yade-dev team mailing list archive

[Václav Šmilauer <eudoxos@xxxxxxxx>]

> I've got an other question about boost::python and dynamic libraries 
> widely used in yade.
> I would like to expose the class SpherePadder (and maybe TetraMesh) to 
> python.
> I think that it could be possible (?) to add something like:
>
> class_<SpherePadder>()
>     .def("pad5",&SpherePadder::pad5)
>     ...
>     ;
>
> in the BOOST_PYTHON_MODULE(_utils).  But, what about the compilation 
> of SpherePadder in this case?
> Must I create a libSpherePadder.so with REGISTER_CLASS_AND_BASE?
1. You should just add python interface to SpherePadder itself, instead 
of putting that to _utils: (a) utils has already way too much things 
inside,  (b) SpherePadder depends on CGAL and yade must compile without 
CGAL as well (for legal reasons).

2. No need to use REGISTER_CLASS_AND_BASE, unless you want yade to know 
about inheritance of your class etc. That is needed for serialization, 
for instance, but in this case I don't see the point.

3. SpherePadder should be added to scons (as an optional part), since it 
will also install the python module tothe appropriate place.

For yade classes I (ab)use the serialization system to get access to 
attributes via the ['attribute'] operator, but in your case you can just 
wrap the variables inside SpherePadder by hand.

sp=SpherePadder()
sp.zmin=-10
sp.gap_max=.01
sp.pack5()

Then you need to think how to pass the result back to python. For 
pack5(), you could for example return list of (x,y,z,r) tuples or 
something similar, which could be used in turn to create spheres.

HTH, Vaclav