[PySide] Binding C++/VTK code using Shiboken

Matthew Woehlke matthew.woehlke at kitware.com
Fri Dec 13 18:43:13 CET 2013


On 2013-12-13 12:18, Nicolas SAUBAT wrote:
> If I have a C++ class based on both Qt and VTK (like this one :
> http://camitk.imag.fr/apidocumentation/3.2/classcamitk_1_1RendererWidget.html),
> I would be able to expose the VTK part using VTK's Python binder OR Qt
> part using Shiboken, but not both.

*AFAIK* this is correct. At least, I know that I've run into the same 
problem and didn't come up with a solution.

> And then the solution seems to be to still use Shiboken,
> and inject code in the wrapped classes (containing CPython code) in
> order to tell it how to convert the Python object (PyObject) into a
> VTKObject when necessary. Which seems difficult, from the C++ side,
> regarding how VTK handles its VTKObjects.
> Please let me know if I've misunderstood some points !

No, that sounds about right.

That said, maybe someone that knows VTK better would have an idea. Have 
you asked on the VTK list?

> And, by the way, would this problem still occur if we decide not to
> expose to Python the VTK part ?

You can wrap many things in Shiboken. If what you wrap uses Qt classes 
(and you are using PySide), then you will be able to wrangle those types 
between C++ and Python (which is really a glorified way of saying you 
can wrap a method that either takes or returns a Qt type). Where you'll 
run into trouble (using Shiboken to do the wrapping) is methods that 
take or return VTK object types. (POD's are not a problem in either 
case, of course.) Basically, if the API only wraps types that are 
themselves wrapped by Shiboken, you'll be fine. (Note however that 
you'll lose base class methods if your base class is not shiboken wrapped.)

You could, conceivably, write your own external shiboken-based Python 
bindings for VTK :-). (After all, PySide is separate from / 'on top of' 
Qt; no need to be tightly integrated into the VTK repo / build system.)

Similarly if you use VTK to do the wrapping, VTK types will be okay but 
Qt types will be a problem. (Also VTK's binding system has other 
limitations, e.g. no STL classes, very poor support for enums...)

> This way, the CPython extension generated would still be able to run the
> VTK methods trought the C++ library,  but forbid to directly handle the
> VTK objects from the Python world (or crashes).

More likely Python just won't "see" methods that deal with VTK objects.

> But I don't know if it would be possible to render the 3D of VTK in some
> Qt widget ...

Hmm... I don't think QVTKWidget is a vtkObject(?); you could write your 
own wrapping for that and only that with Shiboken that would at least 
let you create the widget from Python. Assuming you even need to do that.

-- 
Matthew




More information about the PySide mailing list