Beyond the Glass Box: Constraints as Objects

Constraint Logic Programming (CLP) is a very active research area. One reason being that nite domain CLP systems have been successfully applied to various combinatorial optimization problems such as time tabling, scheduling, frequency allocation, manpower planning, production planning. State of the art nite domain CLP languages o er programming constructs that gives access to the state of the constraint solver. With these constructs, new constraints can be de ned in the CLP language directly, hence the name \glass-box". However, current glass box approaches do not give access to the constraints themselves. This prevents the use of sophisticated constraint solving techniques borrowed from Operations Research. In order to overcome this limitation, we propose to give acces to the constraints as rst class citizen of the CLP language. We implemented this approach into an object oriented language, where constraints are explicitly represented by objects. We provide both theoretical and experimental evidence of the interest of our approach. In particular, we show that some of our extensions provide speedups of more than one order of magnitude over current glass box constraint solvers.