> I am not a visad expert, but I recommend not coupling the graph (adjacency
> matrix or adjacency list) data structure to the existing VisAD data
> structures, until it comes time for the embedding in R^2 or R^3 (which is
> the function performed by the layout algorithm). It will be easier for
> people to import graph data into whatever you design if you keep the
> topology encoding and the cartesian embedding conceptually orthogonal to
> each other. It is also a natural interface boundary to make it easier to
> plug in layout algorithms that can work with arbitrary underlying topology
> I am curious if the rest of the visad community (especially those more
> familiar with the visad themes and design) agrees with this assessment.
> I am also curious if this is the sort of datatype that the main visad team
> feels is appropriate to integrate into the overall framework.
I generally agree about keeping data structure independent.
However, the visad.Delaunay class includes a variety of data
structures for managing irregular topologies in n-dimensions
that may be useful in *some* graph applications. Since that
class has some useful code for manipulating those structures,
its probably worth a quick look.
We would be delighted to integrate any new structures for
graph layout into VisAD. Or just to provide a link to another
system that can work with VisAD. For example, we have a link
to Unidata's Metapps software that uses and extends VisAD.