http://bugs.winehq.org/show_bug.cgi?id=19002
--- Comment #12 from Stefan Dösinger stefandoesinger@gmx.at 2009-06-21 06:01:08 --- Ok, I'll send a patch.
The added negation is correct by the way. device->vs_clipping is true if the GL vertex shader implementation supports clipping, for example with GLSL, or ARB+NVvp2. In that case the vertex shader doesn't set the clipping texcoord, and random vertices are clipped. I don't know why I forgot to negate this in the first place, or why I didn't notice the problem. I did test GLSL, but maybe I was just "lucky" with my random fragment.texcoord contents.
A similar consideration applies to the fixed function pipeline. The KIL should not be used in this case because (a) clipping works the normal way anyway, and (b) the texcoord is not initialized.