breadth_first_search calls tree_edge event point with wrapped edge when run on reverse_graph

[pmachata@…]

Probably a fallout of changeset [73997].

It seems one can get to the original edge by using the member "EdgeDesc underlying_desc". But I'm not sure whether this is the intended interface. The documentation of reverse_graph doesn't mention it.

Besides, with "straight" graph, breadth_first_search calls tree_edge with unwrapped edge, so one can't reuse the same code for both cases. (Or they have to invent accessor wrappers.)

Could this somehow be remedied? It doesn't seem intuitive to call tree_edge (and, presumably, others) with a wrapped edge instead of the edge itself. And it is inelegant that this is different for reverse graph than for "straight" graph.

[Jeremiah Willcock]

The reverse_graph wrapper produces a different graph type, so algorithms don't know about it (to algorithms, it's no different than any other graph type). Thus, it would not be elegant to special-case that one particular type of graph to unwrap its edge descriptors in the various algorithms.