I've uploaded the bipartiteness stuff onto boost vault, as you recommended:
The documentation is in the same manner as the other algorithms, the example creates two graphs, tests them and prints the partitition (1st graph) and the odd-cycle (2nd graph), while the test code calls all implemented interfaces and verifies the certificates. The test code uses boost/test/minimal.hpp structure.
I haven't had a chance to look over your resubmission, but I agree with all of Jeremiah's comments. I just wanted to add some responses to your comments that you sent last week. The memory leaks were detected with valgrind on a Linux system, running against the test file that you provided (and I modified). The memory leaks are trivially solved if you can remove the dynamic allocations and use a container to control the allocations (list, stack, etc.). This also removes the need for a label allocator. With respect to list.size() issue, I there there are only a couple of places where this is still being called, and you're only looking to see if size() > 2. Why not write a function that tries to iterate 3 elements? It's still a O(1) function :) What about the name constrained_shortest_paths? Is that too general? I will try to go over the code in more detail later this week. If you want to address some of Jeremiah's comments before I look at it, let me know. Andrew Sutton andrew.n.sutton@gmail.com