Self-contained Python scripts with uv

Original Article ↗ Hacker News Discussion ↗

Use cases and appeal

  • Many see uv-based scripts as a lightweight alternative to Docker or full packaging, especially for internal tools and small utilities.
  • uv’s speed and automatic Python/version handling are praised; it solves “which Python, which venv, which requirements” issues on heterogeneous machines.
  • Common suggested use: repo-local dev/utility scripts, not necessarily end-user application distribution.

Windows and cross-platform behavior

  • Shebang-based usage is tricky on Windows: native shebang isn’t supported, though workarounds include:
    • Using the py launcher (from the CPython installer) which honors shebangs.
    • Re-associating .py (or custom extensions) with uv run.
    • Running via WSL.
  • Some report success using uv init/add/run on Windows, but flags like --dev currently don’t work with --script.

Licensing, trust, and long-term viability

  • Comparisons to Anaconda’s licensing change prompt questions about uv’s safety as a long-term dependency tool.
  • Commenters note uv’s permissive licensing and that existing open-source code can’t be back-licensed, but also stress:
    • Open source doesn’t guarantee maintenance.
    • uv being written in Rust means relying on a Rust community that cares about Python.

Inline metadata vs comments / PEP 723

  • Debate over using comment blocks for dependency metadata:
    • Critics dislike executable behavior hinging on comments; prefer real Python data structures or JSON.
    • Others argue for a declarative, restricted format; executing Python to discover dependencies would break portability and bootstrapping.
  • Several highlight this is standardized via PEP 723 (inline script metadata), not uv-specific, and is deliberately tool-agnostic.

Editor, LSP, and tooling integration

  • Some struggle getting LSPs (e.g., pyright) to see dependencies from inline metadata.
  • Workarounds involve wrapper scripts that export requirements and re-run editors via uv.
  • Requests for native support in editors (e.g., VS Code Python extension) are linked.

“Self-contained” vs reality

  • Multiple comments challenge the “self-contained” label:
    • uv must be preinstalled; scripts also fetch packages from the network and cache them.
    • Caches/venvs persist unless explicitly cleared, though uv deduplicates packages.
  • Compared approaches:
    • Nix/nix-shell shebangs (stronger system requirements but no separate Python install).
    • Classic bundlers/compilers (PyInstaller, py2exe, Nuitka) that produce true standalone binaries.
    • Simple venv-based bootstrap scripts for environments where Python is already present.