By Brian Curtin
The Python world has come a long way since December 2008 when 3.0 was first released. Books have been released, blogs have been written, and most importantly, projects have been ported. Recently, NumPy and SciPy checked in their porting work. We’ve heard rumblings of Django on 3.x, possibly as early as this summer. Python 3.1.3 was released in the fall and 3.2 final is around a week away, and with 2.7 being the end of the 2.x line, all core hands are on Python 3.
Lennart Regebro knows all about this. He’s the author of a new book, Porting to Python 3, and he’s giving a talk by the same name. The idea to write the book came from a lack of published material on the topic and an interest in writing for the now defunct Python Magazine. “The lack of documentation has been the biggest hurdle, [so] if you want to port to Python 3 you have been stepping into the dark. Since I had been using Python 3 and porting to it on my free time since early 2008 I had a bit of experience to share,” says Lennart. He then took his article series and had the basis for a book which he created with reStructuredText and Sphinx.
While he agrees that the separation of string contents into binary data and Unicode text was the right move, it’s a challenge you’ll have to undertake if your application doesn’t already handle all text as Unicode. “This is where you can expect the biggest problems,” he claims. Luckily he’s taking the time to cover it in his talk. He also covers the important topic of porting strategies, including branching, continuous 2to3 conversion, and single codebase projects.
Asked about his hardest porting project, zope.testing appears to be the winner. The package used doctests from before they were included in the standard library, along with a custom testrunner module, so the first step was to separate and deprecate. “I think I ended up deleting the port and restarting two or three times either because I made a hash of it or the trunk code had changed so much that it was easier to restart than to merge the changes.” Although the port isn’t complete, “porting a package takes between a couple of hours and a couple of days, and is a lot of fun, except if you have a lot of doctests.”
The Zope Component Architecture tops his most-wanted list and is the driving force behind his efforts. “It's really cool, but uses a lot of Python internals so porting it is a challenge,” he says, mentioning that a further complication is the need of writing fixers, for which there is little documentation.” Understanding how 2to3 works internally was another challenge, which led to a chapter in his book. Along with 2to3, Benjamin Peterson’s six package has been helping Lennart along with his porting. “I was planning to write such a module myself, but now I'm glad I didn't, because Benjamin did a much better job than I would have,” he says.
Lennart is a PyCon veteran, coming to the 2008 and 2009 conferences in Chicago, along with several EuroPython events, as well as Polish and French PyCons. The evenings are some of his favorite times, “because there are so many people around you that are much smarter than you are, and are friendly and open and willing to hang with you over coffee or a drink.” Sprints were one of the highlights of his 2009 experience, where he organized a Zope sprint without hopes of a great turnout. He ended up being wrong: “we got a quite a big gathering with many of the top Zope names and had some fantastic discussions on the way forward for Zope, as well as an extremely productive sprint!” He finished the interview by saying, “that was great fun, and those type of things seem to happen a lot on PyCon.”