Re: [Boost-users] compiling parallel boost graph library (PBGL)

Dear PBGL developers,

I'm getting errors when compiling parallel-bgl-0.7.0 (the release from PBGL website).

I'm compiling against boost 1.43.0, MPI implementation by MX (mpich/mx/gcc/64/1.2.7..3) and using gcc version 3.4.6 20060404 (Red Hat 3.4.6-11).

The error I'm getting is: ---- boost/graph/distributed/betweenness_centrality.hpp:574: error: using invalid field `boost::graph::parallel::detail::betweenness_centrality_delta_stepping_impl ::owner' ---- and apparently happens during linking.

I would appreciate your help. I put the printout from 'make' (verbosity on so that you can see the failing command) and my CMakeCache temporarily under:

http://www.cs.vu.nl/~ekr/pbgl-error/error.txt http://www.cs.vu.nl/~ekr/pbgl-error/CMakeCache.txt

Best regards,

Elzbieta Krepska

Hi Elzbieta, It looks like the issue is that extracting the vertex_property_type of a filtered graph where the base graph is a CSR graph using bundled properties the vertex_property_type isn't propagated properly. The SSCA code is trying to declare an explicit property_map type for the vertex_owner and vertex_local maps. I can think of several ways to get around this, the simplest is probably to play some dispatching game to get the compiler to generate the types for me. If that didn't work I'd probably have to take a closer look at the filtered_graph and distributed CSR code. Do you actually need the SSCA code for anything? As a benchmark it's pretty much dead. The failure of this test doesn't affect the rest of the library. It should only be an issue if you want to run betweenness centrality on filtered graphs. I'll add the fact that betweenness centrality doesn't work with filtered graphs in 0.7.0 to my list of bugs, but I'm in the middle of a big rewrite of the library at the moment. Let me know if this is a blocking issue for you and I'll try to carve out some free time to address it in the near term. The version of SSCA/Betweenness Centrality in the boost repository should have already addressed this issue as well so it may be worth trying or looking at to see what the fix is. Cheers, Nick