Comments by "Gilad Barlev" (@GSBarlev) on "Arch Linux's Python Protects Me From Myself" video.

  1. As a professional data scientist, the first rule I learned developing on my company-issued mac was never use the system Python That was largely because of this dependency management shenanigans, but also because the macOS version of Python was extremely outdated at that time (tied to the fact that for corporate security reasons we were always 1-2 OS releases behind the latest from Cupertino). Honestly, the first thing I usually do on any system is install mambaforge and edit my bashrc to auto-activate the base environment. That's a habit I've had to unlearn with Arch given just how many build scripts in the AUR use the session's Python executable rather than explicitly specifying to use the system Python.
    7
  2. FYI, poetry and conda are both extremely mature solutions for this problem. You specify your development or execution environment with a TOML/YAML file and can create your virtual environment in a single command with a little more functionality than you get from pipx.
    4
  3. Yes. By default the apps are fully isolated. You can reuse shared dependencies if you ask nicely, though. I personally use conda, which works like flatpak, storing each dependency in a shared repo and then building the full virtual environment from symlinks. The pro is that you don't have 52 identical copies of the numpy stack on your system. The con is that if you don't periodically run conda clean, you end up with a TON of unused dependencies cluttering your system.
    3
  4. Have you tried mamba? Drop-in replacement for conda, but the solver is ridiculously faster. IMO mambaforge should be everyone's first and last choice of a portable Python installation.
    1