Computing more short or long paths per unit sounds expensive.
It should either only be a direct line obstruction collision test test or an obstruction area test (all points between the center of the attacker and the obstruction area of the attacker).
If a melee unit finds out it can't attack near the wall, it has to compute a path an alternative path to the target that omits the wall. I guess new obstruction grids couldn't help because they afaik don't capture the moving obstructions.
So the number of added collision tests might grow worse than proportional to the number of melee units attempting to attack. So we have to solve it as simply but effective as possible ideally.
