[boost] Re: BGL wrapper for SGB

From: jsiek_at_[hidden]
Date: 2000-10-04 02:50:39

Hi Andreas,

Andreas Scherer writes:
 > --- In boost_at_[hidden], "Andreas Scherer" <as_at_m...> wrote:
 > > Only if my assumption is correct that the PROTOTYPES really are
 > > sufficient for <gb_graph.h> to be #included safely in
 > > <stanford_graph.hpp> in an "extern "C"" group.
 > 
 > Indeed it is!!!
 > 
 > I found that <stanford_graph.hpp> is used exclusively in the
 > "miles_span" example program.  So, I #included the patched SGB header
 > files <gb_graph.h>, <gb_miles.h>, and <gb_save.h> instead of the
 > repeated SGB stuff in <stanford_graph.hpp> and recompiled
 > "miles_span.cpp" with the updated headers and linked it with the newly
 > compiled SGB library.  Alas, this approach works fine.

Great!

 > But here are two remarks: Firstly, <stanford_graph.hpp> repeats only
 > the stuff needed in "miles_span.cpp", so it is _not_ a general
 > interface for other types of SGB graphs.  Why declare the "miles"
 > graph in <stanford_graph.hpp>, but non other generating function?

Laziness :(

Perhaps the way to go is to just remove the repeated stuff and make a
note in the documentation saying that the PROTOTYPES patch is
required.

 > Secondly, BGL's version of "miles_span" reports the minimum spanning
 > tree of length 8075, while SGB's version of "miles_span" reports a
 > length of 14467 (and this value also is "in the book"), although both
 > programs claim to operate on the same graph. I believe, one version
 > must be wrong!  (And guess which...)

Thanks for catching this! I should have included miles_span in my set
of regression tests... A recent change introduced a bug in
boost::prim_minimum_spanning_tree(). I've checked in the fix to the
boost sourceforge CVS repository.

Cheers,

Jeremy