My takeaway, and @RafaelPalomar can correct me if I’m wrong, is that Rafeal succeeded in compiling the SOFA python bindings in a Slicer superbuild in a way that supports many integration strategies:
Integrate SOFA in Slicer event loop by periodically updating the physics on a model constructed by calling SOFA primitives based on content of the scene (e.g. a mesh).
Use SOFA in a thread in Sicer’s process, being cafefull to isolate events.
Use SOFA in a separate process of PythonSlicer, e.g. like ParallelProcessing, which allows, for example, launching the process on a different computer than Slicer is running on. This was technically possible before, but is easier with the superbuild version.