This time, by collapsing the active-weights and augmented-weights neighbour filtering in astar_numpy into one unit, and guarding the reheat-list generation in preheat_upper_bounds.
For Reft + 8 surrounding sectors with default settings, overall pathfinding went from 290s to 258s - an 11.0% saving of runtime.
As name suggests, tune up pathfinding - again.
This time, by collapsing the active-weights and augmented-weights neighbour filtering in astar_numpy into one unit, and guarding the reheat-list generation in preheat_upper_bounds.
For Reft + 8 surrounding sectors with default settings, overall pathfinding went from 290s to 258s - an 11.0% saving of runtime.