The problem of reconfiguring power distribution systems to reduce power losses has been extensively studied because of its significant economic impact. A variety of approximation computational models have recently been proposed. We describe a constraint programming model for this problem, using the Mozart system. To handle real world reconfiguration systems we implemented and integrated into Mozart an efficient constraint propagation system for the real numbers. We show how the CP approach leads to a simpler model and allows more flexible control of reconfiguration parameters. We analyze the performance of our system in canonical distribution networks of up to 60 nodes. We describe how the adaptability of the Mozart search engine allows defining effective strategies for tackling a real distribution system reconfiguration of around 600 nodes.