Using bundled properties as the weight map in dijkstra_shortest_paths

Question

Perhaps this is a stupid question, but I'm trying to use BGL's dijkstra_shortest_paths, and, in particular, to use a field of my Edge bundled property as the weightmap. My attempts have currently led to tens of pages of compiler errors, so I'm hoping that someone knows how to help me. This is essentially what my code looks like:

struct GraphEdge {
    float length;
    // other cruft
};
struct GraphVertex {
    ...
};
typedef boost::adjacency_list
<boost::vecS, boost::vecS, boost::directedS,
 GraphVertex, GraphEdge> GraphType;

I can populate the graph without any problems, however when it comes to calling dijkstra_shortest_paths, I get in trouble. I'd like to use the length field. Specifically, I'd like to know what's the bit of needed boost voodoo to go fit in a call like this:

GraphType m_graph;

vector<int> predecessor(num_vertices(m_graph));
vector<float> distances(num_vertices(m_graph), 0.0f);
vector<int> vertex_index_map(num_vertices(m_graph));
for (size_t i=0; i<vertex_index_map.size(); ++i) {
    vertex_index_map[i] = i;
}

dijkstra_shortest_paths(m_graph, vertex_from, predecessor, distances, 
                        weightmap, vertex_index_map, 
                        std::less<float>(), closed_plus<float>(), 
                        (std::numeric_limits<float>::max)(), 0.0f,
                        default_dijkstra_visitor());
// How do I write the right version of weightmap here?

such that weightmap will somehow associate a particular edge of my graph with the corresponding length field in the property. I'm sure there's an easy way to do this, but the documentation for BGL is incredibly opaque to me. If you can tell me where in the documentation the example is described, I'd be very happy as well.

Thank you in advance!

Answer 1

In case anyone cares about this, using the named parameter version of the call seems to have worked, as follows:

    dijkstra_shortest_paths(m_graph, vertex_from,
                        weight_map(get(&TrafficGraphEdge::length, m_graph))
                        .distance_map(make_iterator_property_map(distances.begin(),
                                                                 get(vertex_index, m_graph))));

This is in the documentation, here. I still don't know how to use the "non-named parameter" version of the call, though.

Answer 2

Ok, I just wasted too much time on this problem. Here is the solution for the posterity:

/**
 * @brief  Example concerning bundled properties.
 * @author Pierre-Andre Noel
 * @date   September 10 2012
 */

#include <iostream>
#include <boost/graph/adjacency_list.hpp>

/// The type of the field we are interested in.
typedef int interesting_type;

/// The struct whose elements will be bundled in each vertex.
struct bundled_in_vertex_type
{
  /// Something interesting.
  interesting_type something;
};

int main()
{
  typedef boost::adjacency_list< boost::vecS, boost::vecS, boost::undirectedS, bundled_in_vertex_type > graph_type;
  typedef graph_type::vertex_descriptor vertex_descriptor_type;

  /// Create a graph of two vertices.
  graph_type g(2);

  /// Name the two nodes.
  const vertex_descriptor_type v1(*boost::vertices(g).first), v2(*(++boost::vertices(g).first));

  // Store some stuff in the two nodes, the "easy" way.
  g[v1].something = interesting_type(42);
  g[v2].something = interesting_type(999);

  // Now what you came here for.
  /// An handle providing direct access to the field "something".
  boost::property_map< graph_type, interesting_type bundled_in_vertex_type::* >::type handle_to_something( boost::get(&bundled_in_vertex_type::something, g) );
  // You can now use "handle_to_something" for whatever deed you are interested in.

  // Just checking that it works.
  std::cout << "Vertex v1's ""something"" field is: " << handle_to_something[v1] << std::endl;
  std::cout << "Vertex v2's ""something"" field is: " << handle_to_something[v2] << std::endl;

  // Thank you and have a nice day.
  return 0;
}

Seriously, this library is great, but the documentation is definitively lacking. This should be a trivial matter.

---

EDIT

If you are using C++11, then you may prefer the following alternative.

    auto handle_to_something( boost::get(&bundled_in_vertex_type::something, g) );

Answer 3

As powerful as the BGL may be, unfortunately, it is not very easy to use in my honest opinion. Getting this to work took some considerable trial and error, but here is a working version compiled with Boost 1.53.0 [we want to use Dijkstra's algorithm on the 'rate' variable in __edge_data]:

struct __edge_data
{
    double rate;
    double edge_thickness;
    size_t colour;
};

struct __vertex_data
{   
   size_t colour; 
   size_t shape_code;
   string name;
};

typedef boost::adjacency_list<boost::vecS, boost::vecS, boost::directedS, __vertex_data, __edge_data> DIgraph;
typedef boost::graph_traits<DIgraph>::vertex_descriptor vertexx;
typedef boost::graph_traits<DIgraph>::vertex_iterator   vertexx_iter;
typedef boost::graph_traits<DIgraph>::edge_descriptor   edgee;

// functor
template<typename T>
struct combine_min : public std::binary_function<T, T, T>
{
        T const operator()(const T& a, const T& b) const
        {
            return b < a ? (b) : (a);
        }
};

// functor
template<typename T>
struct compare_less_than : public std::binary_function<T, T, bool>
{
        bool const operator()(const T& a, const T& b) const
        {
            return a < b;
        }
};

void graph_analysis()
{
     ...

      std::vector<vertexx>   parents(num_vertices(G)); 
      std::vector<double>  distances(num_vertices(G)); 

      auto p_map = boost::make_iterator_property_map(&parents[0], boost::get(boost::vertex_index, G));
      auto d_map = boost::make_iterator_property_map(&distances[0], boost::get(boost::vertex_index, G));
      auto w_map = boost::get(&__edge_data::rate_rate, G); // <=== THIS IS THE TRICK!!!
      auto n_map = boost::get(&__vertex_data::name, G);

      boost::dijkstra_shortest_paths(G, start_vertex_vector,
       boost::weight_map(w_map).
              predecessor_map(p_map).
              distance_map(d_map).
              distance_combine(combine_min<double>()).
              distance_compare(compare_less_than<double>()) );

    ...
}

I sincerely hope this helps! My attempt here was to show how to access all the main 'features' available to the algorithm.