I have been working with and debugging the NURBS portion of SGI's 
sample implementation of GLU and propose the following changes/fixes.

- The following is a bug fix. In ArcTessellator::tessellateNonlinear, 
the initial value of j in the first for loop is incorrect. The 
original code only works if the stride is 2, which is not the case 
for homogenous coordinates.
    libnurbs/internals/arctess.cc, Line 338:
        for(i=1, j=2; i<bezierArc->order; i++, j+= bezierArc->stride)
    Change to:
        for(i=1, j=bezierArc->stride; i<bezierArc->order; i++, j+=
bezierArc->stride)

- The function monoTriangulationFun calls 
monoTriangulationRecFun with zeroes for both of the current 
indices to the inc_chain and dec_chain vertex arrays. If both of 
these containers are empty (which can happen--did for me with one 
surface), monoTriangulationRecFun will assert. I don't know 
if this indicates a flaw in the data, but to avoid the assertion, I 
added the following:
    libnurbs/nurbtess/monoTriangulation.cc, Insert at line 622 
    (before call to monoTriangulationRecFun):
        if (!(0 == inc_chain.getNumElements() && 0 == dec_chain.getNumElements()))

- Initialization of the member variable output_triangles of 
the class OpenGLCurveEvaluator was overlooked. NURBS curves 
won't render in a debug build because of this. Add the following 
initialization in the constructor:
    libnurbs/interface/glcurveval.cc, Insert at line 77:
        output_triangles = 0; //don't output triangles by default

- The following change eliminates a compiler warning (using MS VC6). 
Add the following delete operator to class PooledObj:
    libnurbs/internals/bufpool.h, Insert at line 131:
        inline void     operator delete( void *, Pool & ) { assert( 0 ); }

Another problem this library has, which I have not yet debugged (and 
may not get to), is that it does not render trimmed surfaces 
correctly when using GLU_PARAMETRIC_ERROR for the sampling method. 
The edges of the trim lines are rendered with far too few segments 
(at least with NURBS trim curves--not sure about piecewise trim 
curves).

Note: I posted the above remarks to both the comp.graphics.api.opengl newsgroup
and the OpenGL cading: advanced forum at OpenGL.org.

Thank you for your consideration.

John Shell