2. Specification of the setup.cfg file

version:0.9

This document describes the setup.cfg, an ini-style configuration file used by Packaging to replace the setup.py file used by Distutils. This specification is language-agnostic, and will therefore repeat some information that’s already documented for Python in the configparser.RawConfigParser documentation.

2.1. Syntax

The ini-style format used in the configuration file is a simple collection of sections that group sets of key-value fields separated by = or : and optional whitespace. Lines starting with # or ; are comments and will be ignored. Empty lines are also ignored. Example:

[section1]
# comment
name = value
name2 = "other value"

[section2]
foo = bar

2.1.1. Parsing values

Here are a set of rules to parse values:

  • If a value is quoted with " chars, it’s a string. If a quote character is present in the quoted value, it can be escaped as \" or left as-is.
  • If the value is true, t, yes, y (case-insensitive) or 1, it’s converted to the language equivalent of a True value; if it’s false, f, no, n (case-insensitive) or 0, it’s converted to the equivalent of False.
  • A value can contain multiple lines. When read, lines are converted into a sequence of values. Each line after the first must start with a least one space or tab character; this leading indentation will be stripped.
  • All other values are considered strings.

Examples:

[section]
foo = one
      two
      three

bar = false
baz = 1.3
boo = "ok"
beee = "wqdqw pojpj w\"ddq"

2.1.2. Extending files

A configuration file can be extended (i.e. included) by other files. For this, a DEFAULT section must contain an extends key whose value points to one or more files which will be merged into the current files by adding new sections and fields. If a file loaded by extends contains sections or keys that already exist in the original file, they will not override the previous values.

Contents of one.cfg:

[section1]
name = value

[section2]
foo = foo from one.cfg

Contents of two.cfg:

[DEFAULT]
extends = one.cfg

[section2]
foo = foo from two.cfg
baz = baz from two.cfg

The result of parsing two.cfg is equivalent to this file:

[section1]
name = value

[section2]
foo = foo from one.cfg
baz = baz from two.cfg

Example use of multi-line notation to include more than one file:

[DEFAULT]
extends = one.cfg
          two.cfg

When several files are provided, they are processed sequentially, following the precedence rules explained above. This means that the list of files should go from most specialized to most common.

Tools will need to provide a way to produce a merged version of the file. This will be useful to let users publish a single file.

2.2. Description of sections and fields

Each section contains a description of its options.

  • Options that are marked multi can have multiple values, one value per line.
  • Options that are marked optional can be omitted.
  • Options that are marked environ can use environment markers, as described in PEP 345.

The sections are:

global
Global options not related to one command.
metadata
Name, version and other information defined by PEP 345.
files
Modules, scripts, data, documentation and other files to include in the distribution.
extension sections
Options used to build extension modules.
command sections
Options given for specific commands, identical to those that can be given on the command line.

2.2.1. Global options

Contains global options for Packaging. This section is shared with Distutils.

commands

Defined Packaging command. A command is defined by its fully qualified name. optional, multi

Examples:

[global]
commands =
    package.setup.CustomSdistCommand
    package.setup.BdistDeb
compilers

Defined Packaging compiler. A compiler is defined by its fully qualified name. optional, multi

Example:

[global]
compilers =
    hotcompiler.SmartCCompiler
setup_hooks

Defines a list of callables to be called right after the setup.cfg file is read, before any other processing. Each value is a Python dotted name to an object, which has to be defined in a module present in the project directory alonside setup.cfg or on Python’s sys.path (see Finding hooks). The callables are executed in the order they’re found in the file; if one of them cannot be found, tools should not stop, but for example produce a warning and continue with the next line. Each callable receives the configuration as a dictionary (keys are setup.cfg sections, values are dictionaries of fields) and can make any change to it. optional, multi

Example:

[global]
setup_hooks = _setuphooks.customize_config

2.2.2. Metadata

The metadata section contains the metadata for the project as described in PEP 345. Field names are case-insensitive.

Fields:

name
Name of the project.
version
Version of the project. Must comply with PEP 386.
platform
Platform specification describing an operating system supported by the distribution which is not listed in the “Operating System” Trove classifiers (PEP 301). optional, multi
supported-platform
Binary distributions containing a PKG-INFO file will use the Supported-Platform field in their metadata to specify the OS and CPU for which the binary distribution was compiled. The semantics of the Supported-Platform field are free form. optional, multi
summary
A one-line summary of what the distribution does. (Used to be called description in Distutils1.)
description
A longer description. (Used to be called long_description in Distutils1.) A file can be provided in the description-file field. optional
keywords
A list of additional keywords to be used to assist searching for the distribution in a larger catalog. Comma or space-separated. optional
home-page
The URL for the distribution’s home page.
download-url
The URL from which this version of the distribution can be downloaded. optional
author
Author’s name. optional
author-email
Author’s e-mail. optional
maintainer
Maintainer’s name. optional
maintainer-email
Maintainer’s e-mail. optional
license
A text indicating the term of uses, when a trove classifier does not match. optional.
classifiers
Classification for the distribution, as described in PEP 301. optional, multi, environ
requires-dist
name of another packaging project required as a dependency. The format is name (version) where version is an optional version declaration, as described in PEP 345. optional, multi, environ
provides-dist
name of another packaging project contained within this distribution. Same format than requires-dist. optional, multi, environ
obsoletes-dist
name of another packaging project this version obsoletes. Same format than requires-dist. optional, multi, environ
requires-python
Specifies the Python version the distribution requires. The value is a comma-separated list of version predicates, as described in PEP 345. optional, environ
requires-externals
a dependency in the system. This field is free-form, and just a hint for downstream maintainers. optional, multi, environ
project-url
A label, followed by a browsable URL for the project. “label, url”. The label is limited to 32 signs. optional, multi

One extra field not present in PEP 345 is supported:

description-file
Path to a text file that will be used to fill the description field. Multiple values are accepted; they must be separated by whitespace. description-file and description are mutually exclusive. optional

Example:

[metadata]
name = pypi2rpm
version = 0.1
author = Tarek Ziadé
author-email = tarek@ziade.org
summary = Script that transforms an sdist archive into a RPM package
description-file = README
home-page = http://bitbucket.org/tarek/pypi2rpm/wiki/Home
project-url:
    Repository, http://bitbucket.org/tarek/pypi2rpm/
    RSS feed, https://bitbucket.org/tarek/pypi2rpm/rss
classifier =
    Development Status :: 3 - Alpha
    License :: OSI Approved :: Mozilla Public License 1.1 (MPL 1.1)

You should not give any explicit value for metadata-version: it will be guessed from the fields present in the file.

2.2.3. Files

This section describes the files included in the project.

packages_root
the root directory containing all packages and modules (default: current directory, i.e. the project’s top-level directory where setup.cfg lives). optional
packages
a list of packages the project includes optional, multi
modules
a list of packages the project includes optional, multi
scripts
a list of scripts the project includes optional, multi
extra_files
a list of patterns for additional files to include in source distributions (see Specifying the files to distribute) optional, multi

Example:

[files]
packages_root = src
packages =
    pypi2rpm
    pypi2rpm.command

scripts =
    pypi2rpm/pypi2rpm.py

extra_files =
    setup.py
    README

Note

The setup.cfg configuration file is included by default. Contrary to Distutils, README (or README.txt) and setup.py are not included by default.

2.2.3.1. Resources

This section describes the files used by the project which must not be installed in the same place that python modules or libraries, they are called resources. They are for example documentation files, script files, databases, etc...

For declaring resources, you must use this notation:

source = destination

Data-files are declared in the resources field in the file section, for example:

[files]
resources =
    source1 = destination1
    source2 = destination2

The source part of the declaration are relative paths of resources files (using unix path separator /). For example, if you’ve this source tree:

foo/
   doc/
      doc.man
   scripts/
      foo.sh

Your setup.cfg will look like:

[files]
resources =
    doc/doc.man = destination_doc
    scripts/foo.sh = destination_scripts

The final paths where files will be placed are composed by : source + destination. In the previous example, doc/doc.man will be placed in destination_doc/doc/doc.man and scripts/foo.sh will be placed in destination_scripts/scripts/foo.sh. (If you want more control on the final path, take a look at Defining a base prefix).

The destination part of resources declaration are paths with categories. Indeed, it’s generally a bad idea to give absolute path as it will be cross incompatible. So, you must use resources categories in your destination declaration. Categories will be replaced by their real path at the installation time. Using categories is all benefit, your declaration will be simpler, cross platform and it will allow packager to place resources files where they want without breaking your code.

Categories can be specified by using this syntax:

{category}

Default categories are:

  • config
  • appdata
  • appdata.arch
  • appdata.persistent
  • appdata.disposable
  • help
  • icon
  • scripts
  • doc
  • info
  • man

A special category also exists {distribution.name} that will be replaced by the name of the distribution, but as most of the defaults categories use them, so it’s not necessary to add {distribution.name} into your destination.

If you use categories in your declarations, and you are encouraged to do, final path will be:

source + destination_expanded

For example, if you have this setup.cfg:

[metadata]
name = foo

[files]
resources =
    doc/doc.man = {doc}

And if {doc} is replaced by {datadir}/doc/{distribution.name}, final path will be:

{datadir}/doc/foo/doc/doc.man

Where {datafir} category will be platform-dependent.

2.2.3.1.1. More control on source part
2.2.3.1.1.1. Glob syntax

When you declare source file, you can use a glob-like syntax to match multiples file, for example:

scripts/* = {script}

Will match all the files in the scripts directory and placed them in the script category.

Glob tokens are:

  • *: match all files.
  • ?: match any character.
  • **: match any level of tree recursion (even 0).
  • {}: will match any part separated by comma (example: {sh,bat}).
2.2.3.1.1.2. Order of declaration

The order of declaration is important if one file match multiple rules. The last rules matched by file is used, this is useful if you have this source tree:

foo/
   doc/
      index.rst
      setup.rst
      documentation.txt
      doc.tex
      README

And you want all the files in the doc directory to be placed in {doc} category, but README must be placed in {help} category, instead of listing all the files one by one, you can declare them in this way:

[files]
resources =
    doc/* = {doc}
    doc/README = {help}
2.2.3.1.1.3. Exclude

You can exclude some files of resources declaration by giving no destination, it can be useful if you have a non-resources file in the same directory of resources files:

foo/
   doc/
      RELEASES
      doc.tex
      documentation.txt
      docu.rst

Your files section will be:

[files]
resources =
    doc/* = {doc}
    doc/RELEASES =
2.2.3.1.2. More control on destination part
2.2.3.1.2.1. Defining a base prefix

When you define your resources, you can have more control of how the final path is computed.

By default, the final path is:

destination + source

This can generate long paths, for example (example_final_path):

{datadir}/doc/foo/doc/doc.man

When you declare your source, you can use whitespace to split the source in prefix suffix. So, for example, if you have this source:

docs/ doc.man

The prefix is “docs/” and the suffix is “doc.html”.

Note

Separator can be placed after a path separator or replace it. So these two sources are equivalent:

docs/ doc.man
docs doc.man

Note

Glob syntax is working the same way with standard source and split source. So these rules:

docs/*
docs/ *
docs *

Will match all the files in the docs directory.

When you use split source, the final path is computed this way:

destination + prefix

So for example, if you have this setup.cfg:

[metadata]
name = foo

[files]
resources =
    doc/ doc.man = {doc}

And if {doc} is replaced by {datadir}/doc/{distribution.name}, final path will be:

{datadir}/doc/foo/doc.man
2.2.3.1.3. Overwriting paths for categories

This part is intended for system administrators or downstream OS packagers.

The real paths of categories are registered in the sysconfig.cfg file installed in your python installation. This file uses an ini format too. The content of the file is organized into several sections:

  • globals: Standard categories’s paths.
  • posix_prefix: Standard paths for categories and installation paths for posix system.
  • other ones XXX

Standard categories paths are platform independent, they generally refers to other categories, which are platform dependent. sysconfig will choose these category from sections matching os.name. For example:

doc = {datadir}/doc/{distribution.name}

It refers to datadir category, which can be different between platforms. In posix system, it may be:

datadir = /usr/share

So the final path will be:

doc = /usr/share/doc/{distribution.name}

The platform-dependent categories are:

  • confdir
  • datadir
  • libdir
  • base
2.2.3.1.4. Defining extra categories
2.2.3.1.5. Examples

These examples are incremental but work unitarily.

2.2.3.1.5.1. Resources in root dir

Source tree:

babar-1.0/
   README
   babar.sh
   launch.sh
   babar.py

setup.cfg:

[files]
resources =
    README = {doc}
    *.sh = {scripts}

So babar.sh and launch.sh will be placed in {scripts} directory.

Now let’s move all the scripts into a scripts directory.

2.2.3.1.5.2. Resources in sub-directory

Source tree:

babar-1.1/
   README
   scripts/
      babar.sh
      launch.sh
      LAUNCH
   babar.py

setup.cfg:

[files]
resources =
    README = {doc}
    scripts/ LAUNCH = {doc}
    scripts/ *.sh = {scripts}

It’s important to use the separator after scripts/ to install all the shell scripts into {scripts} instead of {scripts}/scripts.

Now let’s add some docs.

2.2.3.1.5.3. Resources in multiple sub-directories

Source tree:

babar-1.2/
   README
   scripts/
      babar.sh
      launch.sh
      LAUNCH
   docs/
      api
      man
   babar.py

setup.cfg:

[files]
resources =
     README = {doc}
     scripts/ LAUNCH = {doc}
     scripts/ *.sh = {scripts}
     doc/ * = {doc}
     doc/ man = {man}

You want to place all the file in the docs script into {doc} category, instead of man, which must be placed into {man} category, we will use the order of declaration of globs to choose the destination, the last glob that match the file is used.

Now let’s add some scripts for windows users.

2.2.3.1.5.4. Complete example

Source tree:

babar-1.3/
   README
   doc/
      api
      man
   scripts/
      babar.sh
      launch.sh
      babar.bat
      launch.bat
      LAUNCH

setup.cfg:

[files]
resources =
    README = {doc}
    scripts/ LAUNCH = {doc}
    scripts/ *.{sh,bat} = {scripts}
    doc/ * = {doc}
    doc/ man = {man}

We use brace expansion syntax to place all the shell and batch scripts into {scripts} category.

2.2.4. Extension modules sections

If a project includes extension modules written in C or C++, each one of them needs to have its options defined in a dedicated section. Here’s an example:

[files]
packages = coconut

[extension: coconut._fastcoconut]
language = cxx
sources = cxx_src/cononut_utils.cxx
          cxx_src/python_module.cxx
include_dirs = /usr/include/gecode
               /usr/include/blitz
extra_compile_args =
    -fPIC -O2
    -DGECODE_VERSION=$(./gecode_version) -- sys.platform != 'win32'
    /DGECODE_VERSION=win32 -- sys.platform == 'win32'

The section name must start with extension:; the right-hand part is used as the full name (including a parent package, if any) of the extension. Whitespace around the extension name is allowed. If the extension module is not standalone (e.g. _bisect) but part of a package (e.g. thing._speedups), the parent package must be listed in the packages field. Valid fields and their values are listed in the documentation of the packaging.compiler.extension.Extension class; values documented as Python lists translate to multi-line values in the configuration file. In addition, multi-line values accept environment markers on each line, after a --.

2.2.5. Commands sections

To pass options to commands without having to type them on the command line for each invocation, you can write them in the setup.cfg file, in a section named after the command. Example:

[sdist]
# special function to add custom files
manifest-builders = package.setup.list_extra_files

[build]
use-2to3 = True

[build_ext]
inplace = on

[check]
strict = on
all = on

Option values given in the configuration file can be overriden on the command line. See Writing the Setup Configuration File for more information.

These sections are also used to define command hooks.

2.3. Extensibility

Every section can have fields that are not part of this specification. They are called extensions.

An extension field starts with X-. Example:

[metadata]
name = Distribute
X-Debian-Name = python-distribute

2.4. Changes in the specification

The versioning scheme for this specification is MAJOR.MINOR. Changes in the specification will cause the version number to be updated.

Changes to the minor number reflect backwards-compatible changes:

  • New fields and sections (optional or mandatory) can be added.
  • Optional fields can be removed.

The major number will be incremented for backwards-incompatible changes:

  • Mandatory fields or sections are removed.
  • Fields change their meaning.

As a consequence, a tool written to consume 1.5 has these properties:

  • Can read 1.1, 1.2 and all versions < 1.5, since the tool knows what optional fields weren’t there.
  • Can also read 1.6 and other 1.x versions: The tool will just ignore fields it doesn’t know about, even if they are mandatory in the new version. If optional fields were removed, the tool will just consider them absent.
  • Cannot read 2.x and should refuse to interpret such files.

A tool written to produce 1.x should have these properties:

  • Writes all mandatory fields.
  • May write optional fields.

2.5. Acknowledgments

This specification includes work and feedback from these people:

  • Tarek Ziadé
  • Julien Jehannet
  • Boris Feld
  • Éric Araujo

(If your name is missing, please let us know.)