I've sent the mail with the changes requested (on this and patch 3), except for:
hr = IDirectDrawSurface_AddAttachedSurface(surface, ds);
IDirectDrawSurface_Release(ds);if (FAILED(hr))return hr;- }
You've overshot my suggestion for pre-attached depth stencils a bit. Now you no longer detach application-created depth stencils any more in case CreateDevice() fails, but you also don't detach a depth stencil that you attached.
This may even be the correct behavior. If you want you can write a test (e.g. try to pass a color surface without DDSCAPS_3DDEVICE, but *with* a depth stencil attached. See if it is detached. Try to pass a color surface without DDSCAPS_3DDEVICE and see if a depth stencil is attached after the failing CreateDevice call), but I don't think it's worth spending much time on this.
My suggestion is to detach if creating the IM device fails, but only if create_z_surface is TRUE.
As discussed, when CreateDeviceFromSurface returns an error (say, on passing a surface without the 3DDEVICE flag) it'll keep the depth surface it created internally attached, without detaching it from the application-created surface. This bug is apparently not mimicked in Wine's ddraw implementation of AddAttachedSurface, where it will return an error if you attach a z-buffer to a surface without the 3DDEVICE flag. This behavior hasn't been tested on windows yet. We'll probably need to fix this in ddraw and also extend the tests accordingly.
Cheers, Jam