It would have been nice if we could do all of this inside wined3d, but we need
to determine what resources are in use by d3d8/d3d9, not wined3d. This is
mostly an issue for "implicit" resources like swapchain render targets and the
depth / stencil surface. The fact that we're having this problem probably
means we're doing it wrong. Since these implicit resources should only be
referenced by wined3d, we could probably just destroy and recreate the
swapchain, instead of the little updateSurfaceDesc() dance we currently do.
D3DRS_ZBIAS is poorly defined, but it makes sense that the bias should be
format independent. Looking at application behaviour, it seems to include a
slope scale factor as well. This fixes a couple of regressions introduced by
96b758f7b3, although it was broken before as
well, just in a different way.
Henri Verbeet