最近在项目中,涉及到对行为和状态进行建模的需求,尝试用有限状态机(Finite-state machine, FSM)来实现。
1. 概念介绍
1.1 运行机制
基于对有限状态机的粗浅理解,大体的运行机制为:
- 系统所处的状态是明确并且有限的,必定属于状态全集中的某一种;
- 系统接受输入,根据判定条件,决定是维持当前状态,还是切换到某一个新的状态;
- 在维持或切换的过程中,执行一些预设的操作。
可以认为有限状态机是一个离散系统,每接受一次输入,进行一次判断和切换。
1.2 所含要素
一个有限状态机包含如下几个要素:
-
状态:系统所处的状态,在运行过程中又可以分为当前状态和下一阶段状态;
-
事件:也可以理解为每一次运行的输入;
-
条件:根据输入事件执行的判定条件,条件是基于状态的,当前所处的每一种状态,都可以有自己对应的一套判定条件,来决定下一步进入哪一种状态;
-
动作:确定切换路径后,执行的附加操作。
以一个共3种状态的FSM为例,共有3套判定条件,根据当前所处的状态来确定使用哪一种判定条件,共有3*3=9种动作,决定每一种状态切换过程中需要执行的动作。
1.3 分析方法
通常可以用一个表格来对所处理的FSM进行分析,防止情况的遗漏。
在表格中分析清楚每一种状态切换的判定条件和执行动作,再用代码实现,可以最大程度地减轻思考的难度,减少错误的概率。
2. 代码实现
以OOP的方式,做了一个基础的Python实现。
FSM基类:
class StateMachine:def __init__(self, cfg, states, events_handler, actions_handler):# config information for an instanceself.cfg = cfg# define the states and the initial stateself.states = [s.lower() for s in states]self.state = self.states[0]# process the inputs according to current stateself.events = dict()# actions according to current transferself.actions = {state: dict() for state in self.states}# cached data for temporary useself.records = dict()# add events and actionsfor i, state in enumerate(self.states):self._add_event(state, events_handler[i])for j, n_state in enumerate(self.states):self._add_action(state, n_state, actions_handler[i][j])def _add_event(self, state, handler):self.events[state] = handlerdef _add_action(self, cur_state, next_state, handler):self.actions[cur_state][next_state] = handlerdef run(self, inputs):# decide the state-transfer according to the inputsnew_state, outputs = self.events[self.state](inputs, self.states, self.records, self.cfg)# do the actions related with the transferself.actions[self.state][new_state](outputs, self.records, self.cfg)# do the state transferself.state = new_statereturn new_statedef reset(self):self.state = self.states[0]self.records = dict()return# handlers for events and actions, event_X and action_XX are all specific functionsevents_handlers = [event_A, event_B]actions_handlers = [[action_AA, action_AB],[action_BA, action_BB]]# define an instance of StateMachinestate_machine = StateMachine(cfg, states, events_handlers, actions_handlers)
如果对于状态机有具体的要求,可以继承这个基类进行派生。
比如,有对状态机分层嵌套的需求。
class StateGeneral(StateMachine):def __init__(self, cfg, states):super(StateGeneral, self).__init__(cfg, states, events_handler, actions_handler)self.sub_state_machines = dict()def add_sub_fsm(self, name, fsm):self.sub_state_machines[name] = fsmdef run(self, inputs):new_state, outputs = self.events[self.state](inputs, self.states, self.records, self.cfg)# operate the sub_state_machines in actionsself.actions[self.state][new_state](outputs, self.records, self.cfg, \\self.sub_state_machines)self.state = new_statereturn new_statedef reset(self):self.state = self.states[0]self.records = dict()for _, sub_fsm in self.sub_state_machines.items():sub_fsm.reset()return