Perhaps one could even use histograms to estimate selectivity of the query's join predicates to pick a good ordering of joins? This is actually pretty much how

Perhaps one could even use histograms to estimate selectivity of the query's join predicates to pick a good ordering of joins? This is actually pretty much how a basic relational query optimizer works in practice.