Pyenv – lets you easily switch between multiple versions of Python

Original Article ↗ Hacker News Discussion ↗

Role of pyenv

  • Used to install and switch between multiple Python versions, independent of the system Python.
  • Common workflows:
    • pyenv for interpreter versions → venv or pyenv-virtualenv for environments → Poetry/PDM/pip-tools for deps.
    • Matching production runtimes (e.g., specific 3.x.y) or old/new versions on the same machine.
  • Some use it to build optimized interpreters (LTO, native CPU flags) or to keep system site-packages clean.

Critiques & Caveats of pyenv

  • Major concern: it usually compiles CPython from source, which:
    • Requires build toolchains and libraries; often fails on macOS and Linux without careful setup.
    • Increases CI time and image size when used inside containers.
  • Shims add another layer where PATH / resolution bugs can hide.
  • Some ops/devs report repeatedly fixing “mysterious” Python issues by removing pyenv; view it as a wrapper that adds failure modes and is better suited for experts than beginners.
  • Others counter that building Python is trivial, have used pyenv for years without trouble, and value fine-grained control.

Alternatives for Python Versions & Envs

  • System + venv: Many advocate python3.X -m venv .venv plus requirements files as enough for most work; simple and built-in.
    • Counterpoint: venv only uses already-installed interpreters and often symlinks to them, so OS/installer upgrades can silently change the runtime.
  • Multi-language managers: asdf and mise manage Python alongside Node, Ruby, etc.; mise is favored by some for performance and nicer CLI, often reusing asdf/pyenv plugins.
  • Conda/mamba/pixi: Still popular where non-Python binaries (CUDA, geospatial stack) are needed; pixi adds lockfiles and PyPI support (via uv).
  • Docker / Nix:
    • Docker widely used to sidestep host Python issues, especially with NVIDIA base images for ML.
    • Nix highlighted as one of the few options that can truly reproduce full environments (Python + system libs) over time.

Python Packaging Meta

  • Many complain about fragmentation: pip, venv, virtualenv, conda, poetry, PDM, pipenv, Rye, uv, pipx, etc.
  • Consensus that there is no single “silver bullet,” especially for beginners; recommendations range from “just use venv” to “just use Docker” to “use Nix if you can tolerate the complexity.”