I tried out eXpl0it3r's test set up and it caused the same problem for me (just installed catalyst 14.4).
So having CodeXL at my disposal, I fired it up and checked the texture that was being rendered to. It was being cleared as expected... So the problem clearly wasn't that the RenderTexture wasn't clearing properly... it was something else.
I then went ahead and checked the default front- and backbuffers of the RenderWindow, and what do you know... they weren't being cleared, so it's really not the RenderTexture's fault.
Knowing that GL can get quite sensitive in certain situations, I just tried stuff out, and it seems that a resetGLStates() before clearing the RenderWindow fixed it. Then I went ahead and checked which call inside resetGLStates() would fix the problem. It seems that the RenderWindow won't clear properly if the texture that is used as the color attachment of the RenderTexture isn't unbound. A simple setTexture(NULL) inside the clear() method already fixes the problem. I don't know how the driver chose to do things, but it might be trying to access the texture in the RenderWindow's context while it was being written to in the RenderTexture's context, so unbinding it seems to have solved the problem.
I was going to fix it, but I didn't know where to and if calling setTexture(NULL) is the best solution for this problem. It has to be done somewhere, but it could decrease performance by incurring an extra bind per frame (if done while clearing) in the rare case that the user's code only uses a single texture
. This is amortized when more textures are bound every frame. I can push a fix for this if Laurent agrees to unbind in clear.
