Disable pip Outside of Virtual Environments

Python packages everywhere!

I’m a huge fan of virtual environments in Python. They are a convenient way to manage dependencies if you are working on more than one Python project at a time. Well, they are the only way to manage dependencies between projects. In the JavaScript world, if you run npm install it will create a local folder with all the packages and use it in your project (falling back to global packages if a dependency is missing). In Python, all your packages are installed in the same place. And if you want to install a different version of a package, the previous one will be uninstalled:

$ pip install pygments==2.2
Collecting pygments==2.2
  Using cached https://files.pythonhosted.org(...).whl
Installing collected packages: pygments
  Found existing installation: Pygments 2.4.2
    Uninstalling Pygments-2.4.2:
      Successfully uninstalled Pygments-2.4.2
Successfully installed pygments-2.2.0

The best you can do in this situation is to install packages into your user directory (with pip install --user), but that doesn’t really solve the problem.

Plenty of tools have been created to solve the dependencies management problem. From the most popular ones like the pipenv or poetry to less popular like hatch (I have yet to meet someone using it) or dephell (that I have heard about at one of the Python conferences). Still, most of the people I know use the same setup as I do - the virtualenv package (with an optional wrapper like virtualenvwrapper or virtualenv burrito). For a long time I didn’t even know that since Python 3.3, the virtualenv is baked into Python through the venv module. You can create virtual environments without any external tools by simply running python3 -m venv.

There is even a PEP 582 suggesting to use local packages directory (à la node_modules). So the landscape of Python dependencies managers might change in the future.

I can talk for hours about how to set up the most efficient workflow for Python. In fact, I will - next year! I'm preparing a free workshop for some of the Python conferences where I will discuss the dependencies management in Python (among other things, like how to set up VS Code for Python). I don't know which conferences I will attend yet, but you can follow me on Twitter or subscribe to the newsletter where I will announce it once it gets accepted.

And if you want an all-day training on this subject - check out my Modern Python Developer's Toolkit workshop!

In my current setup, I’m using virtualenv, so I can just run workon name-of-environment instead of looking where the activate script is placed. On top of that, since I’m using fish shell, I have installed virtualfish to have even more short commands like vf ls or vf cd (as for a programmer, I really don’t like typing).

And, especially at the beginning, I kept forgetting to activate the virtual environment before I cheerfully ran pip install a-package. Or even worse: pip install -r requirements.txt. Which cluttered my global pip directory with one more package (or hundreds of them in case of requirements.txt file). What’s even worse, sometimes it also uninstalled the previous versions of packages. So other projects that I was building stopped working. And if you have the same package installed in a virtual env and globally - it can get messy sometimes.

There had to be a better way!

Make sure that pip only runs in a virtual environment

So one day I said “That’s it! There has to be a way to at least get a warning that pip is running outside of a virtual environment!”. It turns out that of course there is a way. And it’s even built-in into pip! You can set the PIP_REQUIRE_VIRTUALENV environment variable to true and pip will never run outside of a virtual env! Simply add export PIP_REQUIRE_VIRTUALENV=true to your .bashrc or .zshrc (or set -x PIP_REQUIRE_VIRTUALENV true if you are using fish shell). Now, each time you try to run pip outside of a virtual env, it will simply refuse to do so:

$ pip install requests
ERROR: Could not find an activated virtualenv (required).

If you want to actually install something outside of a virtual environment, you can temporarily clear that env variable: env PIP_REQUIRE_VIRTUALENV='' pip install request. Why would you ever want to do that? For example, to install the great pipx tool that lets you further isolate your command line Python packages.

You can also create a bash command to install pip packages that ignores this setting:

gpip() {
  PIP_REQUIRE_VIRTUALENV="" pip "$@"
}

Now I no longer have to worry about installing dependencies outside of a virtual environment!

---

Photo by Tim Evans on Unsplash

Comments

[Kamal Mustafa]

Wow, I don't know PEP582 exists. Few days ago I also was tinkering with local-packages idea.

[Sebastian Witowski]

Neither did I before I started writing this post (I found about it by accident when searching for virtual environments alternatives) 😉

[Kamal Mustafa]

Unfortunately it doesn't seem to go anywhere. Discussion on it stopped around May and look like they have gave up on the idea.

discuss.python.org/t/pep-582-pytho...