
On Tue, 25 Jun 2013, lizy10b wrote:
Hi Jeremiah Willcock, Can I use the predecessor_map in undirected_dfs() like in dijkstra_shortest_paths()? undirected_dfs(g, predecessor_map(&p[0]).root_vertex(vertex_descriptor(0)).visitor(vis) .edge_color_map(get(edge_color, g)).vertex_color_map(get(vertex_color, g))); It does not work on my end, all elements in predecessor remain 0 after invoke the undirected_dfs() .
If that parameter isn't in the documentation for undirected_dfs, the argument will just be ignored.
In the last section (Extending Algorithms with Visitors) of http://www.boost.org/doc/libs/1_53_0/libs/graph/doc/quick_tour.html, an example shows how to write a predecessor_map using visitor, and the "event point" is edge_relaxed(). But as the undirected_dfs document shows there is no edge_relaxed() for DFS.
So is there any way to save predecessors in DFS? For example an edge (u,v) and start at vertex u
When the discover_vertex(Vertex, Graph&) is invoked at vertex v, then inside the discover_vertex() how could I know the previous discovered vertex is u?
I believe you would want to hook tree_edge() instead of edge_relaxed(). -- Jeremiah Willcock
Hi Jeremiah Willcock , ? It seems undirected_dfs() works. I have not燾heck the result according to the graph, but at least there is no forward edge or cross edge output. Attached is the revised cpp file using undirected_dfs() . My question is why edge_color_map(get(edge_color,爂))?is required when calling the undirected_dfs() . Thanks It looks like what is going wrong with using directed DFS is that edges are examined twice (once in each direction), and so you can have forward edges even in a symmetric graph (which is how undirected graphs are
_________________________________________________________________________________________________________________________________________________________ 发件人: Jeremiah Willcock 发送时间: 2013-06-25 00:44:05 收件人: lizy10b 抄送: boost-users 主题: Re: Re: [Boost-users] Confused with Depth-First-Search method,forwardor cross edge found on a undirected graph ? On Tue, 25 Jun 2013, lizy10b wrote: treated in directed-graph algorithms in BGL). An example is the triangle 0 -- 1 -- 2 -- 0: starting at 0 and processing the edges out of a vertex in numerical order, 0 -> 1 is a tree edge, 1 -> 0 is a back edge, 1 -> 2 is a tree edge, 2 -> 0 is a back edge, 2 -> 1 is a back edge, then 0 -> 2 is a forward edge. An undirected DFS would mark 2 -> 0 as a back edge then ignore 0 -> 2 since it is the same as 2 -> 0. The edge colors in undirected_dfs are used to do this filtering-out of multiple directions for the same edge. -- Jeremiah Willcock