The following are my notes, observations, and resources on the subject of working with Python 3 (with or without Django).
Recently I've become involved in a couple of Django efforts that used Python 3.3. The quick summary of what I learned is pretty much what I expected: Out of the box Django 1.5 (and the pending 1.6 release) works fine with Python 3.3.2.
# Use Python 3.3.2!
Myself and others have encountered problems with using Django 1.5+ and
earlier versions of Python 3. The issues can be tricky; for example
syncdb fails in curious ways on Python 3.3.0.
The answer, for me, is to use Python 3.3.2 and don't look back.
# Checking for Python 3 Compatibility
The steps I use are below. They are in rough order:
- Look up the package on PyPI and see if any of it's trove classifiers mention Python 3 status.
- See if a pull request for Python 3 support is outstanding.
- Run the test suite using Python 3.3
- Use 2to3 to scan the code for issues.
- If a Django project, check the models for
__str__()methods. If it has them, it's a pretty good indicator it's Python 3.3 friendly.
- Make a judgement call.
# Important Packages that work with Python 3
In this section I'm listing a few of the Python and Django packages I'm using that worked without me having to do anything sort of modification or pull request:
- Django 1.5 and 1.6 beta
- Pillow (drop-in replacement for PIL)
- django-bootstrap-registration (templates for django-registration)
# Conversion process
How I convert Python 2 code to Python 3:
- Use 2to3 until you get used to not using it.
- Fix any problems you find in the code. Try to keep solutions as simple as possible.
- Submit the pull request.
- Politely poke the package owner to accept the pull request.
- Once the owner accepts the pull request, gently poke the package owner to push the update to PyPI.
# Packages that needed conversion
Here are four packages worth noting that had to be converted:
- django-stripe-payments (in progress)
Let's get into some detail for each package:
This is a handy, more unicode friendly replacement for Django's django.utils.text.slugify function. It failed on Django 1.6 beta, so I forked it, submitted a successful pull request after testing it on Python 2.6, 2.7, and Python 3.3.2. The Mozilla team pushed it to PyPI and even gave me badges for my efforts!
In the future I would like to see this little package work without the dependency of Django itself, and I've had a couple replacement dependencies suggested.
With the 1.0 release, it's been updated for Django 1.5, unless you use customized User models or Python 3. I really needed this on PyPI, but the maintainer is very busy. Therefore, I forked the project, renamed it to django-reg while referencing the original, and pushed it to PyPI. Not ideal, but sometimes you have to do what you have to do.
Lesson learned: In the future skip these issues and just use django-allauth.
Update 08/08/2013: All my Python 3 work is now on django-allauth. This way I don't have to do the work of maintaining a compatibility branch.
# django-stripe-payments (in progress)
While django-zebra is better known for handling stripe payments, I prefer to use Eldarion's excellent, well-maintained and tested django-stripe-payments. It was delayed because django-nose wasn't Python 3 compatible, but that's been fixed as of at least 7/28/2013. Unfortunately, it's Travis-CI tests fail with the Django 1.5 era custom user model because of what might be a linter misconfiguration. In theory, the just released version of pylint (1.0) should work without any problems.
Of course, I informed the maintainers of my efforts and problems.
Update 07/30/2013: Updated the text to reflect the current status of the linter problems I'm having.
Update 08/08/2013: While django-stripe-payments is useful, it doesn't do everything that I need. I tried extending it's logic into a third app, but that was proving unweildy. Therefore, I forked the project to dj-stripe, which has the features and compatibility that I need.
Today, I asked the maintainer to visit an outstanding pull request that adds Python 3 compatibility. It's a good idea to do this since they might already be working on it, or may have observations they want to share.
Update 7/28/2013: Jeff Balogh pushed a Python 3.3.2+ compatible version up to PyPI a few days ago. Makes everything so much better!
# Dealing with Slow Maintainers
For what it's worth, from experience ranting to or about slow-moving maintainers is absolutely counter-productive. People have lives and jobs that sometimes get in the way of open source. It's more productive to be patient, be polite, and if necessary do an absolutely minimal fork or find a working alternative.
# Handy Resources
The following are two useful resources on converting Python 2 to Python 3. I don't follow their methods exactly, I just use them as rough guidelines for my own workflow.
- Porting Django apps to Python 3 video by Jacob Kaplan-Moss
- Porting to Python 3 book (free HTML or paid PDF, Kindle, ePub) by Lennart Regebro
For various Python 3 recipes, I keep my kindle reader open to this priceless gem:
- Python Cookbook, 3rd Edition book by David Beazley and Brian K. JOnes.
# Closing Thoughts
This blog post makes it seems like I've put in a lot of work getting things to Python 3, but that isn't the case at all. In fact, for the most part the transition has been trivial. There are a few minor hiccups, but nothing that's killing a whole day or destroying a deadline.
Benefit of using Python 3? I get to call myself a Python 3 Hipster.
Downside of using Python 3? A little bit of extra homework on each project. However, each time I fix something it's done and I've made the world a better place.