Mehul Kar

Oct 17, 2017

Using Python Virtualenvs


I’ve been working on a Python project for the last year. Coming from a cushy “convention over configuration” life of Ruby on Rails and ember-cli, it has been confusing and frustrating to find a pattern for myself to work efficiently. When I asked for help from the community, the general answer I got was “this is how I’ve done it for the last 10 years”, or “here’s a 200 line script I wrote for this”. This was incredibly frustrating for someone like me who likes to do things the right way to avoid digging myself into an unnecessary hole.

One of the main things I had trouble with was learning how to specify dependencies and their versions for a project. I was told to “just use virtualenv” many times, but it was not clear how to actually use it. In the beginning, I tried out the pyenv-virtualenv project, because it seemed to be similar to rvm or nvm. I had hoped to be able to define my Python version and a set of packages with it, but it turns out that it was not meant to do that.

I ended up defining my own conventions for using virtualenv. I’m not sure they are the best conventions, but they work. And so I become another Python developer who will impart conventions to you based on personal usage, rather than community consensus:

  • pip install --user virtualenv globally for your user so that you can always virtual environments. As a side note, I now always use the --user flag with pip install if you are installing outside a virtualenv. This means you don’t have to use sudo or change the ownership of paths in /Library to use global packages.
  • Create one virtualenv per project and activate it when you start working on that project. In Ruby, when you cd into a directory rvm will read a .ruby-gemset file to activate the correct named gemset. This means that your gemset can be stored in a shared space outside your project. However, this involves overriding cd, which can be confusing. It worked best for me to activate my virtualenv manually, but I created a shell function to help:

    function v() {
      echo "Activating $name"
      virtualenv $name > /dev/null
      source "$name/bin/activate"

    If you’re wondering, running this function multiple times is safe, because it turns out that creating a virtualenv from inside a virtualenv doesn’t seem to create a Russian doll hierarchy (thank goodness).

  • Don’t attempt to share virtual environments across projects. The only advantage of this seems to be in automation or if you need to be stringent with disk space usage.

  • Define a requirements.txt that simply contains -e ., so that when you run pip install -r requirements.txt, it installs all the dependencies listed in into your virtualenv. This is a convoluted setup, but it turned out to be more confusing to specify all dependencies in requirements.txt directly.

  • Lastly, don’t let experienced Python devs tell you that tooling in Python is straightforward or “simple”. It is not.

There are several other parts of Python development that I found confusing, such as optimal directory structure and writing tests, but I will save those for a different post some other time.

Want to talk about this blag? Email me or send me a toot @mehulkar!