Working in opengl, I am currently working on color picking with glReadPixels. And i have problems deciding who is responsible for calling the pick() function and who should decide what unique values to use for each entity in the scene.

What is the case usually?

-- MORE INFO --
I have a IGraphics interface which essentially is a wrapper for the underlying API i.e. opengl. IGraphics wraps stuff like glDrawElements, SwapBuffers, ReadPixels etc.

currently I have a Renderer::Render() which gets called in the idle loop:



Should I also let Renderer do the picking? Other candidates include the Scene object, but then I will have to share the IGraphics object which is currently contained in an inner-class inside Renderer... that is bad.

It really depends on your overall architecture and how things would most naturally fit. From what you've described it sounds like Renderer makes the most sense for containing the picking logic, at least at a low level; you might consider wrapping that in a higher level interface as well for clean separation in case you want to rework your picking methodology later (e.g. if you decide to do picking via a physics library and a raycast, such as Bullet).