[PySide] PySide - Qt5 - Swig

[Fabien Castan]

*Hi,

The email « PySide is dead ? » has given rise to an interesting discussion
about the evolution of PySide. The conclusion is that there is currently
really limited contribution on the project, and shiboken seems to be a
really big and hard task to maintain.

The idea to rewrite shiboken with a new solution (like LLVM, etc.) is great
in theory but represents a really huge amount of work and doesn't seem
appropriate to the situation. So we would like to propose another solution
for the evolution of PySide.
Instead of rewriting a new binding tool, we could use an existing one and
only do the job of PySide which is to give a python binding to Qt. So only
implement the binding rules.*
*
We started some work with SWIG which is a really wide used binding tool. This
could help to get a wider set of developers on the project. All guys from
SWIG already work on the binding tool and all users of PySide which know
SWIG (or are interested to learn SWIG) could easily contribute.

To explain our history...

   - We have a big C++/Qt application with a Swig binding of the core.
   - We wanted to create new UI tools in python which need to use some of
   our C++ widgets. So we need a binding of our C++ widgets. As our core
   binding is written in Swig (and we are happy with that) we need to bind our
   widgets with the same binding tool for compatibility.
   - So we created a binding of Qt in Swig. As the rules were already
   written in PySide, we write a little tool to convert PySide rules
   (typesystem xml files) to swig files + some manually written files for
   additional things.
   - We worked with this binding during one year, but there were a lot of
   missing rules to get a full pythonic Qt binding. Tools like pyuic and pyrcc
   are also missing. Qt is really big, so there is a lot of things to do to
   adjust the API to another language. But the fact that Swig allows to put
   some code using the target language is really useful for that.
   - The conclusion was that it’s too big to be done and maintain by our
   team.
   - We switch to PySide and do some annoying work to bind our C++ widgets
   with shiboken and write some conversion rules... because some function
   arguments of the widget (binded with shiboken) are core object binded with
   Swig...


Now it works with that, even if we have limited the binded widgets to the
minimum.

But now we would like to switch to Qt5... ;)
It seems that it wont be done by the PySide team. So we think that a
solution is to do a Qt5 binding in Swig.

Our experience proves the feasibility to bind Qt with Swig. Maybe some
really specific things need fixes in Swig... But the main complexity of our
project was the mix of 2 solutions: typesystems and custom Swig rules. So
we would like to write it directly with Swig rules.

Our experience also proves, that it’s only possible to do this as an open
source project with a lot of people involved. But I’m sure, there is a such
community around Qt that are interested in an LGPL python binding that
supports all the outstanding work provided by Qt5 and in particular
QtQuick2!

Are PySide developers interested to explore this approach?
Are other people interested to explore this approach with us?

Regards,
Fabien

PS: We also recommend the use of github for the ease of forking, submitting
« pull request », reviewing, having a bug tracker linked with the code, and
for the responsiveness (gitorious is quite unusable...) and the fact that
its use is widespread. This could help to boost the project.
In all case, having duplicated versions of the project between github and
gitorious is the worst solution. It has created a lot of confusion for us.*
*
*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.qt-project.org/pipermail/pyside/attachments/20130112/c5032650/attachment.html>