Re: [gts-devel] Re: subdivision and non-manifold topologies

[Holger W. <hwa...@ya...>]

On Tue, 18 Apr 2000, Stiphane Popinet wrote:

> This can appear a little strange at first, but a lot of GTS algorithms
> are not deterministic. Indeed, all the algorithms which use
> gts_surface_foreach_(something) can show non-deterministic behaviour. It
> is due to the fact that a GtsSurface stores triangles in a hash table
> for rapid dictionnary operations. A hash table is efficient from a
> probabilistic point of view because it involves some randomness in the
> choices which are made (it's called a "randomized algorithm").

Shure. But I believed, the removed triangles are sorted by quality, so
they shouldn't be in a random order anymore ...


> You don't need to make the topology check yourself. The user of the
> function should assure himself that the preconditions are verified, but
> the algorithm should not crash if these preconditions are not verified.

ok.


> > However, I have still no idea, for what non-manifold segments should be
> > good. Do you need them in the CSG stuff ?
> 
> The initial purpose of GTS is to make physical simulations of problems
> involving interfaces: bubbles, drops, solid/liquid, solid/solid and so
> on. As you know if you have played with soap bubbles, non-manifold
> topologies are quite often encountered. Obviously, in the context of a
> solid modeler this does not make much sense.

Did you already wrote a solver ??


> > I would like much more simple winged edges (which would be much more
> > memory efficient), and don't connect triangles, which don't belong really
> > together. They can get their own edges and points. We would save 16 Byte
> > per 2-manifold segment, we need currently 36 Byte.
> >
> 
> Well, if we want non-manifold topologies, we can't do that.

ok.

- Holger