A few minutes ago, my fingers were poised for a moment above the keyboard as I prepared to backport the essential match_hostname() function (without which the Secure Sockets Layer is not actually secure!) from the Python 3.2 version of the ssl Standard Library to earlier versions of Python. Suddenly, I paused: what would I call the new distribution that I created in the Package Index to hold this small function?

It seemed a shame to consume an entire top-level name in the Package Index for what is, after all, a stopgap measure until older versions of Python are one day retired.

And so I conceived this backports namespace package. It reserves a namespace beneath which we can happily place all of the various features that we want to cut-and-paste from later Python versions. I hope that this will provide two benefits:

1. It should provide greater sanity, and a bit more organization, in the Package Index.

2. When you are ready to port a Python application to a new version of Python, you can search the code for any import statements that name a backports package, and remove the backports for features that have now “arrived” in the version of Python to which you are upgrading.

I have considered calling for all backports packages to issue a warning upon import if they detect that they are running under a version of Python that has now gained the feature they offer, but I think that will be unkind to actual users, since the most widespread versions of Python today still display warnings by default.

Building your own backports module

Placing a module of your own inside of the backports namespace requires only a few simple steps. First, set your project up like::

project/pyproject.toml
project/backports/
project/backports/__init__.py   <--- OPTIONAL - see below!
project/backports/yourpkg/
project/backports/yourpkg/__init__.py
project/backports/yourpkg/foo.py
project/backports/yourpkg/bar.py

This places your own package inside of the backports namespace, so your package and its modules can be imported like this::

import backports.yourpkg
import backports.yourpkg.foo

The file backports.__init__.py is optional for projects that support only Python 3.3 or later. If omitted, the package will be a PEP 420 "native" namespace package.

For projects that support Python 3.2 or earlier, it's absolutely essential that the backports.__init__.py have the following code as its content::

# A Python "namespace package" http://www.python.org/dev/peps/pep-0382/
# This always goes inside of a namespace package's __init__.py

from pkgutil import extend_path
__path__ = extend_path(__path__, __name__)

Failing to include this code will cause the module to be treated as the canonical package for import backports, masking imports for other backports packages and causing errors.

A live example of a package that implements a pkgutil-style namespace can be downloaded from the Python Package Index:

http://pypi.python.org/pypi/backports.ssl_match_hostname/3.2a3

There are currently no working examples of a native backports namespace package. See backports#1 <https://github.com/brandon-rhodes/backports/issues/1>_ for details.

What if the feature is present?

An issue on which I am undecided is whether a backports package, if it finds itself on a modern enough version of Python, should simply import the “real” version of its feature from the Standard Library instead of offering the replacement. My guess is that this is not a good idea, because if — for some reason — an incompatibility crops up bewteen the tweaked code in a backport and the official code in the modern Standard Library, then it would be nice for developers using the backport to be faced with that breakage when they themselves try removing the backport, instead of being faced with it simply because a user tries running their program on more modern version of Python.