Code review comment for lp:~vanvugt/mir/full-gl

* Needs discussion *

I am not quite sure about this. The right way to achieve GL-based compositing should not be this invasive. In my mind, it should go more or less like the following :

1- Finish the work that alf started abstracting the renderer. There should not be any renderer specific code outside the renderer (e.g. no EGL/GL/GLES calls).

2- This should give us a proper (default) renderer with GLESv2. Going one step further, since GL/GLES would probably be desired by many shells potentially, we should have multiple flavors of a default set of renderers already implemented in Mir (e.g. A GL based renderer, a GLESv2 renderer, a GLESv3 renderer).

3- The shells should also be able to override that functionality with their own renderer (say vulkan based), or blitter based renderer). (I believe we already have some form of this but not sure how complete).

4- Also the renderer should be able to be chosen at run-time dynamically and replaced on the fly if needed, though this can be done at a later time (e.g. use blitter based rendering instead of GL for power reasons etc).