On May 13, 2005, at 4:53 AM, Martin Magnusson wrote:

I am sometimes getting segmentation faults (using gcc 3.3.4 under Cygwin) in calls to add_vertex( g ), where g is of the following type: boost::adjacency_list < boost::setS, boost::vecS, boost::undirectedS, Vertex_Properties, Edge_Length >

This happens when I try to merge two graphs with repeated calls to add_vertex (not updating edges or anything, just adding new vertices), and usually when the returned index from add_vertex is around 40000-50000, but it depends on the data. Creating a single graph with more than 50000 vertices in one go is not a problem.

Does anyone have an idea of what is going on in add_vertex, and what could be the problem here?

That's really strange... add_vertex is essentially just a push_back on the vector. Are you iterating over the vertices while calling add_vertex? You mention automatic resizing below: if the resizing did happen, it could invalidate the vertex_iterators into the graph, and might cause a crash.

I suspect it has something to do with the automatic resizing of the internal vector used to store the vertices. Is there a way to reserve memory in advance for this?

No "real" way, but the the "m_vertices" member of adjacency_list stores the vector of vertices, so you could call reserve() on it. In theory, the name could change in future releases, but it's unlikely to happen for a very long time :) Doug