I ran graph-tool's source code through Coverity and it identified a couple of places where the code could be simplified.
Coverity also detected a possible bug on line 877 of graph_blockmodel_multiflip_mcmc.hh. The first argument to sample_split(size_t r, size_t s, RNG& rng) is r and the second is s, however r and s appear to be reversed when it is called with sample_split(s, r, rng). I didn't touch that code because I couldn't or tell whether or not this reversal was intentional, but I wanted to make you aware of it.