I'd like to continue our discussion on Geometric Software that could have continued in the panel discussion if there were no time constraints.

First of all, the CGAL project of building a geometric software library, is definitely an excellent idea and I would think it will have significant impact. We all seem to agree to this point that a library of some kind is needed.

However, there were discussions about the shape of the library. Opinions were expressed in the meeting that it is desirable to have a software library that is not monolithic. I myself would be in favor of having a library that consists of separable functions that can be easily extracted. To make an analogy it would not be convenient to have to get the entire library if one just wants to check out a 'book' in the library. Granted some of the fundamental datatypes required may have to be tagged along, but to ask the user to get the entire software library just to have access to a function, in my opinion, is not something the user would like to have, if they have a choice.

Can we have separable levels of hierarchy? For example, fundamental datatypes, CG_integers, etc. at the bottom-most level, primitives such as sideness, at one level, functions such as Convex hull, Delaunay triangulation, at another level, etc. If one wants to use convex hull, one can easily get relavant primitives required by the convex hull function, without getting primitive functions unrelated to convex hull computation.

This is just a personal suggestion. I just wanted to see a discussion on this issue by the community of computational geometers, and by the potential users of a geometric library outside of this community.

Please join us for the discussion as to how a library should be shaped, if such a library is needed.

D. T. Lee                        http://www.ece.nwu.edu/~dtlee/
Electrical and Computer Engineering   ADDR:  2145 Sheridan Road
Northwestern University	   	     	     Evanston, IL 60208
Phone (847) 491-5007                         Fax (847) 467-4144