Created attachment 35488 [details]
screenshot

I looked into this some time ago and I sent some tests to show the issue (that's afd1fe43890026540703c9bc146fba49da316325 et al.) but I forgot to comment...

Essentially: those tests show that D3D has a specific (and consistent) behavior with fog when there is an orthogonal projection matrix set. Unfortunately, it does not match OpenGL's fixed function pipeline behavior, so in these specific circumstances we get an erroneous fog. This is exactly what causes the purple fishes here: they are actually completely fogged out.

There is no way, apparently, to change OpenGL behavior accordingly (i.e. by means of a GL extension): the only reasonable fix for this issue is to implement a shader-based fixed function replacement pipeline. So, different reason, but same 'prescription' as bug 6955. This doesn't mean that the hacks from bug 6955 are going to help here, just that the "real fix" for both those bugs requires, well, a lot of work...

Austin, could you please re-test this bug with a recent Wine version (>1.5.17).
I believe commit 10d46bc042696bb9e5c45ac88af332c957bbff03 may have helped with this.
At least the fish are colorful here, tested with Nvidia binary drivers 313.18.
You may run into bug #33055.

(In reply to comment #3)
> Austin, could you please re-test this bug with a recent Wine version (>1.5.17).
> I believe commit 10d46bc042696bb9e5c45ac88af332c957bbff03 may have helped with
> this.
> At least the fish are colorful here, tested with Nvidia binary drivers 313.18.
> You may run into bug #33055.

I'm getting bug 33055, so this is blocked.

This is fixed in wine-1.5.28-141-gf663683 / nvidia 313.18.

This is not fixed for me on Mesa radeon. Not sure why it works fine on Nvidia (I can confirm that it does).

In general I wouldn't expect any change from 10d46bc042696bb9e5c45ac88af332c957bbff03 related to this bug. If the debugging I did at the time was correct, the issue is caused by a different handling of fog coordinates between D3D and OpenGL with fixed function rendering. Essentially if there is an orthogonal projection matrix set, the fog is computed with the pre-projection z value on GL, while in D3D it is computed from the projected z. Properly fixing that requires a vertex pipeline replacement, which isn't there yet.

So, no idea why that commit actually made a difference here. Sure enough, the test I added in afd1fe43890026540703c9bc146fba49da316325 still fails, on nvidia as on radeon. I guess I can take another look, but in any case I wouldn't call this bug fixed just yet.

Copied from IRC yesterday:
Apr 24 12:25:12 <austin987>     Mystral, works for me at 715fd2a and f663683, with glsl and arb
Apr 24 12:25:18 <austin987>     on 313.18 / GTX 460

at least on my machine, it appears to have been fixed by the nvidia driver.

Reopening since it's broken for you, though I won't be able to test this going forward (well, without reverting the nvidia-driver).

I forgot to mention at the time, this was fixed by e2cf98eb76b9c8451b5f9d25b0c9c07f9fdc4e04.

(In reply to comment #8)
> I forgot to mention at the time, this was fixed by
> e2cf98eb76b9c8451b5f9d25b0c9c07f9fdc4e04.

Resolving fixed.

Closing bugs fixed in 1.7.13.