Im puzzled why Houdini uses point normals instead of vertex ones on polygonal models… For my opinion vertex would fit it much more elegant, because you wouldn't need to have many points in one place just in order to have sharp edge there…
If they would be vertex it would fit ideology of all the other attributes in Houdini, and it will make at least one thing easier to learn for novices.

Any thoughts?
As for me, I think it may be implemented, i'm not saying easily, but without breaking all the old stuff, i.e leaving point normals but introducing also vertex ones..
What do you think, Houdini people? Is this idea worth the proposal?

I agree with you that splitting points is an awkward approach compared to a vertex-based normal position crease.

I would like to see a solution for imported geometry with hard edges so Subdivision SOP and OpenGL display takes it into account. However not all applications are using hard edge information so cusping edges can be good in some cases. Creasing points are not working as expected either since the faces remaines smooth and not faceted, faceted with single point that is. Along with this should also be a procedural way of ordering edges. As it is now I have not found an easy way of creating a procedural network with PolyBevel since edge information is static and breaks at any change upstreams.

And as you say, keeping the legacy approach is good but introducing a vertex based is faster and gives a greater degree of ways to sharpen edges.

I agree. We've had an RFE in for vertex normal support for quite a while now. It would definitely make a lot of geometry interchange very much easier if this happened.

Here's hoping,
Jason

I completely agree on the RFE for vertex normals.

I've never been a fan of this concept. Unless of course, i'm misunderstanding something quite fundumental…

The sometimes-nasty side effect of this, which is very apparent in maya-land is the fact that every piece of geometry HAS to carry a UV set. Mostly incosequential on smaller meshes, but could translate to huge amounts of extranous data on larger data sets. If any of you ever had to do battle with the infamous map1 vs. uvset1 nonsense, you'll know what i'm talking about.

9 times out of 10, we endup recooking the normals after the fact anyway, which means that the imported normals would get munged up.

*shrug* Just my $0.02 CDN…

Cheers,

G