On Sep 15, 2005, at 11:55 AM, Aaron Windsor wrote:
The main problem I can see with this solution is that if you call get(vertex_index,g) twice you get distinct instances of an auto_index_property_map, containing different associative containers. This could cause some pretty spectacular failures and might be a tough bug to catch within an algorithm. But from what I've seen, getting the vertex_index or edge_index property is usually done once by the algorithm, as in foo's signature above, so that all we would need to do is change the behavior of get(vertex_index,g) and get(edge_index,g) to return an auto_index_property_map. I'd be willing to code up an auto_index_property_map if anyone likes this idea, otherwise I vote for Doug's #4 above.
I like the idea of being able to build an auto_index_map map. However, I don't think it should be returned from get(vertex_index, g) automatically. My concern is that we'd be lying to the user, in effect, about the presence of a vertex index property. Now, if we do exactly the same thing but call the property vertex_auto_index, then I'd be thrilled to have an implementation and integrate it into the BGL. Doug