On Jul 26, 2007, at 1:34 PM, Huseyin Akcan wrote:
I guess my main question is does using descriptors have advantages over iterators in the design or the runtime of the graph library? Also are there other specific reasons to use descriptors that you experienced? Any insight on this issue would be appreciated.
The reason we need descriptors in the Graph library is that there are many different ways to iterate over a particular data structure, all with different iterator types, but we need a common type to talk about what those iterators refer to... that's why all of the iterators into a graph (say, into the adjacency_list) point to vertex or edge descriptors. That extra level of indirection allows us to talk about vertices or edges in the abstract sense, regardless of the way that we found those vertices or edges (enumerating out-edges, in- edges, neighbors, etc.) Now, we have run into similar problems with descriptors not always being the best answer for inserting or removing values from containers. To get around this, there are a few extra overloads for routines such as remove_out_edge (in the adjacency_list) that accept out_edge_iterators rather than edge descriptors. The iterator-based versions are more efficient, but the descriptor-based versions are more flexible (since they can take the descriptors returned from other kinds of iterators). - Doug