Re: [Dri-devel] NURBs + Autonormal peculiarity

[Brian P. <br...@tu...>]

Andy Isaacson wrote:
> On Thu, Jul 11, 2002 at 06:10:01PM -0400, Mike Westall wrote:
> 
>>The attached program illustrates a shading anomaly 
>>we have encountered while attempting to use autonormal
>>in conjunction with nurbs.  
>>
>>The shading of the sphere is seriously screwed up 
>>while running with indirect rendering on the standard
>>X server packaged with RH 7.3.   It can be fixed by
>>uncommenting the /* #define CORRECT 1 */ on line 8. 
> 
> 
> Your example did fail for me on RH7.1 as well -- the sphere was rendered
> in 4 "quarters" of solid shading, but a recent DRI build rendered it
> just fine.  I'll take a screen shot when I'm back at work tomorrow...
> 
> 
>>Oddly enough the EXACT problem that occurs with unaccelerated
>>rendering on XFree86 also occurs with the OpenGL version
>>that comes with recent Sun Solaris boxes.  Thus we 
>>may well have a bug that is simply latent on SGI | NV. 
> 
> 
> I ran the test on a Solaris 8 system with the OpenGL "ogl123_rt32_64"
> downloaded from Sun's OpenGL for Solaris website (Ultra 10 with Elite3D
> graphics) and it rendered correctly.

I expect that this is either an issue in the GLU nurbs code or in
Mesa's evaluator code.

The SGI SI GLU library has better NURBS support that Mesa's original
GLU library.  Make sure you're using the former (identified as version 1.3
with the GLU_EXT_nurbs_tessellator, GLU_EXT_object_space_tess extensions
by glxinfo).

Typically, if you construct a sphere out of Bezier patches two of the
corners get collapsed into one point at +/- 1 on the X,Y or Z axis.
When this happens, computing the partial derivatives of the surface
breaks down, thus producing bad normals.  There are work-arounds for
this but it can be tricky.

-Brian