[Bf-committers] New Booleans & Smooth Meshes (maybe I need math help, maybe a shading glitch)

[Ted Schundler]

So... I was thinking about the uneven faces. Ton's right that they are
the problem. And they are always a problem in Blender it seems,
because you can't have an unevenly subdivided mesh produce correct
vertex normals. That means there is no way to get smooth results after
boolean operations without making the operations redo the entire mesh
geometry (which is exactly what this is trying to avoid). So, maybe
the problem can be fixed somewhere else. I see two possible solutions:

Vertex normal painting - vertex normals can be forced, like normal
maps meet vertex colors. There are problems with that method if the
user tries to edit anything later though...

Weighted vertex normal calcuation - there is already some code to do
this that seems to only get called when autosmoothing is on. However I
think it could be considerably better. I did some tweaking yesterday.
A write up of my experiments and reasoning is here:
http://astro.scu.edu/~ted/oss/blender/smooth/Calculating%20smooth%20vertex%20normals.pdf

I think the weighted vertex normal calculation is probably the right
way to go. I'd be interested in any feedback anyone has on it. Also,
I'd be interested in test meshes anyone has where they have had
problems with awkward not-quite-smooth shading of smooth surfaces with
nonuniform features sizes,


Ted Schundler


On Thu, 16 Dec 2004 10:50:12 -0800, Ted Schundler <tschundler at gmail.com> wrote:
> Ton:
> The first reply you got to this is incompletely and accidentally sent
> just to you.
> 
> Sorry, I just realized I should have prepared a .blend too after I
> posted, but was headed for bed. Here you go:
> http://astro.scu.edu/~ted/oss/blender/boolean/bool_curves2.blend
> There's a text file open in there explaining what's what. The tests
> are separated into different layers.
> 
> As for coding a C/C++ version, that is the plan. But for now I'm doing
> testing in Python since it's a lot easier to test out ideas. (no
> recompiling and such) Once that's working to my satisfaction, I'll
> move to C/C++.
> 
> 
> Dan:
> You are right that AutoSmooth is extremely important when it comes to
> objects that mix smooth and flat faces. Such as if the cut off part of
> the sohere was filled with flat faces. But here the back side that you
> don't see is empty, so the mesh ends where it is cut. Then
> autosmoothing makes no difference.
> 
> Ted
>