A few days ago the project that I worked on for sometime finally went public: PySide was launched!
And “what is PySide?”, you ask me. It is a binding of the Qt4 framework for the Python language created by INdT and Nokia (Moi, tamperelaiset!) under a LGPL license. My team (not that I’m the owner, “mine” in the sense of “our not including the listener” — I think there is a plural for that in Finnish or Quenya) worked as mad and the launching was smooth: we have a positive reception from the community and the news are popping up everywhere.
And we give not only the fish but also the fishing rod: the binding generator was also made available. But what is the importance of this? Since the news was given, the explanation follows.
The main motivation to start the PySide project was to provide Python bindings of the Qt4 library under a LGPL license, to follow the very own Qt4 LGPL license offer from Nokia. Many possibilities on how this should be done where analysed, and before this sentence start an endless technical discussion I must say that all the options have strengths and weaknesses, but aren’t that much different. In the end it was a mix of right-tool-for-the-job, team’s acquaintance with the technologies involved and personal taste. Just to mention a personal favorite, I really appreciate Smoke. In the end we choose to alter an existing binding generator (more on this later) and use Boost.Python as the middle-man to talk to CPython API. To express this with colored boxes this is PySide right now:
Writing bindings for a library so massively huge as Qt is a task… no, it’s not a task, it is a punishment. Beign reasonable people as we are, a previous research has been made and we opt to adapt the code from QtScript Generator,which is a fork of the Qt Jambi’s Generator, and both are binding generators for QtScript and Java, respectively. They’re developed by Trolltech (when it was called Trolltech).
The binding generation scheme works like this:
The global.h file includes all headers (or at least the desired ones) from the library beign wrapped, and is also used to define (and undefine) preprocessor macros. typesystem.xml files contains descriptions of how the library should be exported to the target language and other semantic information: rejeted classes, renames methods, types to be converted, which methods move object ownership from Python to C++ and the otherway around, and specifies where handwritten code for special cases should be inserted. If there is no need for changes, this XML will be a simple list of classes, enums and functions.
Notice that we not only forked the QtScript Generator, we also converted it from an monolithic application to a library + front-end generator scheme. And here is another picture to show the idea:
Theoretically the projects from where we derived code could be changed to use the API Extractor and share this code base (and bugs, and fixes, and improvements). Nevertheless this will not be that straightforward, since we have made a number of changes to the typesystem format, from simple things like tag renaming (mostly changing “java” to “targetlang”) to more complex things that I’ll not exemplify here. Besides that, one could write front-ends that use API Extractor to generate things other than bindings: class diagrams with GraphViz, statistics, something-that-i-did-not-thought-about.
Now for the not so beauty part. In a perfect world the C++ to Python binding generator could generate bindings for any C++ library, although, in the way it is right now the generator is useful only for Qt based libraries. Shame on us! Anyhow, taking into account that our main target was to create Qt bindings and that a beta version of them was released, I suggest you forgive us. :) Of course that it is in our plans to solve this and make the generator a useful tool for as many people as possible.
To much information for now! For the time beign download, test, report bugs and enjoy. And, if you’re feeling social, go to #pyside channel on FreeNode and subscribe to the mailing list. Everybody is welcome.