[Robotics] Extension of the sensor/controller/actuator model in the GE

[Benoit Bolsee]

>> it shows that blender
>> can load DLL, call functions from it, display custom UI, and save 
>> custom data in .blend. Adpating this model to the GE is a complex 
>> project but is possible. The complexity comes partly from that fact 
>> that the GE is
>> C++, and DLL are not good friends of C++.
> 
> Why not? Calling a function from C++ whose implementation 
> resides in the DLL is not that unusual...?

Calling a C function in a DLL is indeed no problem but logic bricks are
C++ classes. To build a new brick you must derive one of the base
sensor/controller/actuator class, which cannot be done in a DLL as the
binary compatibility with the base class cannot be ensured if blender GE
is modified. The only way would be to put the DLL in trunk and compile
it all the time with blender, which goes against the concept of plugin.

The only robust model is to export a C API that allows a "plugin" class
in the GE to export its processing to a DLL and vice versa to allow the
DLL to access base class and scene data. This is where the complexity
lies. A 
short route would be to base this API on the already existing python
API. 

>> Some considerations limit the scope of this plugin model:
>> - The plugin has limited access to the scene data: it cannot directly

>> access the C++ classes, it must go through a C API.
>> - For the same reason, a sensor plugin cannot be triggered by game 
>> internal data
>> - The current logic model may be superseeded by a nodal model in the 
>> future, it may not be the right time to develop a complex plugin 
>> model.
> 
> This latter remark seems to be the most important one! The 
> positive interpretation that I derive from it is that we 
> should try and discuss that "future nodal model" with the 
> responsible developer(s) in order to make our aims clear, and 
> maybe to help realise the whole thing...
> 

I'll make sure you get involved when this project is discussed. Should
happen in the next few weeks.

>> However, a plugin system would be perfect to implement custom 
>> communication protocol, which is an essential need for robotic 
>> applications. Given this, I would still find justified that a project

>> is funded to implement this plugin model.
> 
> If the description of such a project can me made, and the 
> appropriate developers are identified, there might be some 
> money available in some of the projects whose partners are 
> already active in this forum :-)
> 

It would have been perfect for a GSoC project! But slots have been
allocated already. 
It's a project of medium complexity: the theory is well known, the
bricks are in place. 
Unfortunately I don't have the time to prepare a plan but I ask my
co-dev on the BGE.