Using pip-tools to manage my Python dependencies

At last year’s PyCon UK, one of my favourite talks was Aaron Bassett’s session on Python dependency management. He showed us a package called pip-tools, and I’ve been using it ever since.

pip-tools is used to manage your pip dependencies. It allows you to write a top-level summary of the packages you need, for example:

$ cat requirements.in
pytest >= 1.4
requests

Here I want a version of pytest that’s at least 1.4, and any version of requests.

Then I run pip-compile, which turns that into a full requirements.txt:

$ pip-compile
$ cat requirements.txt
certifi==2017.7.27.1      # via requests
chardet==3.0.4            # via requests
idna==2.6                 # via requests
py==1.4.34                # via pytest
pytest==3.2.2
requests==2.18.4
urllib3==1.22             # via requests

I can install these dependencies with pip install -r requirements.txt.

The generated file is pinned: every package has a fixed version. This means that I get the same versions whenever I run pip install, no matter what the new version is. If you don’t pin your dependencies, your package manager may silently install a new version when it’s released – and that’s an easy way for bugs to sneak in.

Instead, check in both files into version control, so you can see exactly when a dependency version was changed. This makes it easier to see if a version bump introduced a bump.

There are also comments to explain why you need a particular package: for example, I’m installing certifi because it’s required by requests.

I’ve been using pip-tools since Aaron’s recommendation, and it’s been really nice. It’s not had an earth-shattering impact on my workflow, but it shaves off a bunch of rough edges. If you do any work with Python, I recommend giving it a look.

For more about pip-tools itself, I recommend Better Package Management by Vincent Driessen, one of the pip-tools authors. This human-readable/pinned-package distinction is coming to vanilla pip in the form of Pipfile, but that was in its infancy last September. pip-tools has been stable for over two years.


Recently, I’ve been trying to push more of my tools inside Docker. Every tool I run in Docker is one less tool I have to install locally, so I can get up-and-running that much faster. Handily, there’s already a Docker image for running pip-tools.

You run it as follows:

$ docker run --volume /path/to/repo:/src --rm micktwomey/pip-tools

It looks for a requirements.in in /src, so we mount the repo in that directory — this gives the container the ability to read the file, and write a requirements.txt back into a file on the host system. I also add the --rm flag, which cleans up the countainer after it’s finished running.

If you already have Docker, this is a nice way to use pip-tools without installing it locally.


Alongside Docker, I’ve been defining more of my build processes in Makefiles. Having Docker commands is useful, but I don’t want to have to remember all the flags every time I use them. Writing a Makefile gives me shortcuts for common tasks.

This is the Make task I have for updating a requirements.txt:

requirements.txt: requirements.in
    docker run --volume $(CURDIR):/src --rm micktwomey/pip-tools
    touch requirements.txt

To use it, run make requirements.txt.

The first line specifies the Make target (requirements.txt), and tells Make that it depends on requirements.in. So when the Make task is invoked, it checks that the .in file exists, and then whether the .in file was updated more recently than .txt. If yes — the .txt file needs rebuilding. If no — we’re up-to-date, there’s nothing to do.

The second line runs the Docker command explained above, using the Make variable $(CURDIR) to get the current directory.

Finally, touch ensures that the last modified time of requirements.txt is always updated. pip-tools will only change the modification time if there are changes to the dependency pins — I change it manually so that make knows the task has run, and the “should I run this task” logic explained above doesn’t spin endlessly.

Once I have this Make task, I can invoke it from other tasks — for example, build tasks that install from requirements.txt — and so it gets run when required, but without an explicit action from me. It’s just another step that happens transparently when I run make build.

If you’d like to see an example of this in use, check out the Makefile changes in the same patch as this post.