Negative edge weight check in boost's dijkstra_shortest_paths

Question

I am using the boost graph library to make a call to dijkstra_shortest_paths . However, I have something special setup in that the weight_map is actually a functor. Hence, everytime the boost library requires the weight of an edge, my functor is called, makes a complicated computation and delivers the result back to boost.

Unfortunately, in dijkstra_shortest_paths.hpp the struct dijkstra_bfs_visitor 's method examine_edge has a get call to the weightmap, only to check if the returned value is negative. I am fully aware that I cannot use Dijkstra's algorithm with negative values and I am certain that my functor only returns positive values. However, this check causes my functor to be called twice for each edge. As it performs a complicated computation I'd like to avoid performing it twice (the results don't change between calls.. each edge gets the same wait during a dijkstra_shortest_paths run).

So far, I am manually checking the edge passed to the functor and in case of the duplicate call I am returning the previous remembered result. This clearly is more of a workaround than a solution.

I tried to pass my own visitor that overwrites examine_edge , however, the original method defined by boost's dijkstra_bfs_visitor is still applied.

Does anyone know if there is a better way to handle this situation and somehow avoid the negate edge weight check?

Answer 1

You are right, even supplying you own visitor the test of negativity will be performed.

  void examine_edge(Edge e, Graph& g) {
    if (m_compare(get(m_weight, e), m_zero))
        boost::throw_exception(negative_edge());
    m_vis.examine_edge(e, g);
  } 

But anyway the weight_map is supposed to be called multiple time (see boost doc ):

 for each vertex v in Adj[u]
  if (w(u,v) + d[u] < d[v])
    d[v] := w(u,v) + d[u]

Just call djikstra with a precomputed weight map, every edge weight will have to be computed anyway.

Answer 2

I suggest you use lazy computation to ensure that each expensive computation is carried out at most once. The most straightforward way to implement this would be give your Edge class a getWeight() lazy getter, I'm not familiar with the library you're using so I'm not sure how you'd integrate that with the functor.