Packaging namespace packages#

Namespace packages allow you to split the sub-packages and modules within a single package across multiple, separate distribution packages (referred to as distributions in this document to avoid ambiguity). For example, if you have the following package structure:

mynamespace/
    __init__.py
    subpackage_a/
        __init__.py
        ...
    subpackage_b/
        __init__.py
        ...
    module_b.py
pyproject.toml

And you use this package in your code like so:

from mynamespace import subpackage_a
from mynamespace import subpackage_b

Then you can break these sub-packages into two separate distributions:

mynamespace-subpackage-a/
    pyproject.toml
    src/
        mynamespace/
            subpackage_a/
                __init__.py

mynamespace-subpackage-b/
    pyproject.toml
    src/
        mynamespace/
            subpackage_b/
                __init__.py
            module_b.py

Each sub-package can now be separately installed, used, and versioned.

Namespace packages can be useful for a large collection of loosely-related packages (such as a large corpus of client libraries for multiple products from a single company). However, namespace packages come with several caveats and are not appropriate in all cases. A simple alternative is to use a prefix on all of your distributions such as import mynamespace_subpackage_a (you could even use import mynamespace_subpackage_a as subpackage_a to keep the import object short).

Creating a namespace package#

There are currently two different approaches to creating namespace packages, from which the latter is discouraged:

  1. Use native namespace packages. This type of namespace package is defined in PEP 420 and is available in Python 3.3 and later. This is recommended if packages in your namespace only ever need to support Python 3 and installation via pip.

  2. Use legacy namespace packages. This comprises pkgutil-style namespace packages and pkg_resources-style namespace packages.

Native namespace packages#

Python 3.3 added implicit namespace packages from PEP 420. All that is required to create a native namespace package is that you just omit __init__.py from the namespace package directory. An example file structure (following src-layout):

mynamespace-subpackage-a/
    pyproject.toml # AND/OR setup.py, setup.cfg
    src/
        mynamespace/ # namespace package
            # No __init__.py here.
            subpackage_a/
                # Regular import packages have an __init__.py.
                __init__.py
                module.py

It is extremely important that every distribution that uses the namespace package omits the __init__.py or uses a pkgutil-style __init__.py. If any distribution does not, it will cause the namespace logic to fail and the other sub-packages will not be importable.

The src-layout directory structure allows automatic discovery of packages by most build backends. See src layout vs flat layout for more information. If however you want to manage exclusions or inclusions of packages yourself, this is possible to be configured in the top-level pyproject.toml:

[build-system]
...

[tool.setuptools.packages.find]
where = ["src/"]
include = ["mynamespace.subpackage_a"]

[project]
name = "mynamespace-subpackage-a"
...

The same can be accomplished with a setup.cfg:

[options]
package_dir =
    =src
packages = find_namespace:

[options.packages.find]
where = src

Or setup.py:

from setuptools import setup, find_namespace_packages

setup(
    name='mynamespace-subpackage-a',
    ...
    packages=find_namespace_packages(where='src/', include=['mynamespace.subpackage_a']),
    package_dir={'': 'src'},
)

Setuptools will search the directory structure for implicit namespace packages by default.

A complete working example of two native namespace packages can be found in the native namespace package example project.

Note

Because native and pkgutil-style namespace packages are largely compatible, you can use native namespace packages in the distributions that only support Python 3 and pkgutil-style namespace packages in the distributions that need to support Python 2 and 3.

Legacy namespace packages#

These two methods, that were used to create namespace packages prior to PEP 420, are now considered to be obsolete and should not be used unless you need compatibility with packages already using this method. Also, pkg_resources has been deprecated.

To migrate an existing package, all packages sharing the namespace must be migrated simultaneously.

Warning

While native namespace packages and pkgutil-style namespace packages are largely compatible, pkg_resources-style namespace packages are not compatible with the other methods. It’s inadvisable to use different methods in different distributions that provide packages to the same namespace.

pkgutil-style namespace packages#

Python 2.3 introduced the pkgutil module and the pkgutil.extend_path() function. This can be used to declare namespace packages that need to be compatible with both Python 2.3+ and Python 3. This is the recommended approach for the highest level of compatibility.

To create a pkgutil-style namespace package, you need to provide an __init__.py file for the namespace package:

mynamespace-subpackage-a/
    src/
        pyproject.toml # AND/OR setup.cfg, setup.py
        mynamespace/
            __init__.py  # Namespace package __init__.py
            subpackage_a/
                __init__.py  # Regular package __init__.py
                module.py

The __init__.py file for the namespace package needs to contain the following:

__path__ = __import__('pkgutil').extend_path(__path__, __name__)

Every distribution that uses the namespace package must include such an __init__.py. If any distribution does not, it will cause the namespace logic to fail and the other sub-packages will not be importable. Any additional code in __init__.py will be inaccessible.

A complete working example of two pkgutil-style namespace packages can be found in the pkgutil namespace example project.

pkg_resources-style namespace packages#

Setuptools provides the pkg_resources.declare_namespace function and the namespace_packages argument to setup(). Together these can be used to declare namespace packages. While this approach is no longer recommended, it is widely present in most existing namespace packages. If you are creating a new distribution within an existing namespace package that uses this method then it’s recommended to continue using this as the different methods are not cross-compatible and it’s not advisable to try to migrate an existing package.

To create a pkg_resources-style namespace package, you need to provide an __init__.py file for the namespace package:

mynamespace-subpackage-a/
    src/
        pyproject.toml # AND/OR setup.cfg, setup.py
        mynamespace/
            __init__.py  # Namespace package __init__.py
            subpackage_a/
                __init__.py  # Regular package __init__.py
                module.py

The __init__.py file for the namespace package needs to contain the following:

__import__('pkg_resources').declare_namespace(__name__)

Every distribution that uses the namespace package must include such an __init__.py. If any distribution does not, it will cause the namespace logic to fail and the other sub-packages will not be importable. Any additional code in __init__.py will be inaccessible.

Note

Some older recommendations advise the following in the namespace package __init__.py:

try:
    __import__('pkg_resources').declare_namespace(__name__)
except ImportError:
    __path__ = __import__('pkgutil').extend_path(__path__, __name__)

The idea behind this was that in the rare case that setuptools isn’t available packages would fall-back to the pkgutil-style packages. This isn’t advisable because pkgutil and pkg_resources-style namespace packages are not cross-compatible. If the presence of setuptools is a concern then the package should just explicitly depend on setuptools via install_requires.

Finally, every distribution must provide the namespace_packages argument to setup() in setup.py. For example:

from setuptools import find_packages, setup

setup(
    name='mynamespace-subpackage-a',
    ...
    packages=find_packages()
    namespace_packages=['mynamespace']
)

A complete working example of two pkg_resources-style namespace packages can be found in the pkg_resources namespace example project.