Closed
Description
Currently we use 1 short to store the forward and 1 for the backward weight of the landmarks heuristic. We could improve precision if we would store the forward weight and the delta instead and then use e.g. 20 bits for the forward weight and only 12 for the delta (as the delta is usually small). This way we should be able to avoid most of the 'too many weights were maxed out' warnings for world wide preparation.
Related to #969