As for spatial toolkits, I’m not sure what he wants but we probably don’t have it. We have an implementation of Fortune’s Voronoi algorithm, and we assume the GTS library has been installed. And some basic computational geometry things.
As for integrating this into Graphviz, it would go with the other edge routing algorithms (spline, line, polyline, orthogonal) in the neato library. (It’s not clear how this would fit into the dot layout.) We would create a new type of edge. The code would switch on that type and do its thing. I can point out exactly where this occurs in the code when the time comes. Based on figure 26, this routing is different from the others in that it seems to also move the nodes as well as computing the edge layouts. That shouldn’t be a problem for non-dot layouts.
The implementer can best determine how to incrementally introduce the code, probably starting with new types, auxiliary functions, etc. From our standpoint, the whole thing should probably be done as a separate library, as we currently handle various parts of neato/sfdp. At the global level, the code would be entirely isolated until the new edge type is introduced into the switch statement.