simulate the mechanism of how BigWorld handles priorities of entities in the aoi of a witness. it seems that the priorities will finally all be cleared to zero, though very slow, hundreds of thousands iterations usually required at least and sometimes 100M iterations won't do, if all other entities are static.

simulate.py

#!/usr/bin/env pypy-c

from __future__ import print_function
import random
import math
import sys
import signal
import os
import time


MAX_DISTANCE = 100 ** 2


def calc(count, size):
    # a: [[priority, distance]]
    a = [[math.sqrt(random.randint(1, MAX_DISTANCE))] * 2 for _ in range(size)]
    for _ in range(count):
        a.sort()
        #print(a)
        mp = a[0][0] + 5
        i = 0
        for i in range(size):
            if a[i][0] <= mp:
                a[i][0] += a[i][1]
            else:
                break
        else:
            print("done, i = %d, %s" % (_, str(map(lambda x: round(x[1], 2), a))))
            return
        for j in range(i):
            if a[j][0] < a[i][0]:
                a[j][0] = a[i][0]
    print("failed, i = %d, %s" % (_, str(map(lambda x: x[1], a))))


if __name__ == '__main__':
    children = []
    for _ in range(4):
        # use as many children as our cpu cores to speedup
        pid = os.fork()
        if pid == 0:
            random.seed()  # otherwise all children will end up with the same a
            t1 = time.time()
            calc(100 * 1000 * 1000, 10)
            t2 = time.time()
            print("time cost:", t2 - t1)
            sys.exit(0)
        else:
            children.append(pid)

    try:
        try:
            while True:
                print(os.waitpid(-1, 0))
        except OSError:
            print("done")
    except KeyboardInterrupt:
        # kill all children or they become zombies that eat a lot cpu
        for pid in children:
            os.kill(pid, signal.SIGKILL)