View source for A simple turn scheduling system -- Python implementation

==Introduction==

I have been looking at time systems for roguelike games. I have found really simple systems that are not powerful enough and powerful systems that are very complicated. I decided to try to put something together that would be simple and powerful (enough). This system is a simple schedule of turns based on the monster's speed. I provide Python code for the scheduling system--the system is a good match for use in Python.

==Algorithm==

Define a quantum of time, call it a tick. Time is kept in a variable called ticks. A monster's speed is defined by the number of ticks between turns. For example, the pc might have a speed of 10; a faster monster might have a speed of 8, and really fast monster might have a speed of 2.

A schedule is kept of future moves, specifically a list of moves that are scheduled for future values of ticks. Implementing a schedule as a Python dict might look like this at ticks = 100:
<syntaxhighlight lang="python">
sched = { 101: [mon1, mon2], 105: [mon3, pc], 109: [mon4, mon5] }
</syntaxhighlight>

In that example, mon1 gets the next move followed by mon2. Then the ticks would get incremented to 105 and mon3 and the pc get to move.

When an actor makes a move it schedules its next move, so if mon1 and mon2 have speeds of 7 and 10, then the schedule in the above example would look like this at ticks = 102:
<syntaxhighlight lang="python">
sched = { 105: [mon3, pc], 107: [mon1], 109: [mon4, mon5], 111: [mon2] }
</syntaxhighlight>

==Code==

Here is a simple implementation in Python:
<syntaxhighlight lang="python" line>
#!/usr/bin/env python
# -*- coding: utf-8-*-

import random  # only needed for the fake monster

class Ticker(object):
    """Simple timer for roguelike games."""

    def __init__(self):
        self.ticks = 0  # current ticks--sys.maxint is 2147483647
        self.schedule = {}  # this is the dict of things to do {ticks: [obj1, obj2, ...], ticks+1: [...], ...}

    def schedule_turn(self, interval, obj):
        self.schedule.setdefault(self.ticks + interval, []).append(obj)

    def next_turn(self):
        things_to_do = self.schedule.pop(self.ticks, [])
        for obj in things_to_do:
            obj.do_turn()

###################################################################
#    Example main program

if __name__== "__main__":
    class Monster(object):
        """Fake monster for demo."""
        def __init__(self, ticker):
            self.ticker = ticker
            self.speed = 6 + random.randrange(1, 6)  # random speed in 7 - 12
            self.ticker.schedule_turn(self.speed, self) # schedule monsters 1st move

        def do_turn(self):
            print self, "gets a turn at", self.ticker.ticks # just print a message
            self.ticker.schedule_turn(self.speed, self)     # and schedule the next turn

    ticker = Ticker()  #  create our ticker
    print ticker.schedule

    monsters = []  #  create some monsters for the demo
    while len(monsters) < 5:
        monsters.append(Monster(ticker))
    print ticker.schedule

    while ticker.ticks < 51:  #  our main program loop
        if ticker.ticks in [10, 20, 30, 40, 50]:
            print ticker.schedule
        ticker.ticks += 1
        ticker.next_turn()
</syntaxhighlight>

Class Ticker has three methods (in this minimal implementation): __init__, which initializes the empty schedule and the ticks variable; schedule turn, which is called by the monster to insert its next turn into the schedule; and next_turn, which finds the next tick with scheduled items and runs them.

The main loop just increments the ticks and calls next_turn to make the next moves.

Some sample output looks like this:
<syntaxhighlight lang="text">
{}
{8: [<__main__.Monster instance at 0xb8056dcc>], 
 9: [<__main__.Monster instance at 0xb8056e0c>], 
11: [<__main__.Monster instance at 0xb8056dec>], 
 7: [<__main__.Monster instance at 0xb8056d8c>, <__main__.Monster instance at 0xb8056e2c>]}
<__main__.Monster instance at 0xb8056d8c> gets a turn at 7
<__main__.Monster instance at 0xb8056e2c> gets a turn at 7
<__main__.Monster instance at 0xb8056dcc> gets a turn at 8
<__main__.Monster instance at 0xb8056e0c> gets a turn at 9
{11: [<__main__.Monster instance at 0xb8056dec>], 
 14: [<__main__.Monster instance at 0xb8056d8c>, <__main__.Monster instance at 0xb8056e2c>],
 16: [<__main__.Monster instance at 0xb8056dcc>],
 18: [<__main__.Monster instance at 0xb8056e0c>]}
<__main__.Monster instance at 0xb8056dec> gets a turn at 11
<__main__.Monster instance at 0xb8056d8c> gets a turn at 14
<__main__.Monster instance at 0xb8056e2c> gets a turn at 14
<__main__.Monster instance at 0xb8056dcc> gets a turn at 16
<__main__.Monster instance at 0xb8056e0c> gets a turn at 18
{21: [<__main__.Monster instance at 0xb8056d8c>, <__main__.Monster instance at 0xb8056e2c>],
 22: [<__main__.Monster instance at 0xb8056dec>],
 24: [<__main__.Monster instance at 0xb8056dcc>],
 27: [<__main__.Monster instance at 0xb8056e0c>]}
<__main__.Monster instance at 0xb8056d8c> gets a turn at 21
<__main__.Monster instance at 0xb8056e2c> gets a turn at 21
<__main__.Monster instance at 0xb8056dec> gets a turn at 22
...</syntaxhighlight>
==Scheduling non-monster actions==

So far all discussion has been about scheduling of the moves of dungeon monsters (and the pc). Most roguelikes will have other actions that need to be scheduled. In this system any object (in the OO programming sense) that implements a do_turn method and calls the ticker's schedule_turn method can be timed. This should be general enough for all timing considerations.

For example, here is example code for a class that would control a protection spell that degraded over time.
<syntaxhighlight lang="python" line>
    class Protection(object):

        def __init__(self, ticker, ac, monster):
            self.ticker = ticker
            self.ac = ac # the amount of ac enhancement
            self.monster = monster
            self.decay_speed = 250  # protection decrements by 1 every 250 ticks
            self.monster.increment_ac(self.ac)  # this would increment the monster's ac
            self.ticker.schedule_turn(self.decay_speed, self) # schedule 1st ac decay

        def do_turn(self):
            self.ac += -1
            self.monster.increment_ac(-1) # so decrement the ac
            if self.ac > 0:   # if the ac is back to normal, don't reschedule
                self.ticker.schedule_turn(self.decay_speed, self)</syntaxhighlight>
==Other Considerations==

We have a variable that keeps track of the number of ticks since the beginning of the game:
<syntaxhighlight lang="python">        self.ticks = 0</syntaxhighlight>

Some may complain that ticking through the future schedule until we find scheduled actions is inelegant and wastes resources. In a real roguelike, with lots of monsters and other actions to be scheduled, I would expect that the percentage of ticks with no actions will not be large. Therefore, the overhead of iterating through the schedule will insignificant. Other time systems have to search through the schedule to insert actions or have to touch every monster every tick to update energy, or etc. I think the efficiency of this system will compare favorably with those.

It is inevitable that more than one turn will be scheduled for some ticks. Within a tick, the turns are taken in the same order as they were scheduled (i.e., FIFO). I think this is the correct implementation; if not, it can be easily modified.

The example is quite simplified in that the monster's speed is static. In a game it should be calculated from health, enchantments, weight carried, etc. This can be calculated on the fly for every call to schedule_turn, or it could be calculated when something changes, that would require storing both the base speed and current speed. 

Some game actions might take more (reading a spellbook) or less (reloading a shotgun) time than a full turn. That is also easily handled by calculating a pseudo speed for those operations before calling schedule_turn. One problem is when a long turn is interrupted (e.g., "you stop digging" in NetHack). In that case the currently scheduled action would have to be canceled and a new action scheduled. Canceling the scheduled action is a small problem. One approach would be to store the ticks of the next action in the monster, so that the monster can ask the ticker to cancel the now unneeded action.

==Kotlin Implementation==
See https://gist.github.com/Larkenx/5e1007eac05768865a77f5d2e75a5cda for a Kotlin implementation based off of this article.

[[Category:Scheduling]]