Conda: A package management disaster?

Original Article ↗ Hacker News Discussion ↗

Article & Site Presentation

  • Many readers found the blog page nearly unreadable due to CSS (e.g., forced word-breaking), on both mobile and desktop.
  • Several note the article is largely a curated email thread from the Python mailing list.
  • Some claim parts of the article are inaccurate or confused (e.g., multiple NumPy versions in one process, Jupyter/kernel behavior, “current directory” module shadowing being unrelated to Conda).

Conda: Pain Points

  • Frequent complaints about extremely slow and sometimes “stuck” dependency resolution, especially with conda‑forge and larger environments.
  • Mixing conda and pip in the same environment is widely viewed as a major source of breakage.
  • Some describe Conda environments as fragile when shared or reproduced across machines; others say it works fine if you treat envs as disposable and recreate from YAML.
  • Newer libmamba-based solving is reported as faster, but several still consider performance bad.
  • Some avoid Conda entirely due to prior bad experiences or due to new licensing limits for large companies.

Conda: Strengths & Use Cases

  • Strong support for Windows and compiled scientific stacks was the original killer feature (SciPy/NumPy, CUDA, GDAL, etc.).
  • Handles non-Python dependencies and multi-language stacks (C/C++/Fortran, R, Java, Node, command-line tools), which pip/PyPI generally do not.
  • Popular in bioinformatics and scientific computing because it can install almost all domain tools from one ecosystem.
  • Seen as valuable for corporate environments needing central control, mirroring, access policies, and reproducibility.

Alternatives & New Tools

  • Many users now prefer uv for Python-only workflows: very fast, PEP-compliant, and a potential replacement for pip/poetry/pipx.
  • pixi is highlighted as “Conda done right”: project-local environments, fast solving, conda-style binary ecosystem plus PyPI via uv.
  • mamba is recommended as a drop-in, faster Conda CLI.
  • Nix (and tools built on it) is praised for cross-language, fully reproducible environments, sometimes replacing Conda altogether.

Broader Python Packaging Debate

  • Many see Python packaging as unusually fragmented (pip, venv, conda, poetry, etc.) and historically under-designed compared to ecosystems with a single “blessed” tool.
  • Others argue venv + pip (and now wheels) are adequate for many projects, especially outside Windows and heavy scientific/ML workloads.