness, and how PEPs 723 and 735 simplify CI scripts and dependency group declarations. A novel shared-libraries pattern using symlinks and vendoring solves the DRY-vs-coupling tradeoff. The guests argue the monorepo-vs-multirepo debate is effectively settled by modern tooling, and share how active collaboration with tool authors (Astral/uv, prek) shaped the ecosystem. Additional topics include IDE configuration scripts for large repos, AI-generated contribution challenges, and how enforced isolation improved Airflow's internal architecture.
Sort: