path: root/system/doc/design_principles/statem.xml

<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">

<chapter>
  <header>
    <copyright>
      <year>2016</year><year>2018</year>
      <holder>Ericsson AB. All Rights Reserved.</holder>
    </copyright>
    <legalnotice>
      Licensed under the Apache License, Version 2.0 (the "License");
      you may not use this file except in compliance with the License.
      You may obtain a copy of the License at

          http://www.apache.org/licenses/LICENSE-2.0

      Unless required by applicable law or agreed to in writing, software
      distributed under the License is distributed on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
      See the License for the specific language governing permissions and
      limitations under the License.

    </legalnotice>

    <title>gen_statem Behavior</title>
    <prepared></prepared>
    <docno></docno>
    <date></date>
    <rev></rev>
    <file>statem.xml</file>
  </header>
  <marker id="gen_statem Behaviour" />
  <p>
    This section is to be read with the
    <seealso marker="stdlib:gen_statem"><c>gen_statem(3)</c></seealso>
    manual page in STDLIB, where all interface functions and callback
    functions are described in detail.
  </p>

<!-- =================================================================== -->

  <section>
    <marker id="Event-Driven State Machines" />
    <title>Event-Driven State Machines</title>
    <p>
      Established Automata Theory does not deal much with
      how a state transition is triggered,
      but assumes that the output is a function
      of the input (and the state) and that they are
      some kind of values.
    </p>
    <p>
      For an Event-Driven State Machine, the input is an event
      that triggers a state transition and the output
      is actions executed during the state transition.
      It can analogously to the mathematical model of a
      Finite-State Machine be described as
      a set of relations of the following form:
    </p>
    <pre>
State(S) x Event(E) -> Actions(A), State(S')</pre>
    <p>These relations are interpreted as follows:
      if we are in state <c>S</c> and event <c>E</c> occurs, we
      are to perform actions <c>A</c> and make a transition to
      state <c>S'</c>. Notice that <c>S'</c> can be equal to <c>S</c>
      and that <c>A</c> can be empty.
    </p>
    <p>
      As <c>A</c> and <c>S'</c> depend only on
      <c>S</c> and <c>E</c>, the kind of state machine described
      here is a Mealy machine
      (see, for example, the Wikipedia article "Mealy machine").
    </p>
    <p>
      Like most <c>gen_</c> behaviors, <c>gen_statem</c> keeps
      a server <c>Data</c> besides the state. Because of this, and as
      there is no restriction on the number of states
      (assuming that there is enough virtual machine memory)
      or on the number of distinct input events,
      a state machine implemented with this behavior
      is in fact Turing complete.
      But it feels mostly like an Event-Driven Mealy machine.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="When to use gen_statem" />
    <title>When to use gen_statem</title>
    <p>
      If your process logic is convenient to describe as a state machine,
      and you want any of these <c>gen_statem</c> key features:
    </p>
    <list type="bulleted">
      <item>
	Co-located callback code for each state,
	regardless of
	<seealso marker="#Event Types">Event Type</seealso>
	(such as <em>call</em>, <em>cast</em> and <em>info</em>)
      </item>
      <item>
	<seealso marker="#Postponing Events">
	  Postponing Events
	</seealso>
	(a substitute for selective receive)
      </item>
      <item>
	<seealso marker="#Inserted Events">
	  Inserted Events
	</seealso>
	that is: events from the state machine to itself
	(in particular purely internal events)
      </item>
      <item>
	<seealso marker="#State Enter Calls">
	  State Enter Calls
	</seealso>
	(callback on state entry co-located with the rest
	of each state's callback code)
      </item>
      <item>
	Easy-to-use timeouts
	(<seealso marker="#State Time-Outs">State Time-Outs</seealso>,
	<seealso marker="#Event Time-Outs">Event Time-Outs</seealso>
	and
	<seealso marker="#Generic Time-Outs">Generic Time-outs</seealso>
	(named time-outs))
      </item>
    </list>
    <p>
      If so, or if possibly needed in future versions,
      then you should consider using <c>gen_statem</c> over
      <seealso marker="stdlib:gen_server"><c>gen_server</c></seealso>.
    </p>
    <p>
      For simple state machines not needing these features
      <seealso marker="stdlib:gen_server"><c>gen_server</c></seealso>
      works just fine.
      It also has got smaller call overhead,
      but we are talking about something like 2 vs 3.3 microseconds
      call roundtrip time here, so if the server callback
      does just a little bit more than just replying,
      or if the call is not extremely frequent,
      that difference will be hard to notice.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Callback Module" />
    <title>Callback Module</title>
    <p>
      The callback module contains functions that implement
      the state machine.
      When an event occurs,
      the <c>gen_statem</c> behaviour engine
      calls a function in the callback module with the event,
      current state and server data.
      This function performs the actions for this event,
      and returns the new state and server data
      and also actions to be performed by the behaviour engine.
    </p>
    <p>
      The behaviour engine holds the state machine state,
      server data, timer references, a queue of posponed messages
      and other metadata.  It receives all process messages,
      handles the system messages, and calls the callback module
      with machine specific events.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Callback Modes" />
    <title>Callback Modes</title>
    <p>
      The <c>gen_statem</c> behavior supports two callback modes:
    </p>
    <taglist>
      <tag>
	<seealso marker="stdlib:gen_statem#type-callback_mode">
	  <c>state_functions</c>
	</seealso>
      </tag>
      <item>
        <p>
	  Events are handled by one callback function per state.
	</p>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-callback_mode">
	  <c>handle_event_function</c>
	  </seealso>
      </tag>
      <item>
        <p>
	  Events are handled by one single callback function.
	</p>
      </item>
    </taglist>
    <p>
      The callback mode is selected at server start
      and may be changed with a code upgrade/downgrade.
    </p>
    <p>
      See the section
      <seealso marker="#Event Handler">Event Handler</seealso>
      that describes the event handling callback function(s).
    </p>
    <p>
      The callback mode is selected by implementing
      a mandatory callback function
      <seealso marker="stdlib:gen_statem#Module:callback_mode/0">
	<c>Module:callback_mode()</c>
      </seealso>
      that returns one of the callback modes.
    </p>
    <p>
      The
      <seealso marker="stdlib:gen_statem#Module:callback_mode/0">
	<c>Module:callback_mode()</c>
      </seealso>
      function may also return a list containing the callback mode
      and the atom <c>state_enter</c> in which case
      <seealso marker="#State Enter Calls">State Enter Calls</seealso>
      are activated for the callback mode.
    </p>

    <section>
      <marker id="Choosing the Callback Mode" />
      <title>Choosing the Callback Mode</title>
      <p>
	The short version: choose <c>state_functions</c> -
	it is the one most like <c>gen_fsm</c>.
	But if you do not want the restriction that the state
	must be an atom, or if you do not want to write
	one event handler function per state; please read on...
      </p>
      <p>
	The two
	<seealso marker="#Callback Modes">Callback Modes</seealso>
	give different possibilities and restrictions,
	with one common goal:
	to handle all possible combinations of events and states.
      </p>
      <p>
	This can be done, for example, by focusing on one state at the time
	and for every state ensure that all events are handled.
	Alternatively, you can focus on one event at the time
	and ensure that it is handled in every state.
	You can also use a mix of these strategies.
      </p>
      <p>
	With <c>state_functions</c>, you are restricted to use
	atom-only states, and the <c>gen_statem</c> engine
	branches depending on state name for you.
	This encourages the callback module to co-locate
	the implementation of all event actions particular
	to one state in the same place in the code,
	hence to focus on one state at the time.
      </p>
      <p>
	This mode fits well when you have a regular state diagram,
	like the ones in this chapter, which describes all events and actions
	belonging to a state visually around that state,
	and each state has its unique name.
      </p>
      <p>
	With <c>handle_event_function</c>, you are free to mix strategies,
	as all events and states are handled in the same callback function.
      </p>
      <p>
	This mode works equally well when you want to focus on
	one event at the time or on
	one state at the time, but function
	<seealso marker="stdlib:gen_statem#Module:handle_event/4">
	  <c>Module:handle_event/4</c>
	</seealso>
	quickly grows too large to handle without branching to
	helper functions.
      </p>
      <p>
	The mode enables the use of non-atom states, for example,
	complex states or even hierarchical states.
	See section
	<seealso marker="#Complex State">Complex State</seealso>.
	If, for example, a state diagram is largely alike
	for the client side and the server side of a protocol,
	you can have a state <c>{StateName,server}</c> or
	<c>{StateName,client}</c>,
	and make <c>StateName</c> determine where in the code
	to handle most events in the state.
	The second element of the tuple is then used to select
	whether to handle special client-side or server-side events.
      </p>
    </section>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Event Handler" />
    <title>Event Handler</title>
    <p>
      Which callback function that handles an event
      depends on the callback mode:
    </p>
    <taglist>
      <tag><c>state_functions</c></tag>
      <item>
	The event is handled by:<br />
	<seealso marker="stdlib:gen_statem#Module:StateName/3">
	  <c>Module:StateName(EventType, EventContent, Data)</c>
	</seealso>
	<p>
	  This form is the one mostly used in the
	  <seealso marker="#Example">Example</seealso>
	  section.
	</p>
      </item>
      <tag><c>handle_event_function</c></tag>
      <item>
	The event is handled by:<br />
	<seealso marker="stdlib:gen_statem#Module:handle_event/4">
	  <c>Module:handle_event(EventType, EventContent, State, Data)</c>
	</seealso>
	<p>
	  See section
	  <seealso marker="#One Event Handler">One Event Handler</seealso>
	  for an example.
	</p>
      </item>
    </taglist>
    <p>
      The state is either the name of the function itself or an argument to it.
      The other arguments are the <c>EventType</c> described in section
      <seealso marker="#Event Types">Event Types</seealso>,
      the event dependent <c>EventContent</c>, and the current server <c>Data</c>.
    </p>
    <p>
      State enter calls are also handled by the event handler and have
      slightly different arguments. See the section
      <seealso marker="#State Enter Calls">State Enter Calls</seealso>.
    </p>
    <p>
      The event handler return values are defined in the description of
      <seealso marker="stdlib:gen_statem#Module:StateName/3">
	<c>Module:StateName/3</c>
      </seealso>
      in the <c>gen_statem</c> manual page, but here is
      a more readable list:
    </p>
    <taglist>
      <tag>
	<c>{next_state, NextState, NewData, Actions}</c><br />
	<c>{next_state, NextState, NewData}</c>
      </tag>
      <item>
	<p>
	  Set next state and update the server data.
	  If the <c>Actions</c> field is used, execute state transition actions.
	  An empty <c>Actions</c> list is equivalent to not returning the field.
	</p>
	<p>
	  See section
	  <seealso marker="#State Transition Actions">
	    State Transition Actions
	  </seealso>
	  for a list of possible
	  state transition actions.
	</p>
	<p>
	  If <c>NextState =/= State</c> the state machine changes
	  to a new state.  A
	  <seealso marker="#State Enter Calls">state enter call</seealso>
	  is performed if enabled and all
	  <seealso marker="#Postponing Events">postponed events</seealso>
	  are retried.
	</p>
      </item>
      <tag>
	<c>{keep_state, NewData, Actions}</c><br />
	<c>{keep_state, NewData}</c>
      </tag>
      <item>
	<p>
	  Same as the <c>next_state</c> values with
	  <c>NextState =:= State</c>, that is, no state change.
	</p>
      </item>
      <tag>
	<c>{keep_state_and_data, Actions}</c><br />
	<c>keep_state_and_data</c>
      </tag>
      <item>
	<p>
	  Same as the <c>keep_state</c> values with
	  <c>NextData =:= Data</c>, that is, no change in server data.
	</p>
      </item>
      <tag>
	<c>{repeat_state, NewData, Actions}</c><br />
	<c>{repeat_state, NewData}</c><br />
	<c>{repeat_state_and_data, Actions}</c><br />
	<c>repeat_state_and_data</c>
      </tag>
      <item>
	<p>
	  Same as the <c>keep_state</c> or <c>keep_state_and_data</c> values,
	  and if
	  <seealso marker="#State Enter Calls">
	    State Enter Calls
	  </seealso>
	  are enabled, repeat the state enter call
	  as if this state was entered again.
	</p>
      </item>
      <tag>
	<c>{stop, Reason, NewData}</c><br />
	<c>{stop, Reason}</c>
      </tag>
      <item>
	<p>
	  Stop the server with reason <c>Reason</c>.
	  If the <c>NewData</c> field is used, first update the server data.
	</p>
      </item>
      <tag>
	<c>{stop_and_reply, Reason, NewData, ReplyActions}</c><br />
	<c>{stop_and_reply, Reason, ReplyActions}</c>
      </tag>
      <item>
	<p>
	  Same as the <c>stop</c> values, but first execute the given
	  state transition actions that may only be reply actions.
	</p>
      </item>
    </taglist>

    <section>
      <marker id="The First State" />
      <title>The First State</title>
      <p>
	To decide the first state the
	<seealso marker="stdlib:gen_statem#Module:init/1">
	  <c>Module:init(Args)</c>
	</seealso>
	callback function is called before any
	<seealso marker="#Event Handler">Event Handler</seealso>
	is called.  This function behaves like an event handler
	function, but gets its only argument <c>Args</c> from 
	the <c>gen_statem</c>
	<seealso marker="stdlib:gen_statem#start/3">
	  <c>start/3,4</c>
	</seealso>
	or
	<seealso marker="stdlib:gen_statem#start_link/3">
	  <c>start_link/3,4</c>
	</seealso>
	function, and returns <c>{ok, State, Data}</c>
	or <c>{ok, State, Data, Actions}</c>.
	If you use the
	<seealso marker="#Postponing Events"><c>postpone</c></seealso>
	action from this function, that action is ignored,
	since there is no event	to postpone.
      </p>
    </section>

  </section>

<!-- =================================================================== -->

  <section>
    <marker id="State Transition Actions" />
    <title>State Transition Actions</title>
    <p>
      In the first section
      <seealso marker="#Event-Driven State Machines">
	Event-Driven State Machines
      </seealso>
      actions were mentioned as a part of
      the general state machine model. These general actions
      are implemented with the code that callback module
      <c>gen_statem</c> executes in an event-handling
      callback function before returning
      to the <c>gen_statem</c> engine.
    </p>
    <p>
      There are more specific state-transition actions
      that a callback function can command the <c>gen_statem</c>
      engine to do after the callback function return.
      These are commanded by returning a list of
      <seealso marker="stdlib:gen_statem#type-action">actions</seealso>
      in the
      <seealso marker="stdlib:gen_statem#type-state_callback_result">
	return value
      </seealso>
      from the
      <seealso marker="stdlib:gen_statem#Module:StateName/3">callback function</seealso>.
      These are the possible state transition actions:
    </p>
    <taglist>
      <tag>
	<seealso marker="stdlib:gen_statem#type-postpone">
	  <c>postpone</c>
	</seealso>
	<br />
	<c>{postpone, Boolean}</c>
      </tag>
      <item>
	If set postpone the current event, see section
	<seealso marker="#Postponing Events">Postponing Events</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-hibernate">
	  <c>hibernate</c>
	</seealso>
	<br />
	<c>{hibernate, Boolean}</c>
      </tag>
      <item>
	If set hibernate the <c>gen_statem</c>, treated in section
	<seealso marker="#Hibernation">Hibernation</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-state_timeout">
	  <c>{state_timeout, Time}</c>
	</seealso>
	<br />
	<c>{state_timeout, Time, Opts}</c>
      </tag>
      <item>
	Start a state time-out, read more in section
	<seealso marker="#State Time-Outs">State Time-Outs</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-generic_timeout">
	  <c>{{timeout, Name}, Time}</c>
	</seealso>
	<br />
	<c>{{timeout, Name}, Time, Opts}</c>
      </tag>
      <item>
	Start a generic time-out, read more in section
	<seealso marker="#Generic Time-Outs">Generic Time-Outs</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-event_timeout">
	  <c>{timeout, Time}</c>
	</seealso>
	<br />
	<c>{timeout, Time, Opts}</c><br />
	<c>Time</c>
      </tag>
      <item>
	Start an event time-out, see more in section
	<seealso marker="#Event Time-Outs">Event Time-Outs</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-reply_action">
	  <c>{reply, From, Reply}</c>
	</seealso>
      </tag>
      <item>
	Reply to a caller, mentioned at the end of section
	<seealso marker="#All State Events">All State Events</seealso>
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-action">
	  <c>{next_event, EventType, EventContent}</c>
	</seealso>
      </tag>
      <item>
	Generate the next event to handle, see section 
	<seealso marker="#Inserted Events">Inserted Events</seealso>
      </item>
    </taglist>
    <p>
      For details, see the <c>gen_statem(3)</c>
      manual page for type
      <seealso marker="stdlib:gen_statem#type-action"><c>action()</c></seealso>.
      You can, for example, reply to many callers,
      generate multiple next events,
      and set a time-out to use absolute instead of relative time
      (using the <c>Opts</c> field).
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Event Types" />
    <title>Event Types</title>
    <p>
      Events are categorized in different
      <seealso marker="stdlib:gen_statem#type-event_type">event types</seealso>.
      Events of all types are for a given state
      handled in the same callback function, and that function gets
      <c>EventType</c> and <c>EventContent</c> as arguments.
    </p>
    <p>
      The following is a complete list of event types and where
      they come from:
    </p>
    <taglist>
      <tag>
	<seealso marker="stdlib:gen_statem#type-external_event_type">
	  <c>cast</c>
	</seealso>
      </tag>
      <item>
	Generated by
	<seealso marker="stdlib:gen_statem#cast/2"><c>gen_statem:cast</c></seealso>.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-external_event_type">
	  <c>{call,From}</c>
	</seealso>
      </tag>
      <item>
	Generated by
	<seealso marker="stdlib:gen_statem#call/2"><c>gen_statem:call</c></seealso>,
	where <c>From</c> is the reply address to use
	when replying either through the state transition action
	<c>{reply,From,Msg}</c> or by calling
	<seealso marker="stdlib:gen_statem#reply/1"><c>gen_statem:reply</c></seealso>.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-external_event_type">
	  <c>info</c>
	</seealso>
      </tag>
      <item>
	Generated by any regular process message sent to
	the <c>gen_statem</c> process.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-timeout_event_type">
	  <c>state_timeout</c>
	</seealso>
      </tag>
      <item>
	Generated by state transition action
	<seealso marker="stdlib:gen_statem#type-state_timeout">
	  <c>{state_timeout,Time,EventContent}</c>
	</seealso>
	state timer timing out.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-timeout_event_type">
	  <c>{timeout,Name}</c>
	</seealso>
      </tag>
      <item>
	Generated by state transition action
	<seealso marker="stdlib:gen_statem#type-generic_timeout">
	  <c>{{timeout,Name},Time,EventContent}</c>
	</seealso>
	generic timer timing out.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-timeout_event_type">
	  <c>timeout</c>
	</seealso>
      </tag>
      <item>
	Generated by state transition action
	<seealso marker="stdlib:gen_statem#type-event_timeout">
	  <c>{timeout,Time,EventContent}</c>
	</seealso>
	(or its short form <c>Time</c>)
	event timer timing out.
      </item>
      <tag>
	<seealso marker="stdlib:gen_statem#type-event_type">
	  <c>internal</c>
	</seealso>
      </tag>
      <item>
	Generated by state transition
	<seealso marker="stdlib:gen_statem#type-action">action</seealso>
	<c>{next_event,internal,EventContent}</c>.
	All event types above can also be generated using
	<c>{next_event,EventType,EventContent}</c>.
      </item>
    </taglist>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="State Enter Calls" />
    <title>State Enter Calls</title>
    <p>
      The <c>gen_statem</c> behavior can if this is enabled,
      regardless of callback mode,
      automatically
      <seealso marker="stdlib:gen_statem#type-state_enter">
	call the state callback
      </seealso>
      with special arguments whenever the state changes
      so you can write state enter actions
      near the rest of the state transition rules.
      It typically looks like this:
    </p>
    <pre>
StateName(enter, OldState, Data) ->
    ... code for state enter actions here ...
    {keep_state, NewData};
StateName(EventType, EventContent, Data) ->
    ... code for actions here ...
    {next_state, NewStateName, NewData}.</pre>
    <p>
      Since the state enter call is not an event there are restrictions
      on the allowed return value and
      <seealso marker="#State Transition Actions">State Transition Actions</seealso>.
      You may not change the state,
      <seealso marker="#Postponing Events">postpone</seealso>
      this non-event, or
      <seealso marker="#Inserted Events">insert events</seealso>.
    </p>
    <p>
      The first state that is entered will get a state enter call
      with <c>OldState</c> equal to the current state.
    </p>
    <p>
      You may repeat the state enter call using the <c>{repeat_state,...}</c>
      return value from the 
      <seealso marker="#Event Handler">Event Handler</seealso>.
      In this case <c>OldState</c> will also be equal to the current state.
    </p>
    <p>
      Depending on how your state machine is specified,
      this can be a very useful feature,
      but it forces you to handle the state enter calls in all states.
      See also the
      <seealso marker="#State Enter Actions">
	State Enter Actions
      </seealso>
      chapter.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Example" />
    <title>Example</title>
    <p>
      A door with a code lock can be seen as a state machine.
      Initially, the door is locked. When someone presses a button,
      an event is generated.
      The pressed buttons are collected, up to the number of buttons
      in the correct code.
      If correct, the door is unlocked for 10 seconds.
      If not correct, we wait for a new button to be pressed.
    </p>
    <!-- The image is edited with dia in a .dia file,
	 then exported to Scalable Vector Graphics. -->
    <image file="../design_principles/code_lock.svg" width="80%">
      <icaption>Code Lock State Diagram</icaption>
    </image>
    <p>
      This code lock state machine can be implemented using
      <c>gen_statem</c> with the following callback module:
    </p>
    <code type="erl"><![CDATA[
-module(code_lock).
-behaviour(gen_statem).
-define(NAME, code_lock).

-export([start_link/1]).
-export([button/1]).
-export([init/1,callback_mode/0,terminate/3]).
-export([locked/3,open/3]).

start_link(Code) ->
    gen_statem:start_link({local,?NAME}, ?MODULE, Code, []).

button(Button) ->
    gen_statem:cast(?NAME, {button,Button}).

init(Code) ->
    do_lock(),
    Data = #{code => Code, length => length(Code), buttons => []},
    {ok, locked, Data}.

callback_mode() ->
    state_functions.
    ]]></code>
    <code type="erl"><![CDATA[
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
    NewButtons =
        if
            length(Buttons) < Length ->
                Buttons;
            true ->
                tl(Buttons)
        end ++ [Button],
    if
        NewButtons =:= Code -> % Correct
	    do_unlock(),
            {next_state, open, Data#{buttons := []},
             [{state_timeout,10000,lock}]}; % Time in milliseconds
	true -> % Incomplete | Incorrect
            {next_state, locked, Data#{buttons := NewButtons}}
    end.
    ]]></code>
    <code type="erl"><![CDATA[
open(state_timeout, lock,  Data) ->
    do_lock(),
    {next_state, locked, Data};
open(cast, {button,_}, Data) ->
    {next_state, open, Data}.
    ]]></code>
    <code type="erl"><![CDATA[
do_lock() ->
    io:format("Lock~n", []).
do_unlock() ->
    io:format("Unlock~n", []).

terminate(_Reason, State, _Data) ->
    State =/= locked andalso do_lock(),
    ok.
    ]]></code>
    <p>The code is explained in the next sections.</p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Starting gen_statem" />
    <title>Starting gen_statem</title>
    <p>
      In the example in the previous section, <c>gen_statem</c> is
      started by calling <c>code_lock:start_link(Code)</c>:
    </p>
    <code type="erl"><![CDATA[
start_link(Code) ->
    gen_statem:start_link({local,?NAME}, ?MODULE, Code, []).
    ]]></code>
    <p>
      <c>start_link</c> calls function
      <seealso marker="stdlib:gen_statem#start_link/4"><c>gen_statem:start_link/4</c></seealso>,
      which spawns and links to a new process, a <c>gen_statem</c>.
    </p>
    <list type="bulleted">
      <item>
        <p>
          The first argument, <c>{local,?NAME}</c>, specifies
          the name. In this case, the <c>gen_statem</c> is locally
          registered as <c>code_lock</c> through the macro <c>?NAME</c>.
        </p>
        <p>
          If the name is omitted, the <c>gen_statem</c> is not registered.
          Instead its pid must be used. The name can also be specified
          as <c>{global,Name}</c>, then the <c>gen_statem</c> is
          registered using
          <seealso marker="kernel:global#register_name/2"><c>global:register_name/2</c></seealso>
          in Kernel.
        </p>
      </item>
      <item>
        <p>
          The second argument, <c>?MODULE</c>, is the name of
          the callback module, that is, the module where the callback
          functions are located, which is this module.
        </p>
        <p>
          The interface functions (<c>start_link/1</c> and <c>button/1</c>)
          are located in the same module as the callback functions
          (<c>init/1</c>, <c>locked/3</c>, and <c>open/3</c>).
          It is normally good programming practice to have the client-side
          code and the server-side code contained in one module.
        </p>
      </item>
      <item>
        <p>
          The third argument, <c>Code</c>, is a list of digits, which
          is the correct unlock code that is passed
          to callback function <c>init/1</c>.
	</p>
      </item>
      <item>
        <p>
          The fourth argument, <c>[]</c>, is a list of options.
          For the available options, see
          <seealso marker="stdlib:gen_statem#start_link/3"><c>gen_statem:start_link/3</c></seealso>.
	</p>
      </item>
    </list>
    <p>
      If name registration succeeds, the new <c>gen_statem</c> process
      calls callback function <c>code_lock:init(Code)</c>.
      This function is expected to return <c>{ok, State, Data}</c>,
      where <c>State</c> is the initial state of the <c>gen_statem</c>,
      in this case <c>locked</c>; assuming that the door is locked to begin
      with. <c>Data</c> is the internal server data of the <c>gen_statem</c>.
      Here the server data is a <seealso marker="stdlib:maps">map</seealso>
      with key <c>code</c> that stores the correct button sequence,
      key <c>length</c> store its length,
      and key <c>buttons</c> that stores the collected buttons
      up to the same length.
    </p>

    <code type="erl"><![CDATA[
init(Code) ->
    do_lock(),
    Data = #{code => Code, length => length(Code), buttons => []},
    {ok, locked, Data}.
    ]]></code>
    <p>Function
      <seealso marker="stdlib:gen_statem#start_link/3"><c>gen_statem:start_link</c></seealso>
      is synchronous. It does not return until the <c>gen_statem</c>
      is initialized and is ready to receive events.
    </p>
    <p>
      Function
      <seealso marker="stdlib:gen_statem#start_link/3"><c>gen_statem:start_link</c></seealso>
      must be used if the <c>gen_statem</c>
      is part of a supervision tree, that is, started by a supervisor.
      Another function,
      <seealso marker="stdlib:gen_statem#start/3"><c>gen_statem:start</c></seealso>
      can be used to start a standalone <c>gen_statem</c>, that is,
      a <c>gen_statem</c> that is not part of a supervision tree.
    </p>

    <p>
      Function
      <seealso marker="stdlib:gen_statem#Module:callback_mode/0"><c>Module:callback_mode/0</c></seealso>
      selects the
      <seealso marker="#Callback Modes"><c>CallbackMode</c></seealso>
      for the callback module, in this case
      <seealso marker="stdlib:gen_statem#type-callback_mode"><c>state_functions</c></seealso>.
      That is, each state has got its own handler function:
    </p>
    <code type="erl"><![CDATA[
callback_mode() ->
    state_functions.
    ]]></code>

  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Handling Events" />
    <title>Handling Events</title>
    <p>The function notifying the code lock about a button event is
      implemented using
      <seealso marker="stdlib:gen_statem#cast/2"><c>gen_statem:cast/2</c></seealso>:
    </p>
    <code type="erl"><![CDATA[
button(Button) ->
    gen_statem:cast(?NAME, {button,Button}).
    ]]></code>
    <p>
      The first argument is the name of the <c>gen_statem</c> and must
      agree with the name used to start it. So, we use the
      same macro <c>?NAME</c> as when starting.
      <c>{button,Button}</c> is the event content.
    </p>
    <p>
      The event is sent to the <c>gen_statem</c>.
      When the event is received, the <c>gen_statem</c> calls
      <c>StateName(cast, Event, Data)</c>, which is expected to
      return a tuple <c>{next_state, NewStateName, NewData}</c>,
      or <c>{next_state, NewStateName, NewData, Actions}</c>.
      <c>StateName</c> is the name of the current state and
      <c>NewStateName</c> is the name of the next state to go to.
      <c>NewData</c> is a new value for the server data of
      the <c>gen_statem</c>, and <c>Actions</c> is a list of
      actions to be performed by the <c>gen_statem</c> engine.
    </p>

    <code type="erl"><![CDATA[
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
    NewButtons =
        if
            length(Buttons) < Length ->
                Buttons;
            true ->
                tl(Buttons)
        end ++ [Button],
    if
        NewButtons =:= Code -> % Correct
	    do_unlock(),
            {next_state, open, Data#{buttons := []},
             [{state_timeout,10000,lock}]}; % Time in milliseconds
	true -> % Incomplete | Incorrect
            {next_state, locked, Data#{buttons := NewButtons}}
    end.
    ]]></code>
    <p>
      In state <c>locked</c>, when a button is pressed,
      it is collected with the last pressed buttons
      up to the length of the correct code,
      and compared with the correct code.
      Depending on the result, the door is either unlocked
      and the <c>gen_statem</c> goes to state <c>open</c>,
      or the door remains in state <c>locked</c>.
    </p>
    <p>
      When changing to state <c>open</c>, the collected
      buttons are reset, the lock unlocked, and a state timer
      for 10 s is started.
    </p>

    <code type="erl"><![CDATA[
open(cast, {button,_}, Data) ->
    {next_state, open, Data}.
    ]]></code>
    <p>
      In state <c>open</c>, a button event is ignored
      by staying in the same state.  This can also be done
      by returning <c>{keep_state, Data}</c> or in this case
      since <c>Data</c> unchanged even by returning
      <c>keep_state_and_data</c>.
    </p>
  </section>

  <section>
    <marker id="State Time-Outs" />
    <title>State Time-Outs</title>
    <p>
      When a correct code has been given, the door is unlocked and
      the following tuple is returned from <c>locked/2</c>:
    </p>
    <code type="erl"><![CDATA[
{next_state, open, Data#{buttons := []},
 [{state_timeout,10000,lock}]}; % Time in milliseconds
 ]]></code>
    <p>
      10,000 is a time-out value in milliseconds.
      After this time (10 seconds), a time-out occurs.
      Then, <c>StateName(state_timeout, lock, Data)</c> is called.
      The time-out occurs when the door has been in state <c>open</c>
      for 10 seconds. After that the door is locked again:
    </p>
    <code type="erl"><![CDATA[
open(state_timeout, lock,  Data) ->
    do_lock(),
    {next_state, locked, Data};
    ]]></code>
    <p>
      The timer for a state time-out is automatically cancelled
      when the state machine changes states.  You can restart
      a state time-out by setting it to a new time, which cancels
      the running timer and starts a new.  This implies that
      you can cancel a state time-out by restarting it with
      time <c>infinity</c>.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="All State Events" />
    <title>All State Events</title>
    <p>
      Sometimes events can arrive in any state of the <c>gen_statem</c>.
      It is convenient to handle these in a common state handler function
      that all state functions call for events not specific to the state.
    </p>
    <p>
      Consider a <c>code_length/0</c> function that returns
      the length of the correct code.
      We dispatch all events that are not state-specific
      to the common function <c>handle_common/3</c>:
    </p>
    <code type="erl"><![CDATA[
...
-export([button/1,code_length/0]).
...

code_length() ->
    gen_statem:call(?NAME, code_length).

...
locked(...) -> ... ;
locked(EventType, EventContent, Data) ->
    handle_common(EventType, EventContent, Data).

...
open(...) -> ... ;
open(EventType, EventContent, Data) ->
    handle_common(EventType, EventContent, Data).

handle_common({call,From}, code_length, #{code := Code} = Data) ->
    {keep_state, Data,
     [{reply,From,length(Code)}]}.
    ]]></code>

    <p>
      Another way to do it is through a convenience macro
      <c>?HANDLE_COMMON/0</c>:
    </p>
    <code type="erl"><![CDATA[
...
-export([button/1,code_length/0]).
...

code_length() ->
    gen_statem:call(?NAME, code_length).

-define(HANDLE_COMMON,
    ?FUNCTION_NAME(T, C, D) -> handle_common(T, C, D)).
%%
handle_common({call,From}, code_length, #{code := Code} = Data) ->
    {keep_state, Data,
     [{reply,From,length(Code)}]}.

...
locked(...) -> ... ;
?HANDLE_COMMON.

...
open(...) -> ... ;
?HANDLE_COMMON.
]]></code>

    <p>
      This example uses
      <seealso marker="stdlib:gen_statem#call/2"><c>gen_statem:call/2</c></seealso>,
      which waits for a reply from the server.
      The reply is sent with a <c>{reply,From,Reply}</c> tuple
      in an action list in the <c>{keep_state, ...}</c> tuple
      that retains the current state.  This return form is convenient
      when you want to stay in the current state but do not know or
      care about what it is.
    </p>
    <p>
      If the common event handler needs to know the current state
      a function <c>handle_common/4</c> can be used instead:
    </p>
    <code type="erl"><![CDATA[
-define(HANDLE_COMMON,
    ?FUNCTION_NAME(T, C, D) -> handle_common(T, C, ?FUNCTION_NAME, D)).
    ]]></code>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="One Event Handler" />
    <title>One Event Handler</title>
    <p>
      If
      <seealso marker="#Callback Modes">
	Callback Mode
      </seealso>
      <c>handle_event_function</c> is used,
      all events are handled in
      <seealso marker="stdlib:gen_statem#Module:handle_event/4"><c>Module:handle_event/4</c></seealso>
      and we can (but do not have to) use an event-centered approach
      where we first branch depending on event
      and then depending on state:
    </p>
    <code type="erl"><![CDATA[
...
-export([handle_event/4]).

...
callback_mode() ->
    handle_event_function.

handle_event(cast, {button,Button}, State, #{code := Code} = Data) ->
    case State of
	locked ->
            #{length := Length, buttons := Buttons} = Data,
            NewButtons =
                if
                    length(Buttons) < Length ->
                        Buttons;
                    true ->
                        tl(Buttons)
                end ++ [Button],
            if
                NewButtons =:= Code -> % Correct
                    do_unlock(),
                    {next_state, open, Data#{buttons := []},
                     [{state_timeout,10000,lock}]}; % Time in milliseconds
                true -> % Incomplete | Incorrect
                    {keep_state, Data#{buttons := NewButtons}}
            end;
	open ->
            keep_state_and_data
    end;
handle_event(state_timeout, lock, open, Data) ->
    do_lock(),
    {next_state, locked, Data};
handle_event(
  {call,From}, code_length, _State, #{code := Code} = Data) ->
    {keep_state, Data,
     [{reply,From,length(Code)}]}.

...
]]></code>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Stopping" />
    <title>Stopping</title>

    <section>
      <marker id="In a Supervision Tree" />
      <title>In a Supervision Tree</title>
      <p>
	If the <c>gen_statem</c> is part of a supervision tree,
	no stop function is needed.
	The <c>gen_statem</c> is automatically terminated by its supervisor.
	Exactly how this is done is defined by a
	<seealso marker="sup_princ#shutdown">shutdown strategy</seealso>
	set in the supervisor.
      </p>
      <p>
	If it is necessary to clean up before termination, the shutdown
	strategy must be a time-out value and the <c>gen_statem</c> must
	in function <c>init/1</c> set itself to trap exit signals
	by calling
	<seealso marker="erts:erlang#process_flag/2"><c>process_flag(trap_exit, true)</c></seealso>:
      </p>
      <code type="erl"><![CDATA[
init(Args) ->
    process_flag(trap_exit, true),
    do_lock(),
    ...
    ]]></code>
      <p>
	When ordered to shut down, the <c>gen_statem</c> then calls
	callback function <c>terminate(shutdown, State, Data)</c>.
      </p>
      <p>
	In this example, function <c>terminate/3</c>
	locks the door if it is open, so we do not accidentally leave the door
	open when the supervision tree terminates:
      </p>
      <code type="erl"><![CDATA[
terminate(_Reason, State, _Data) ->
    State =/= locked andalso do_lock(),
    ok.
    ]]></code>
    </section>

    <section>
      <marker id="Standalone gen_statem" />
      <title>Standalone gen_statem</title>
      <p>
	If the <c>gen_statem</c> is not part of a supervision tree,
	it can be stopped using
	<seealso marker="stdlib:gen_statem#stop/1"><c>gen_statem:stop</c></seealso>,
	preferably through an API function:
      </p>
      <code type="erl"><![CDATA[
...
-export([start_link/1,stop/0]).

...
stop() ->
    gen_statem:stop(?NAME).
    ]]></code>
      <p>
	This makes the <c>gen_statem</c> call callback function
	<c>terminate/3</c> just like for a supervised server
	and waits for the process to terminate.
      </p>
    </section>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Event Time-Outs" />
    <title>Event Time-Outs</title>
    <p>
      A time-out feature inherited from <c>gen_statem</c>'s predecessor
      <seealso marker="stdlib:gen_fsm"><c>gen_fsm</c></seealso>,
      is an event time-out, that is,
      if an event arrives the timer is cancelled.
      You get either an event or a time-out, but not both.
    </p>
    <p>
      It is ordered by the state transition action
      <c>{timeout,Time,EventContent}</c>, or just an integer <c>Time</c>,
      even without the enclosing actions list
      (the latter is a form inherited from <c>gen_fsm</c>.
    </p>
    <p>
      This type of time-out is useful for example to act on inactivity.
      Let us restart the code sequence
      if no button is pressed for say 30 seconds:
    </p>
    <code type="erl"><![CDATA[
...

locked(timeout, _, Data) ->
    {next_state, locked, Data#{buttons := []}};
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
...
	true -> % Incomplete | Incorrect
            {next_state, locked, Data#{buttons := NewButtons},
             30000} % Time in milliseconds
...
]]></code>
    <p>
      Whenever we receive a button event we start an event time-out
      of 30 seconds, and if we get an event type <c>timeout</c>
      we reset the remaining code sequence.
    </p>
    <p>
      An event time-out is cancelled by any other event so you either
      get some other event or the time-out event.  It is therefore
      not possible nor needed to cancel or restart an event time-out.
      Whatever event you act on has already cancelled
      the event time-out...
    </p>
    <p>
      Note that an event time-out does not work well with
      when you have for example a status call as in
      <seealso marker="#All State Events">All State Events</seealso>,
      or handle unknown events, since all kinds of events
      will cancel the event time-out.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Generic Time-Outs" />
    <title>Generic Time-Outs</title>
    <p>
      The previous example of state time-outs only work if
      the state machine stays in the same state during the
      time-out time.  And event time-outs only work if no
      disturbing unrelated events occur.
    </p>
    <p>
      You may want to start a timer in one state and respond
      to the time-out in another, maybe cancel the time-out
      without changing states, or perhaps run multiple
      time-outs in parallel. All this can be accomplished with
      <seealso marker="stdlib:gen_statem#type-generic_timeout">generic time-outs</seealso>.
      They may look a little bit like
      <seealso marker="stdlib:gen_statem#type-event_timeout">event time-outs</seealso>
      but contain a name to allow for any number of them simultaneously
      and they are not automatically cancelled.
    </p>
    <p>
      Here is how to accomplish the state time-out
      in the previous example by instead using a generic time-out
      named for example <c>open</c>:
    </p>
    <code type="erl"><![CDATA[
...
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
...
    if
        NewButtons =:= Code -> % Correct
	    do_unlock(),
            {next_state, open, Data#{buttons := []},
             [{{timeout,open},10000,lock}]}; % Time in milliseconds
...

open({timeout,open}, lock, Data) ->
    do_lock(),
    {next_state,locked,Data};
open(cast, {button,_}, Data) ->
    {keep_state,Data};
...
]]></code>
    <p>
      Specific generic time-outs can just as
      <seealso marker="#State Time-Outs">State Time-Outs</seealso>
      be restarted or cancelled
      by setting it to a new time or <c>infinity</c>.
    </p>
    <p>
      In this particular case we do not need to cancel the timeout
      since the timeout event is the only possible reason to
      change the state from <c>open</c> to <c>locked</c>.
    </p>
    <p>
      Instead of bothering with when to cancel a time-out,
      a late time-out event can be handled by ignoring it
      if it arrives in a state where it is known to be late.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Erlang Timers" />
    <title>Erlang Timers</title>
    <p>
      The most versatile way to handle time-outs is to use
      Erlang Timers; see
      <seealso marker="erts:erlang#start_timer/4"><c>erlang:start_timer/3,4</c></seealso>.
      Most time-out tasks can be performed with the
      time-out features in <c>gen_statem</c>,
      but an example of one that can not is if you should need
      the return value from
      <seealso marker="erts:erlang#cancel_timer/2"><c>erlang:cancel_timer(Tref)</c></seealso>, that is; the remaining time of the timer.
    </p>
    <p>
      Here is how to accomplish the state time-out
      in the previous example by instead using an Erlang Timer:
    </p>
    <code type="erl"><![CDATA[
...
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
...
    if
        NewButtons =:= Code -> % Correct
	    do_unlock(),
	    Tref =
                 erlang:start_timer(
                     10000, self(), lock), % Time in milliseconds
            {next_state, open, Data#{buttons := [], timer => Tref}};
...

open(info, {timeout,Tref,lock}, #{timer := Tref} = Data) ->
    do_lock(),
    {next_state,locked,maps:remove(timer, Data)};
open(cast, {button,_}, Data) ->
    {keep_state,Data};
...
]]></code>
    <p>
      Removing the <c>timer</c> key from the map when we
      change to state <c>locked</c> is not strictly
      necessary since we can only get into state <c>open</c>
      with an updated <c>timer</c> map value.  But it can be nice
      to not have outdated values in the state <c>Data</c>!
    </p>
    <p>
      If you need to cancel a timer because of some other event, you can use
      <seealso marker="erts:erlang#cancel_timer/2"><c>erlang:cancel_timer(Tref)</c></seealso>.
      Note that a time-out message cannot arrive after this,
      unless you have postponed it before (see the next section),
      so ensure that you do not accidentally postpone such messages.
      Also note that a time-out message may have arrived
      just before you cancelling it, so you may have to read out
      such a message from the process mailbox depending on
      the return value from 
      <seealso marker="erts:erlang#cancel_timer/2"><c>erlang:cancel_timer(Tref)</c></seealso>.
    </p>
    <p>
      Another way to handle a late time-out can be to not cancel it,
      but to ignore it if it arrives in a state
      where it is known to be late.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Postponing Events" />
    <title>Postponing Events</title>
    <p>
      If you want to ignore a particular event in the current state
      and handle it in a future state, you can postpone the event.
      A postponed event is retried after the state has
      changed, that is, <c>OldState =/= NewState</c>.
    </p>
    <p>
      Postponing is ordered by the state transition
      <seealso marker="#State Transition Actions">
	State Transition Action
      </seealso>
      <c>postpone</c>.
    </p>
    <p>
      In this example, instead of ignoring button events
      while in the <c>open</c> state, we can postpone them
      and they are queued and later handled in the <c>locked</c> state:
    </p>
    <code type="erl"><![CDATA[
...
open(cast, {button,_}, Data) ->
    {keep_state,Data,[postpone]};
...
]]></code>
    <p>
      Since a postponed event is only retried after a state change,
      you have to think about where to keep a state data item.
      You can keep it in the server <c>Data</c>
      or in the <c>State</c> itself,
      for example by having two more or less identical states
      to keep a boolean value, or by using a complex state
      (see section
      <seealso marker="#Complex State">Complex State</seealso>)
      with
      <seealso marker="#Callback Modes">Callback Mode</seealso>
      <seealso marker="stdlib:gen_statem#type-callback_mode"><c>handle_event_function</c></seealso>.
      If a change in the value changes the set of events that is handled,
      then the value should be kept in the State.
      Otherwise no postponed events will be retried
      since only the server Data changes.
    </p>
    <p>
      This is not important if you do not postpone events.
      But if you later decide to start postponing some events,
      then the design flaw of not having separate states
      when they should be, might become a hard to find bug.
    </p>

    <section>
      <marker id="Fuzzy State Diagrams" />
      <title>Fuzzy State Diagrams</title>
      <p>
	It is not uncommon that a state diagram does not specify
	how to handle events that are not illustrated
	in a particular state in the diagram.
	Hopefully this is described in an associated text
	or from the context.
      </p>
      <p>
	Possible actions: ignore as in drop the event
	(maybe log it) or deal with the event in some other state
	as in postpone it.
      </p>
    </section>

    <section>
      <marker id="Selective Receive" />
      <title>Selective Receive</title>
      <p>
        Erlang's selective receive statement is often used to
        describe simple state machine examples in straightforward
        Erlang code. The following is a possible implementation of
        the first example:
      </p>
    <code type="erl"><![CDATA[
-module(code_lock).
-define(NAME, code_lock_1).
-export([start_link/1,button/1]).

start_link(Code) ->
    spawn(
      fun () ->
	      true = register(?NAME, self()),
	      do_lock(),
	      locked(Code, length(Code), [])
      end).

button(Button) ->
    ?NAME ! {button,Button}.
    ]]></code>
    <code type="erl"><![CDATA[
locked(Code, Length, Buttons) ->
    receive
        {button,Button} ->
            NewButtons =
                if
                    length(Buttons) < Length ->
                        Buttons;
                    true ->
                        tl(Buttons)
                end ++ [Button],
            if
                NewButtons =:= Code -> % Correct
                    do_unlock(),
		    open(Code, Length);
                true -> % Incomplete | Incorrect
                    locked(Code, Length, NewButtons)
            end
    end.
    ]]></code>
    <code type="erl"><![CDATA[
open(Code, Length) ->
    receive
    after 10000 -> % Time in milliseconds
	    do_lock(),
	    locked(Code, Length, [])
    end.

do_lock() ->
    io:format("Locked~n", []).
do_unlock() ->
    io:format("Open~n", []).
    ]]></code>
    <p>
      The selective receive in this case causes implicitly <c>open</c>
      to postpone any events to the <c>locked</c> state.
    </p>
    <p>
      A selective receive cannot be used from a <c>gen_statem</c>
      behavior as for any <c>gen_*</c> behavior,
      as the receive statement is within the <c>gen_*</c> engine itself.
      It must be there because all
      <seealso marker="stdlib:sys"><c>sys</c></seealso>
      compatible behaviors must respond to system messages and therefore
      do that in their engine receive loop,
      passing non-system messages to the callback module.
    </p>
    <p>
      The
      <seealso marker="#State Transition Actions">
	State Transition Action
      </seealso>
      <c>postpone</c> is designed to model
      selective receives. A selective receive implicitly postpones
      any not received events, but the <c>postpone</c>
      state transition action explicitly postpones one received event.
    </p>
    <p>
      Both mechanisms have the same theoretical
      time and memory complexity, while the selective receive
      language construct has smaller constant factors.
    </p>
    </section>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="State Enter Actions" />
    <title>State Enter Actions</title>
    <p>
      Say you have a state machine specification
      that uses state enter actions.
      Allthough you can code this using inserted events
      (described in the next section), especially if just
      one or a few states has got state enter actions,
      this is a perfect use case for the built in
      <seealso marker="#State Enter Calls">State Enter Calls</seealso>.
    </p>
    <p>
      You return a list containing <c>state_enter</c> from your
      <seealso marker="stdlib:gen_statem#Module:callback_mode/0"><c>callback_mode/0</c></seealso>
      function and the <c>gen_statem</c> engine will call your
      state callback once with the arguments
      <c>(enter, OldState, ...)</c> whenever the state changes.
      Then you just need to handle these event-like calls in all states.
    </p>
    <code type="erl"><![CDATA[
...
init(Code) ->
    process_flag(trap_exit, true),
    Data = #{code => Code, length = length(Code)},
    {ok, locked, Data}.

callback_mode() ->
    [state_functions,state_enter].

locked(enter, _OldState, Data) ->
    do_lock(),
    {keep_state,Data#{buttons => []}};
locked(
  cast, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
...
    if
        NewButtons =:= Code -> % Correct
            {next_state, open, Data};
...

open(enter, _OldState, _Data) ->
    do_unlock(),
    {keep_state_and_data,
     [{state_timeout,10000,lock}]}; % Time in milliseconds
open(state_timeout, lock, Data) ->
    {next_state, locked, Data};
...
]]></code>
    <p>
      You can repeat the state enter code by returning one of
      <c>{repeat_state, ...}</c>, <c>{repeat_state_and_data,_}</c>
      or <c>repeat_state_and_data</c> that otherwise behaves
      exactly like their <c>keep_state</c> siblings.
      See the type
      <seealso marker="stdlib:gen_statem#type-state_callback_result">
	<c>state_callback_result()</c>
      </seealso>
      in the reference manual.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Inserted Events" />
    <title>Inserted Events</title>
    <p>
      It can sometimes be beneficial to be able to generate events
      to your own state machine.
      This can be done with the
      <seealso marker="#State Transition Actions">
	State Transition Action
      </seealso>
      <c>{next_event,EventType,EventContent}</c>.
    </p>
    <p>
      You can generate events of any existing
      <seealso marker="stdlib:gen_statem#type-action">type</seealso>,
      but the <c>internal</c> type can only be generated through action
      <c>next_event</c>. Hence, it cannot come from an external source,
      so you can be certain that an <c>internal</c> event is an event
      from your state machine to itself.
    </p>
    <p>
      One example for this is to pre-process incoming data, for example
      decrypting chunks or collecting characters up to a line break.
    </p>
    <p>
      Purists may argue that this should be modelled with a separate
      state machine that sends pre-processed events
      to the main state machine,
      but to decrease overhead the small pre-processing state machine
      can be implemented in the common state event handling
      of the main state machine using a few state data variables
      that then sends the pre-processed events as internal events
      to the main state machine.
      Using internal events also can make it easier
      to synchronize the state machines.
    </p>
    <p>
      A variant of this is to use a
      <seealso marker="#Complex State">
	Complex State
      </seealso>
      with
      <seealso marker="#One Event Handler">One Event Handler</seealso>.
      The state is then modeled with for example a tuple
      <c>{MainFSMState,SubFSMState}</c>.
    </p>
    <p>
      To illustrate this we make up an example where the buttons
      instead generate down and up (press and release) events,
      and the lock responds to an up event only after
      the corresponding down event.
    </p>
    <code type="erl"><![CDATA[
...
-export([down/1, up/1]).
...
down(Button) ->
    gen_statem:cast(?NAME, {down,Button}).

up(Button) ->
    gen_statem:cast(?NAME, {up,Button}).

...

locked(enter, _OldState, Data) ->
    do_lock(),
    {keep_state,Data#{buttons => []}};
locked(
  internal, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
...
]]></code>
    <code type="erl"><![CDATA[
handle_common(cast, {down,Button}, Data) ->
    {keep_state, Data#{button => Button}};
handle_common(cast, {up,Button}, Data) ->
    case Data of
        #{button := Button} ->
            {keep_state,maps:remove(button, Data),
             [{next_event,internal,{button,Button}}]};
        #{} ->
            keep_state_and_data
    end;
...

open(internal, {button,_}, Data) ->
    {keep_state,Data,[postpone]};
...
]]></code>
    <p>
      If you start this program with <c>code_lock:start([17])</c>
      you can unlock with <c>code_lock:down(17), code_lock:up(17).</c>
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Example Revisited" />
    <title>Example Revisited</title>
    <p>
      This section includes the example after most of the mentioned
      modifications and some more using state enter calls,
      which deserves a new state diagram:
    </p>
    <!-- The image is edited with dia in a .dia file,
	 then exported to Scalable Vector Graphics. -->
    <image file="../design_principles/code_lock_2.svg" width="80%">
      <icaption>Code Lock State Diagram Revisited</icaption>
    </image>
    <p>
      Notice that this state diagram does not specify how to handle
      a button event in the state <c>open</c>. So, you need to
      read in some side notes, that is, here: that unspecified events
      shall be postponed (handled in some later state).
      Also, the state diagram does not show that the <c>code_length/0</c>
      call must be handled in every state.
    </p>

    <section>
      <marker id="Callback Mode: state_functions" />
      <title>Callback Mode: state_functions</title>
      <p>
	Using state functions:
      </p>
      <code type="erl"><![CDATA[
-module(code_lock).
-behaviour(gen_statem).
-define(NAME, code_lock_2).

-export([start_link/1,stop/0]).
-export([down/1,up/1,code_length/0]).
-export([init/1,callback_mode/0,terminate/3]).
-export([locked/3,open/3]).

start_link(Code) ->
    gen_statem:start_link({local,?NAME}, ?MODULE, Code, []).
stop() ->
    gen_statem:stop(?NAME).

down(Button) ->
    gen_statem:cast(?NAME, {down,Button}).
up(Button) ->
    gen_statem:cast(?NAME, {up,Button}).
code_length() ->
    gen_statem:call(?NAME, code_length).
    ]]></code>
      <code type="erl"><![CDATA[
init(Code) ->
    process_flag(trap_exit, true),
    Data = #{code => Code, length => length(Code), buttons => []},
    {ok, locked, Data}.

callback_mode() ->
    [state_functions,state_enter].

-define(HANDLE_COMMON,
    ?FUNCTION_NAME(T, C, D) -> handle_common(T, C, D)).
%%
handle_common(cast, {down,Button}, Data) ->
    {keep_state, Data#{button => Button}};
handle_common(cast, {up,Button}, Data) ->
    case Data of
        #{button := Button} ->
            {keep_state, maps:remove(button, Data),
             [{next_event,internal,{button,Button}}]};
        #{} ->
            keep_state_and_data
    end;
handle_common({call,From}, code_length, #{code := Code}) ->
    {keep_state_and_data,
     [{reply,From,length(Code)}]}.
    ]]></code>
    <code type="erl"><![CDATA[
locked(enter, _OldState, Data) ->
    do_lock(),
    {keep_state, Data#{buttons := []}};
locked(state_timeout, button, Data) ->
    {keep_state, Data#{buttons := []}};
locked(
  internal, {button,Button},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
    NewButtons =
        if
            length(Buttons) < Length ->
                Buttons;
            true ->
                tl(Buttons)
        end ++ [Button],
    if
        NewButtons =:= Code -> % Correct
            {next_state, open, Data};
	true -> % Incomplete | Incorrect
            {keep_state, Data#{buttons := NewButtons},
             [{state_timeout,30000,button}]} % Time in milliseconds
    end;
?HANDLE_COMMON.
]]></code>
    <code type="erl"><![CDATA[
open(enter, _OldState, _Data) ->
    do_unlock(),
    {keep_state_and_data,
     [{state_timeout,10000,lock}]}; % Time in milliseconds
open(state_timeout, lock, Data) ->
    {next_state, locked, Data};
open(internal, {button,_}, _) ->
    {keep_state_and_data, [postpone]};
?HANDLE_COMMON.

do_lock() ->
    io:format("Locked~n", []).
do_unlock() ->
    io:format("Open~n", []).

terminate(_Reason, State, _Data) ->
    State =/= locked andalso do_lock(),
    ok.
    ]]></code>
    </section>

    <section>
      <marker id="Callback Mode: handle_event_function" />
      <title>Callback Mode: handle_event_function</title>
      <p>
        This section describes what to change in the example
        to use one <c>handle_event/4</c> function.
        The previously used approach to first branch depending on event
	does not work that well here because of the state enter calls,
        so this example first branches depending on state:
      </p>
      <code type="erl"><![CDATA[
-export([handle_event/4]).
]]></code>
      <code type="erl"><![CDATA[
callback_mode() ->
    [handle_event_function,state_enter].
    ]]></code>
      <code type="erl"><![CDATA[
%%
%% State: locked
handle_event(enter, _OldState, locked, Data) ->
    do_lock(),
    {keep_state, Data#{buttons := []}};
handle_event(state_timeout, button, locked, Data) ->
    {keep_state, Data#{buttons := []}};
handle_event(
  internal, {button,Button}, locked,
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
    NewButtons =
        if
            length(Buttons) < Length ->
                Buttons;
            true ->
                tl(Buttons)
        end ++ [Button],
    if
        NewButtons =:= Code -> % Correct
            {next_state, open, Data};
	true -> % Incomplete | Incorrect
            {keep_state, Data#{buttons := NewButtons},
             [{state_timeout,30000,button}]} % Time in milliseconds
    end;
    ]]></code>
      <code type="erl"><![CDATA[
%%
%% State: open
handle_event(enter, _OldState, open, _Data) ->
    do_unlock(),
    {keep_state_and_data,
     [{state_timeout,10000,lock}]}; % Time in milliseconds
handle_event(state_timeout, lock, open, Data) ->
    {next_state, locked, Data};
handle_event(internal, {button,_}, open, _) ->
    {keep_state_and_data,[postpone]};
    ]]></code>
      <code type="erl"><![CDATA[
%% Common events
handle_event(cast, {down,Button}, _State, Data) ->
    {keep_state, Data#{button => Button}};
handle_event(cast, {up,Button}, _State, Data) ->
    case Data of
        #{button := Button} ->
            {keep_state, maps:remove(button, Data),
             [{next_event,internal,{button,Button}},
              {state_timeout,30000,button}]}; % Time in milliseconds
        #{} ->
            keep_state_and_data
    end;
handle_event({call,From}, code_length, _State, #{length := Length}) ->
    {keep_state_and_data,
     [{reply,From,Length}]}.
    ]]></code>
    </section>
    <p>
      Notice that postponing buttons from the <c>open</c> state
      to the <c>locked</c> state feels like a strange thing to do
      for a code lock, but it at least illustrates event postponing.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Filter the State" />
    <title>Filter the State</title>
    <p>
      The example servers so far in this chapter
      print the full internal state in the error log, for example,
      when killed by an exit signal or because of an internal error.
      This state contains both the code lock code
      and which digits that remain to unlock.
    </p>
    <p>
      This state data can be regarded as sensitive,
      and maybe not what you want in the error log
      because of some unpredictable event.
    </p>
    <p>
      Another reason to filter the state can be
      that the state is too large to print, as it fills
      the error log with uninteresting details.
    </p>
    <p>
      To avoid this, you can format the internal state
      that gets in the error log and gets returned from
      <seealso marker="stdlib:sys#get_status/1"><c>sys:get_status/1,2</c></seealso>
      by implementing function
      <seealso marker="stdlib:gen_statem#Module:format_status/2"><c>Module:format_status/2</c></seealso>,
      for example like this:
    </p>
    <code type="erl"><![CDATA[
...
-export([init/1,terminate/3,format_status/2]).
...

format_status(Opt, [_PDict,State,Data]) ->
    StateData =
	{State,
	 maps:filter(
	   fun (code, _) -> false;
	       (_, _) -> true
	   end,
	   Data)},
    case Opt of
	terminate ->
	    StateData;
	normal ->
	    [{data,[{"State",StateData}]}]
    end.
    ]]></code>
    <p>
      It is not mandatory to implement a
      <seealso marker="stdlib:gen_statem#Module:format_status/2"><c>Module:format_status/2</c></seealso>
      function. If you do not, a default implementation is used that
      does the same as this example function without filtering
      the <c>Data</c> term, that is, <c>StateData = {State,Data}</c>,
      in this example containing sensitive information.
    </p>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Complex State" />
    <title>Complex State</title>
    <p>
      The callback mode
      <seealso marker="stdlib:gen_statem#type-callback_mode"><c>handle_event_function</c></seealso>
      enables using a non-atom state as described in section
      <seealso marker="#Callback Modes">Callback Modes</seealso>,
      for example, a complex state term like a tuple.
    </p>
    <p>
      One reason to use this is when you have a state item
      that when changed should cancel the
      <seealso marker="#State Time-Outs">State Time-Out</seealso>,
      or one that affects the event handling
      in combination with postponing events.
      We will go for the latter and complicate the previous example
      by introducing a configurable lock button
      (this is the state item in question),
      which in the <c>open</c> state immediately locks the door,
      and an API function <c>set_lock_button/1</c> to set the lock button.
    </p>
    <p>
      Suppose now that we call <c>set_lock_button</c>
      while the door is open,
      and we have already postponed a button event
      that was the new lock button:
    </p>
    <code type="erl"><![CDATA[
1> code_lock:start_link([a,b,c], x).
{ok,<0.666.0>}
2> code_lock:button(a).
ok
3> code_lock:button(b).
ok
4> code_lock:button(c).
ok
Open
5> code_lock:button(y).
ok
6> code_lock:set_lock_button(y).
x
% What should happen here?  Immediate lock or nothing?
]]></code>
    <p>
      We could say that the button was pressed too early
      so it is not to be recognized as the lock button.
      Or we can make the lock button part of the state so
      when we then change the lock button in the locked state,
      the change becomes a state change
      and all postponed events are retried,
      therefore the lock is immediately locked!
    </p>
    <p>
      We define the state as <c>{StateName,LockButton}</c>,
      where <c>StateName</c> is as before
      and <c>LockButton</c> is the current lock button:
    </p>
    <code type="erl"><![CDATA[
-module(code_lock).
-behaviour(gen_statem).
-define(NAME, code_lock_3).

-export([start_link/2,stop/0]).
-export([button/1,set_lock_button/1]).
-export([init/1,callback_mode/0,terminate/3]).
-export([handle_event/4]).

start_link(Code, LockButton) ->
    gen_statem:start_link(
        {local,?NAME}, ?MODULE, {Code,LockButton}, []).
stop() ->
    gen_statem:stop(?NAME).

button(Button) ->
    gen_statem:cast(?NAME, {button,Button}).
set_lock_button(LockButton) ->
    gen_statem:call(?NAME, {set_lock_button,LockButton}).
    ]]></code>
    <code type="erl"><![CDATA[
init({Code,LockButton}) ->
    process_flag(trap_exit, true),
    Data = #{code => Code, length => length(Code), buttons => []},
    {ok, {locked,LockButton}, Data}.

callback_mode() ->
    [handle_event_function,state_enter].

%% State: locked
handle_event(enter, _OldState, {locked,_}, Data) ->
    do_lock(),
    {keep_state, Data#{buttons := []}};
handle_event(state_timeout, button, {locked,_}, Data) ->
    {keep_state, Data#{buttons := []}};
handle_event(
  cast, {button,Button}, {locked,LockButton},
  #{code := Code, length := Length, buttons := Buttons} = Data) ->
    NewButtons =
        if
            length(Buttons) < Length ->
                Buttons;
            true ->
                tl(Buttons)
        end ++ [Button],
    if
        NewButtons =:= Code -> % Correct
            {next_state, {open,LockButton}, Data};
	true -> % Incomplete | Incorrect
            {keep_state, Data#{buttons := NewButtons},
             [{state_timeout,30000,button}]} % Time in milliseconds
    end;
    ]]></code>
    <code type="erl"><![CDATA[
%%
%% State: open
handle_event(enter, _OldState, {open,_}, _Data) ->
    do_unlock(),
    {keep_state_and_data,
     [{state_timeout,10000,lock}]}; % Time in milliseconds
handle_event(state_timeout, lock, {open,LockButton}, Data) ->
    {next_state, {locked,LockButton}, Data};
handle_event(cast, {button,LockButton}, {open,LockButton}, Data) ->
    {next_state, {locked,LockButton}, Data};
handle_event(cast, {button,_}, {open,_}, _Data) ->
    {keep_state_and_data,[postpone]};
    ]]></code>
    <code type="erl"><![CDATA[
%%
%% Common events
handle_event(
  {call,From}, {set_lock_button,NewLockButton},
  {StateName,OldLockButton}, Data) ->
    {next_state, {StateName,NewLockButton}, Data,
     [{reply,From,OldLockButton}]}.
    ]]></code>
    <code type="erl"><![CDATA[
do_lock() ->
    io:format("Locked~n", []).
do_unlock() ->
    io:format("Open~n", []).

terminate(_Reason, State, _Data) ->
    State =/= locked andalso do_lock(),
    ok.
    ]]></code>
  </section>

<!-- =================================================================== -->

  <section>
    <marker id="Hibernation" />
    <title>Hibernation</title>
    <p>
      If you have many servers in one node
      and they have some state(s) in their lifetime in which
      the servers can be expected to idle for a while,
      and the amount of heap memory all these servers need
      is a problem, then the memory footprint of a server
      can be mimimized by hibernating it through
      <seealso marker="stdlib:proc_lib#hibernate/3"><c>proc_lib:hibernate/3</c></seealso>.
    </p>
    <note>
      <p>
        It is rather costly to hibernate a process; see
        <seealso marker="erts:erlang#hibernate/3"><c>erlang:hibernate/3</c></seealso>.
        It is not something you want to do after every event.
      </p>
    </note>
    <p>
      We can in this example hibernate in the <c>{open,_}</c> state,
      because what normally occurs in that state is that
      the state time-out after a while
      triggers a transition to <c>{locked,_}</c>:
    </p>
    <code type="erl"><![CDATA[
...
%%
%% State: open
handle_event(enter, _OldState, {open,_}, _Data) ->
    do_unlock(),
    {keep_state_and_data,
     [{state_timeout,10000,lock}, % Time in milliseconds
      hibernate]};
...
]]></code>
    <p>
      The atom
      <seealso marker="stdlib:gen_statem#type-hibernate"><c>hibernate</c></seealso>
      in the action list on the last line
      when entering the <c>{open,_}</c> state is the only change.
      If any event arrives in the <c>{open,_},</c> state, we
      do not bother to rehibernate, so the server stays
      awake after any event.
    </p>
    <p>
      To change that we would need to insert
      action <c>hibernate</c> in more places.
      For example, the state-independent <c>set_lock_button</c>
      operation would have to use <c>hibernate</c> but only in the
      <c>{open,_}</c> state, which would clutter the code.
    </p>
    <p>
      Another not uncommon scenario is to use the
      <seealso marker="#Event Time-Outs">Event Time-Out</seealso>
      to trigger hibernation after a certain time of inactivity.
      There is also a server start option
      <seealso marker="stdlib:gen_statem#type-hibernate_after_opt">
	<c>{hibernate_after, Timeout}</c>
      </seealso>
      for
      <seealso marker="stdlib:gen_statem#start/3">
	<c>start/3,4</c>
      </seealso>
      or
      <seealso marker="stdlib:gen_statem#start_link/3">
	<c>start_link/3,4</c>
      </seealso>
      that may be used to automatically hibernate the server.
    </p>
    <p>
      This particular server probably does not use
      heap memory worth hibernating for.
      To gain anything from hibernation, your server would
      have to produce non-insignificant garbage
      during callback execution,
      for which this example server can serve as a bad example.
    </p>
  </section>

</chapter>