Posted on 10 mins read


This blog post aims to demonstrate the most practical way to install multiple versions of Python, and of setting up ‘virtual environments’ for macOS userso

We’ll also dig into how to manage our project dependencies (e.g. we’ll be discussing the classic Pip and requirements.txt format) + our approach for avoiding newer tools such as Poetry and Pipenv, which I feel are just too complicated/overkill for the majority of use cases.

Why is this not a simple thing to do? I hear you ask! Well, because we don’t live in a perfect world and sometimes the overhead of ‘convenience’ can be worth the cost otherwise incurred, and sometimes it can’t.

I’ll start by describing briefly what a virtual environment is, and then I’ll move onto the point of this article and the options we have available to us.

But before we get into it... time for some self-promotion 🙊

Virtual Environments

A cooperatively isolated runtime environment that allows Python users and applications to install and upgrade Python distribution packages without interfering with the behaviour of other Python applications running on the same system. – Python Glossary

To simplify: if you work on multiple Python projects and each project uses the same external dependency (e.g. the Requests HTTP library), then it’s possible your application (for each project) will be written to support a specific version of that dependency.

Having a virtual environment setup for each project means you can have project specific Python package installs. For example, project ‘foo’ can use the Request library version 1.0 while project ‘bar’ can use version 2.0 (which might introduce a different API).

Note: the official documentation on Virtual Environments can be found here.

Creating Virtual Environments

If you installed Python using Homebrew (e.g. brew install python3) and you don’t care what version of Python you have, you only care about having virtual environments, then you can utilize the built-in venv module like so:

python3 -m venv /foo/bar
source /foo/bar/bin/activate
python3 -m pip install <dependencies>

This will ensure you only install third-party packages/modules into the specific virtual environment you’ve just activated.

The only downside of this very simple approach is that installing Python via Homebrew means you’ll have only a single version of Python installed and you have no control over what that version is, let alone have multiple versions installed (as the python3 command will be overwritten each time).

This consideration is actually critical to understand when you consider the Linux OS internals use the same Python version as exposed to the end user. Meaning if you mess up your Python install on Linux, then there is a high chance you’ll break the entire operating system!

Note: if you want to prevent accidentally executing pip install outside of a virtual environment then use export PIP_REQUIRE_VIRTUALENV=true (it can also be set in a ~/.pip/pip.conf)

If you require virtual environments across multiple Python versions, then read the following couple of sections…

Installing Python Versions

So there are in fact two ways of installing Python versions, but it’s much less practical in reality and so i’m only going to demonstrate one of them.

The two ways are:

  • Manual compilation
  • External build tool

As you can imagine, manually compiling Python would be the more flexible solution, but the harsh reality is that compiling Python requires very specific dependencies that can be hard to get right (and most often it goes wrong).

So instead we’ll focus on the latter: an external tool called pyenv which internally uses another external tool called python-build.

  • pyenv: let’s us switch Python versions easily
  • python-build let’s us install Python versions easily.

You don’t need to install python-build directly as it’ll be installed when you install pyenv.

To install pyenv execute:

brew install pyenv

Note: yes, this means you need to be using Homebrew, but let’s face it, it’s the defacto standard for macOS package management.

Once installed you’ll be able to use the following commands:

  • python-build --definitions: list all versions of Python available to be installed
  • pyenv install <version>: install the version of Python you need

Once you have installed the version of Python you need, now you just need to remember to ‘activate’ it whenever you’re working on your project that requires that specific version of Python. To do that, you’ll need to do two things (and you only need to do them once):

  • add eval "$(pyenv init -)" to your .bashrc (or shell of choice)
  • execute pyenv local <version> in your project directory

What the first point does is it’ll allow your shell to respond to any .python-version file found within a directory on your computer. This file will contain a Python version.

What generates the .python-version file is the latter point.

Virtual Environments for multiple Pythons

To setup virtual environments with Python is actually very simple (as we saw in the earlier part of this post), but not compatible when using an external build tool such as pyenv because of where pyenv installs Python binaries and how it switches between versions.

But luckily there is an extension to pyenv called pyenv-virtualenv which can be installed like so:

brew install pyenv-virtualenv

Once installed, setting up a new virtual environment is as simple as:

pyenv virtualenv foobar
pyenv activate foobar
pyenv deactivate foobar

Note: if you want you can specify the Python version to create the virtual environment for: pyenv virtualenv <version> <name>.

Shell Configuration

This is just a quick summary of the configuration lines added to my .bashrc:

eval "$(python3 -m pip completion --bash)"
eval "$(pyenv init -)"
eval "$(pyenv virtualenv-init -)"

Managing Dependencies

When it comes to dealing with specific dependency versions, I like to use the method Kenneth Reitz published back in 2016.

Note: this method keeps with the traditional requirements.txt file as utilized by Pip. I mention this as you’ll notice with other tools (such as Pipenv or Poetry), that they move away from this established format and that can be a bit disruptive in terms of how Python teams have traditionally worked. I’m not saying it’s a bad thing, but change isn’t always good.

Problem Summary

Here is a summary of the problem we’re trying to solve:

The requirements.txt file typically doesn’t include the sub-dependencies required by your top-level dependencies (because that would require you to manually identify them and to type them all out, something that should be an automated process and so in practice never happens manually).

e.g. you specify a top-level dependency of foo (which might install version 1.0), but that internally requires the use of other third-party packages such as bar and baz (and specific versions for each of them).

But a pip install from a file that only includes the top-level dependencies could (over time) result in different sub-dependency versions being installed by either different members of your team or via your deployment platform.

e.g. if you don’t specify a version for foo, then in a months time when someone else (or your deployment platform) runs pip install it will attempt to install the latest version of foo which might be version 2.0 (and subsequently the third-party packages it uses might also change).

To avoid that people have come to use pip freeze after doing a pip install to overwrite their requirements.txt with a list of all dependencies and their explicit versions.

This solves the issue of installing from requirements.txt in a months time when lots of your top-level dependencies release new breaking versions.

The problem now is that you have to manually search for the top-level dependencies (in this new larger/more-indepth requirements.txt file) and update them manually. Doing this might break things as you now don’t know what the sub-dependency versions should be set to.


So the approach we take with any project is to define a requirements-to-freeze.txt file. This file will contain all your project’s top-level dependencies (inc. any explicit versions required), for example:


Next we can generate our actual requirements.txt file based upon the contents of requirements-to-freeze.txt using the pip freeze command, like so:

python -m pip install -r requirements-to-freeze.txt
python -m pip freeze > requirements.txt

Which will result in a requirements.txt file that looks something like:


This means you’ll never manually update requirements.txt again. Any time you need to update a dependency you’ll do it in requirements-to-freeze.txt, then re-run:

python -m pip install -r requirements-to-freeze.txt
python -m pip freeze > requirements.txt

Or instead of manually updating the dependencies in requirements-to-freeze.txt you could use the --upgrade flag:

python -m pip install -r requirements-to-freeze.txt --upgrade
python -m pip freeze > requirements.txt

To make the process easier (see notes below for details) we can utilize a bash shell script (and docker) to help us:


# A quick and easy way to exhaustively freeze a set of "top level" Python
# dependencies for services following the "Better Pip Workflow" approach:
# This script makes it easy to generate the frozen requirements without needing
# to juggle fresh virtualenvs or compile-time dependencies by instead
# installing the requirements in an ephemeral docker container.
# Example usage:
#     .freeze-requirements path/to/requirements-to-freeze.txt > path/to/requirements.txt
# To install extra base OS packages (e.g. mysql-dev), specify them in PACKAGES on the command line:
#     PACKAGES="mysql-dev" ./freeze-requirements ...

set -e
set -u
set -o pipefail


DEFAULT_PACKAGES="gcc python-dev libssl-dev"



if [ "$requirements_file" == "" ]; then
    echo "Usage: $(basename $0) REQUIREMENTS_FILE" >&2
    exit 1

if [ ! -f "$requirements_file" ] && [ "$requirements_file" != "-" ]; then
    echo "File not found: $requirements_file" >&2
    exit 1

docker build -t $TAG - >&2 <<EOF
FROM python:${PYTHON_VERSION}-slim

RUN apt-get update && apt-get install -y ${PACKAGES}
RUN pip install virtualenv && virtualenv /venv

cat $requirements_file | exec docker run --rm -i -a stdin -a stdout -a stderr $TAG sh -c '
cat >/tmp/requirements-to-freeze.txt
/venv/bin/pip install -r /tmp/requirements-to-freeze.txt >&2
/venv/bin/pip freeze -r /tmp/requirements-to-freeze.txt'

Caching Dependencies

Starting with pip version 6.0 you can prevent having to reinstall dependencies that are used across multiple virtual environments by caching them (this is especially useful with Continuous Integration builds).

To do so, add the following to your ~/.bashrc file:

export PIP_DOWNLOAD_CACHE=$HOME/.pip/cache

Alternatively add it to your ~/.pip/pip.conf file:

require-virtualenv = true
download-cache = $HOME/.pip/cache

Command Line Packages

As a bonus section I’m going to quickly mention the tool pipx which allows us to install Python command line tools such that they are isolated binaries and so they don’t clutter up our top-level Python runtime space.

To install pipx, the official instructions state you can install it via either Homebrew or via an existing Python interpreter.

Homebrew install is as follows:

brew install pipx
pipx ensurepath

Using an existing Python interpreter:

python3 -m pip install --user pipx
python3 -m pipx ensurepath

But I found it didn’t work unless I omitted the --user flag:

python3 -m pip install pipx
python3 -m pipx ensurepath

After that you can add the following to your ~/.bashrc (or similar for whatever shell you use):

pipx completions
# eval "$(register-python-argcomplete pipx)"

Now you’re able to safely install command line Python tools, like so:

pipx install pycowsay

Note: if you use the PIP_REQUIRE_VIRTUALENV setting (mentioned earlier in this post) but you also installed pipx via Homebrew, then you’ll find that doing so can cause problems because pipx can’t use its internal list function (as no virtual environment is currently activated). So to fix the issue always set the environment variable to false: PIP_REQUIRE_VIRTUALENV=false pipx list.

But before we wrap up... time (once again) for some self-promotion 🙊