[Bug 11819] The Longest Journey Demo: characters rendered partially because of artifacts in the z-buffer

http://bugs.winehq.org/show_bug.cgi?id=11819

Octavian Voicu  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]
            Summary|The longest journey demo:   |The Longest Journey Demo:
                   |Depth Stencil buffer        |characters rendered
                   |locking not implemented     |partially because of
                   |                            |artifacts in the z-buffer

--- Comment #37 from Octavian Voicu  2011-11-10 13:23:32 CST ---
There are a bunch of issues here:

1. First of all, game uses only one z-buffer (no stencil buffers). It locks it
then immediately unlocks it, without ever accessing the memory (probably for
some kind of synchronization). I do have an initial implementation of
locking/unlocking, but depth/stencil surface code needs to be refactored.

Until that, we need to silence two fixmes (show only once) that pollute the
output to make the game playable:

fixme:d3d_surface:surface_load_location Unimplemented location SFLAG_INSYSMEM
for depth/stencil buffers.
fixme:d3d_surface:surface_unmap Depth / stencil buffer locking is not
implemented.

2. The black rendered characters is a lighting-related issue, which I described
in http://bugs.winehq.org/show_bug.cgi?id=29052 and also included a patch.

3. Not specifically related to this bug report, but there is another issue with
the message box asking whether you want to quit the game showing up completely
gray. I have a hack for this in http://bugs.winehq.org/show_bug.cgi?id=28678,
but there's much work to be done until a patch is ready for submission.

4. Finally, the partially rendered characters issue is indeed related to the
z-buffer, but indirectly. Game uses IDirect3DViewport2::Clear(D3DCLEAR_ZBUFFER)
to clear the z-buffer, but the rectangles it clears are wrong. It decides what
rectangles need to be cleared in the z-buffer by transforming the bounding box
of characters using IDirect3DDevice2::TransformVertices(). Wine's
implementation of this method doesn't work when
IDirect3DViewport2::SetViewport2() is used to set the viewport. I already
tested a hackish patch that solves the issue, and currently working on tests
and a proper one.

With patches for 1, 2, 3, 4 graphics are flawless, but didn't test very much
yet.

I changed the title of the bug to reflect to be more accurate after my new
findings, and will update progress of fixing TransformVertices and conversion
between D3DVIEWPORT and D3DVIEWPORT2 here.