CGAL users discussion list

[Fernando Cacciola <>]

Hello Rob,

> Fernando,
>
>  In order to split the vertex, v1 back into v1 and v0, we need
> to call split_vertex( vy0_v1, vx0_v1 ).  However, the problem is that
> the edge collapse (not General_case_step::do, but the edge collapse
> called directly after Visitor::OnCollapsing) will alter these edges as
> they are obtained prior to the collapse.

You are right in that the real edge collapse won't be reversible with a
call to step::undo, that's why in my sketch there is a step::do method
as well. What I had in mind was that you would run the algorithm with
the PM visitor to collect the sequence of steps in decimating M0 to M1,
but then, in the context of the PM, discard M1 and decimate *from M0*
once again, but this time just executing step::do in order.

This way, each step::do being the one doing the actual changes (instead
of the halfedge_collapse funtion used in the algorithm) knows exactly 
what elements have been invalidated, so the corresponding step::undo can
operate specifically on those elements with are known to be still valid
after a previous call to step::do.

Even if I'm missing something and there are is still need to denote the 
mesh elements in a more stable way, or to fixup handles in the remaning 
steps, or any such thing, you now has complete control since changes are 
now applied by these step:do and step::undo methods, not by the 
halfedge_collapse function the algorithm used first.

I know this is suboptimal since the decimation has to be done twice and
goes totally astray with the current simplification code, but support
for progressive meshes where not considered during the algorithm design,
so this is as ugly as a workaround can be.
Yet off the top of my head it should work.
Consider that the algorithm does much more than just collapsing edges: 
it needs to decide which ones, which is the bulk of the problem, so 
there is still gain in calling the algorithm even if afterwards you end 
up doing the collapses "by hand" and once again.


Best

Fernando Cacciola

www.geometryfactory.com