[DebianGIS] Re: thuban strict wxPython version check?

[Francesco Paolo Lovergine]

On Thu, Jun 08, 2006 at 05:22:33PM +0200, Bernhard Herzog wrote:
> Paul Wise <pabs3 at bonedaddy.net> writes:
> 
> > I'm wondering what the following check from Thuban/version.py is for? Is
> > the wxWidgets ABI really so fragile that wxProj breaks on new minor
> > 2.6.X versions or new 2.4.X versions? Surely using the soname to do
> > version checks with the normal shared library support is enough?
> 
> The symptom people often see when using Thuban with a different
> wxWidgets version -- even when it's only a different minor version -- is
> that it segfaults as soon as it tries to actually render a map on the
> screen.  In the past there were enough problem reports from users about
> this for us to put in the version check so that users would at least get
> a meaningful error message instead of just a segfault.
> 
> The reason for the crashes is that wxproj accesses wxPython objects at
> the C++-level to extract the corresponding wxWidgets object so that it
> can call wxWidgets methods directly.  That way, Thuban can render data
> read from e.g. a shapefile without having to funnel the data through
> Python which improves rendering speed considerably.
> 
> Unfortunately, this means that the wxWidgets object used by wxproj has
> been created by the wxWidgets library wxPython is linked with and that
> may be different from the one wxproj is linked with.  The libraries may
> differ in the layout of the classes and/or virtual tables depending on
> the version and perhaps compile time options.
> 
>   Bernhard
> 

I'm not a python addicted, but I suspect this issue should be managed
at python-wxgtk2.4 level. Incompatibilities in ABIs should be managed
using different source packages to avoid conflicts. I'm CC to d-python
list for analysis...

-- 
Francesco P. Lovergine