pbr - Python Build Reasonableness

A library for managing setuptools packaging needs in a consistent manner.

pbr reads and then filters the setup.cfg data through a setup hook to fill in default values and provide more sensible behaviors, and then feeds the results in as the arguments to a call to setup.py - so the heavy lifting of handling python packaging needs is still being done by setuptools.

Note that we don’t support the easy_install aspects of setuptools: while we depend on setup_requires, for any install_requires we recommend that they be installed prior to running setup.py install - either by hand, or by using an install tool such as pip.

What It Does

PBR can and does do a bunch of things for you:

  • Version: Manage version number based on git revisions and tags
  • AUTHORS: Generate AUTHORS file from git log
  • ChangeLog: Generate ChangeLog from git log
  • Sphinx Autodoc: Generate autodoc stub files for your whole module
  • Requirements: Store your dependencies in a pip requirements file
  • long_description: Use your README file as a long_description
  • Smart find_packages: Smartly find packages under your root package


Versions can be managed two ways - postversioning and preversioning. Postversioning is the default, and preversioning is enabeld by setting version in the setup.cfg metadata section. In both cases version strings are inferred from git.

If a given revision is tagged, that’s the version.

If it’s not, then we take the last tagged version number and increment it to get a minimum target version.

We then walk git history back to the last release. Within each commit we look for a Sem-Ver: pseudo header, and if found parse it looking for keywords. Unknown symbols are not an error (so that folk can’t wedge pbr or break their tree), but we will emit an info level warning message. Known symbols: feature, api-break, deprecation, bugfix. A missing Sem-Ver line is equivalent to Sem-Ver: bugfix. The bugfix symbol causes a patch level increment to the version. The feature and deprecation symbols cause a minor version increment. The api-break symbol causes a major version increment.

If postversioning is in use, we use the resulting version number as the target version.

If preversioning is in use - that is if there is a version set in setup.cfg metadata - then we check that that version is higher than the target version we inferred above. If it is not, we raise an error, otherwise we use the version from setup.cfg as the target.

We then generate dev version strings based on the commits since the last release and include the current git sha to disambiguate multiple dev versions with the same number of commits since the release.


Note that pbr expects git tags to be signed, for using it to calculate version.

The versions are expected to be compliant with Linux/Python Compatible Semantic Versioning 3.0.0.

The version.SemanticVersion class can be used to query versions of a package and present it in various forms - debian_version(), release_string(), rpm_string(), version_string(), or version_tuple().

AUTHORS and ChangeLog

Why keep an AUTHORS or a ChangeLog file, when git already has all of the information you need. AUTHORS generation supports filtering/combining based on a standard .mailmap file.

Sphinx Autodoc

Sphinx can produce auto documentation indexes based on signatures and docstrings of your project- but you have to give it index files to tell it to autodoc each module. That’s kind of repetitive and boring. PBR will scan your project, find all of your modules, and generate all of the stub files for you.

Sphinx documentation setups are altered to generate man pages by default. They also have several pieces of information that are known to setup.py injected into the sphinx config.

See the pbr section for details on configuring your project for autodoc.


You may not have noticed, but there are differences in how pip requirements.txt files work and how distutils wants to be told about requirements. The pip way is nicer, because it sure does make it easier to populate a virtualenv for testing, or to just install everything you need. Duplicating the information, though, is super lame. So PBR will let you keep requirements.txt format files around describing the requirements for your project, will parse them and split them up appropriately, and inject them into the install_requires and/or tests_require and/or dependency_links arguments to setup. Voila!

You can also have a requirement file for each specific major version of Python. If you want to have a different package list for Python 3, just drop a requirements-py3.txt, and it will be used instead.

The requirement files are tried in that order (N being the Python major version number used to install the package):

  • requirements-pyN.txt
  • tools/pip-requires-py3
  • requirements.txt
  • tools/pip-requires

Only the first file found is used to install the list of packages it contains.

Extra requirements

Groups of optional dependencies (“extra” requirements) can be described in your setup.cfg, rather than needing to be added to setup.py. An example (which also demonstrates the use of environment markers) is shown below.

Environment markers

Environment markers are conditional dependencies which can be added to the requirements (or to a group of extra requirements) automatically, depending on the environment the installer is running in. They can be added to requirements in the requirements file, or to extras definied in setup.cfg - but the format is slightly different for each.

For requirements.txt:

argparse; python=='2.6'

will result in the package depending on argparse only if it’s being installed into python2.6

For extras specifed in setup.cfg, add an extras section. For instance, to create two groups of extra requirements with additional constraints on the environment, you can use:

security =
testing =


There is no need to maintain two long descriptions- and your README file is probably a good long_description. So we’ll just inject the contents of your README.rst, README.txt or README file into your empty long_description. Yay for you.


pbr requires a distribution to use setuptools. Your distribution must include a distutils2-like setup.cfg file, and a minimal setup.py script.

A simple sample can be found in pbr’s own setup.cfg (it uses its own machinery to install itself):

name = pbr
author = OpenStack Foundation
author-email = openstack-dev@lists.openstack.org
summary = OpenStack's setup automation in a reusable form
description-file = README
home-page = https://launchpad.net/pbr
license = Apache-2
classifier =
    Development Status :: 4 - Beta
        Environment :: Console
        Environment :: OpenStack
        Intended Audience :: Developers
        Intended Audience :: Information Technology
        License :: OSI Approved :: Apache Software License
        Operating System :: OS Independent
        Programming Language :: Python
keywords =
packages =
data_files =
    etc/pbr = etc/*
    etc/init =
console_scripts =
    pbr = pbr.cmd:main
pbr.config.drivers =
    plain = pbr.cfg.driver:Plain

The minimal setup.py should look something like this:

#!/usr/bin/env python

from setuptools import setup


Note that it’s important to specify pbr=True or else the pbr functionality will not be enabled.

It should also work fine if additional arguments are passed to setup(), but it should be noted that they will be clobbered by any options in the setup.cfg file.


The format of the files section is worth explaining. There are three fundamental keys one is likely to care about, packages, namespace_packages, and data_files.

packages is a list of top-level packages that should be installed. The behavior of packages is similar to setuptools.find_packages in that it recurses the python package hierarchy below the given top level and installs all of it. If packages is not specified, it defaults to the name given in the [metadata] section.

namespace_packages is the same, but is a list of packages that provide namespace packages.

data_files lists files to be installed. The format is an indented block that contains key value pairs which specify target directory and source file to install there. More than one source file for a directory may be indicated with a further indented list. Source files are stripped of leading directories. Additionally, pbr supports a simple file globbing syntax for installing entire directory structures, so:

data_files =
    etc/pbr = etc/pbr/*
    etc/neutron =
    etc/init.d = neutron.init

Will result in /etc/neutron containing api-paste.ini and dhcp-agent.ini, both of which pbr will expect to find in the etc directory in the root of the source tree. Additionally, neutron.init from that dir will be installed in /etc/init.d. All of the files and directories located under etc/pbr in the source tree will be installed into /etc/pbr.

Note that this behavior is relative to the effective root of the environment into which the packages are installed, so depending on available permissions this could be the actual system-wide /etc directory or just a top-level etc subdirectory of a virtualenv.


The general syntax of specifying entry points is a top level name indicating the entry point group name, followed by one or more key value pairs naming the entry point to be installed. For instance:

console_scripts =
    pbr = pbr.cmd:main
pbr.config.drivers =
    plain = pbr.cfg.driver:Plain
    fancy = pbr.cfg.driver:Fancy

Will cause a console script called pbr to be installed that executes the main function found in pbr.cmd. Additionally, two entry points will be installed for pbr.config.drivers, one called plain which maps to the Plain class in pbr.cfg.driver and one called fancy which maps to the Fancy class in pbr.cfg.driver.


The pbr section controls pbr specific options and behaviours.

The autodoc_tree_index_modules is a boolean option controlling whether pbr should generate an index of modules using sphinx-apidoc. By default, setup.py is excluded. The list of excluded modules can be specified with the autodoc_tree_excludes option. See the sphinx-apidoc man page for more information.

The autodoc_index_modules is a boolean option controlling whether pbr should itself generates documentation for Python modules of the project. By default, all found Python modules are included; some of them can be excluded by listing them in autodoc_exclude_modules. This list of modules can contains fnmatch style pattern (e.g. myapp.tests.*) to exclude some modules.

The warnerrors boolean option is used to tell Sphinx builders to treat warnings as errors which will cause sphinx-build to fail if it encounters warnings. This is generally useful to ensure your documentation stays clean once you have a good docs build.


When using autodoc_tree_excludes or autodoc_index_modules you may also need to set exclude_patterns in your Sphinx configuration file (generally found at doc/source/conf.py in most OpenStack projects) otherwise Sphinx may complain about documents that are not in a toctree. This is especially true if the warnerrors=True option is set. See http://sphinx-doc.org/config.html for more information on configuring Sphinx.


Comments may be used in setup.cfg, however all comments should start with a # and may be on a single line, or in line, with at least one white space character immediately preceding the #. Semicolons are not a supported comment delimiter. For instance:

# A comment at the start of a dedicated line
key =
    value1 # An in line comment
    # A comment on a dedicated line

Indices and tables