This is an in-progress draft, enjoy.


You've probably installed Python libraries before using pip like so:

pip install Django==4.9.3

And then like magic you could start using Django scripts and importing Django code into your own project.

But how did that command get Django installed? We're not going to do a deep dive on packaging here because it's a lengthy subject, but we are going to explore enough so that you understand how to take your app and make it installable and usable by other users.

The basics:

The basics of a Python package include a (1) Python module and (2) a setup file that identifies your package and locates the code.

At a bare minimum you'll need a file that tells the package manager infomration about your Python package such as it's name, the version, and what source code to include when installing.

def hello_world():
    return {}
from distutils.core import setup


Okay, it's a silly module. But it's an installable silly module. The first part should be self explanatory. It's a Python module. The does one very important thing, it calls setup from distutils whcih is Python's way of declaring this a module. This file is what you will use when intsalling a package, either directly or indirectly. Without getting into the details of all of the possible arguments you can see that we're providing meta information, e.g. who wrote it, the name of the package, and where to find the packages.

This file provides data at the installation level so that when you run python install Python knows what to get out to install on your Python path.

  • where the package itself is, i.e. the source of the Python modules
  • where tests are
  • what else is required for the package to install and/or test
  • descriptive information about the app itself

It also provides useufl data for package indexes, like PyPI (nee Cheeseshop). Additional argumetns like license make it easy for people to search for modules by how the software is licences, and classifiers provies useful information for searching for software by targeted purpose, environment, and Python version.


The version is critical. It needs to be a string, and you can hard code it here, but that turns out to be a pain in the tuchus. Look, you probably declare the version elsewhere, so declare it one place and be done with it. That one place should be in the package root itself, either in or mypackage/ The canonoical way of definit it is with the variable name __version__.

To get this into your package's file, either import or, if that causes problems, read it from the file directly as text.


The most important thing you'll use classifiers for is specifying which Python versions your code works with.

Classifiers won't create hard constraints on how your package can be installed - at least using typical tools - but they do provide strucuted information in the package index that allow people to find useful software and identify if itwill work for them.

As we're still in the transition phase between Python 2 and Python 3, including the Pyton version classifier is one of the most important you can fill out. Beyond that you can include other classifiers such as the stability of your package (e.g. is it alpha? stable?).

Everything else

Your Python package will need to include more than just your code and a file. Here's a short overview of what else should be included.


Hopefully this one is obvious. But what should this include? At a minimum it's helpful to include a brief description of the project, author identification with contact info, and initial installation, configuration, and usage instructions. It needn't cover everything.

Thanks to the GitHubification of open source software, README files are read just as much as web pages as they are in text editors, and so the content has changed. They can include project diagrams, logos, dynamic images for build progress, in short they often serve as much a purpose as marketing material as they do mini-documentation.

If you're keen on promoting your app then you'll probably want to delve into all of that, but consider that a well written README, written for the developer trying to understand the code in front of them, will typically server as decent marketing, whereas great marketing READMEs dont' necessarily work for the developer's aid.

Your README can be in text format, but if you're going to put your package on PyPI you should use reStructuredText.


More on this later, but you should include a file explaining ownerhip of your code and how it's licensed for end users.


You probably know what this is already, but it's common to see project requirements, dependencies, included in a pip requirements file. This may be joined by additional files like a requirements-test.txt file specifying dependencies used solely for testing.

If the former is present by itself it's frequently read from to populate the file in an effort to break out more of the information.

A separate testing requirements file is very helpful and these dependencies are not required and needn't be installed in the deployed environment and in many cases testing requires a few additional packages.

One thing the file doesn't include is project assets. There's a package specification and a test packages spec, but it may come as a surpise that this only includes Python files. If you're writing a number crunching library this might of no import to you (rimshot) but if you've got templates or images, you're at risk of losing those.

What's this manifest business? We need to tell what to include in the package. By default it will include Python files, but often we want to include other files, like a README or assets like HTML and CSS files. These we specify in the file.

include README.rst
include LICENSE
recursive-include myapp/templates *


What's this setup.cfg file? That's a configuration file that we're using here for creating a wheel file. A wheel is a way of packaging Python programs that replaces the previous style, called an 'egg'. I kind of prefer the term 'egg', it fits more with the Python namespace, if you will, although Larch might have been even better. Wheel can be configured in several ways and we want to just make universal wheels.

universal = 1

There are other uses for this file that go beyond the scope of what we're discussing here.


This is easily my favorite of these files (yes, I have a favorite, and no I don't get out much). We'll go into greater depth about how this file is used in the advanced section on testing, but is short it's a configuraiton file used primarily but exclusively by tox, a tool for managing test environments. It's worth mentioning here because you'll likely see this in many Python packages.


Sometimes named HISTORY, this file should be used to catalog what changed in the package from version to version. When properly updated this tells your users when you introduced a backwards incompatible change (and why) explaining the justification for a new release.

There's a chapter on releases, but suffice to say there's a lot of value in keeping this kind of thing uptodate even for small projects that the public will never see.