Weird Bumpmap Bug

[selden]

Mike,

My posting was to let Chris know which of the five "render paths" has the bug and that it's easy to reproduce on a recent model Nvidia card. The problem affects more than just the older Nvidia cards.

Which of the Control-V render paths works for you?

[selden]

Unfortunately, close examination reveals that the bumpmap code in Celestia v1.3.0 is broken not just in the "Nvidia Combiners" but also in the "OpenGL vertex shaders" rendering path. Superficially it looks OK, but when you watch the western side of Olympus Mons during sunrise, the western escarpment is lit when it should be in shadow.

This is using the medres mars surface texture and bumpmap as shipped with Celestia. As best I can tell, those images are identical to the ones that shipped with v1.2.5.

Here's an example:


Here's the corresponding Cel:// URL
cel://SyncOrbit/Sol:Mars/2003-04-23T20:19:23.47212?x=mISsNileBPG4DA&y=DrdtY+e30qL//////////w&z=jEx0sgA9heoW&ow=-0.632684&ox=-0.062483&oy=0.765674&oz=-0.097722&select=Sol:Mars&fov=45.000000&ts=0.000000&rf=515&lm=2

System configuration:
256MB, 500MHz P3, Windows XP Pro SP1
128MB Nvidia GeForce4 Ti4200, Detonator 40.72 and 43.45, OpenGL 1.4
Celestia v1.3.0

This bug is *not* present in Celestia v1.3.0pre4, pre5.1 or pre6. (I don't have a copy of the 6.1 executable.)

(written later:
diff claims that the bump*.vp files in the shaders directory are identical between pre6 and final, so that suggests a problem in the executable.)

[chris]

I screwed this up in a last minute attempt to make 'standard' normal maps work. The problem with the NVIDIA combiners path is something I knew about and chose to disregard . . . The problem with the OpenGL vertex shaders path is a surprise. This is high priority, and I'll get it worked out very soon.

--Chris

[billybob884]

Which of the Control-V render paths works for you?

basic and multitexture are the only ones that work "correctly"