Re: [Plib-users] <<code review>> Bounding sphere of a branch node

[Steve B. <sjb...@ai...>]

Rahul Choudhury wrote:

> The current ssg code marks a branch node's bouding
> sphere empty as and when an entity is added to the
> branch or removed from it. But the bounding sphere of
> the branch node really does not get updated. The
> places where it does get updated are in flatten() of
> ssgOptimiser or the constructor of ssgLeaf (which in
> turn calls the recalcBSphere of ssgBranch).
> 
> But think about a scenario where a Branch group has a
> leaf ssgRahulLeaf which is derived from ssgLeaf and
> has a different constrcutor which does not invoke
> recalcBSphere in it. Also imagine that the user of the
> ssg API did not call flatten on the scene graph before
> renedering it. Will the bounding spheres of the branch
> nodes be updated in this case ?
> 
> Is this better to update the bounding sphere of a
> branch node right when an entity is added to or
> removed from it ?

I don't exactly mark the sphere 'empty' - I mark it
'dirty' (ie invalid) - and SSG will recompute it
only when something asks for the value of the bsphere.

The thing to bear in mind is that dirtying a bounding
sphere also dirties the spheres above it in the scene
graph.

What SSG does is to only recompute a bsphere when it's
actually needed.

That gives you LARGE speedups when a lot of leaf nodes
are dirtied between each update because the higher up
nodes are not repeatedly recomputed.

In general, bspheres will get recomputed a minimal number
of times in the CullAndDraw routine - or the next isect
request or whatever.

Clever - eh!

One of the keys to fast scene graph API's is 'lazy evaluation'.
 
> Let me know your views.

I think I dodged *that* bullet :-)

-- 
Steve Baker                  http://web2.airmail.net/sjbaker1
sjb...@ai... (home)  http://www.woodsoup.org/~sbaker
sj...@ht...      (work)