Erlang logo
User's Guide
PDF
Top

OTP Design Principles
User's Guide
Version 7.3


Expand All
Contract All

Chapters

3 gen_fsm Behaviour

This section is to be read with the gen_fsm(3) manual page in STDLIB, where all interface functions and callback functions are described in detail.

3.1  Finite-State Machines

A Finite-State Machine (FSM) can be described as a set of relations of the form:

State(S) x Event(E) -> Actions(A), State(S')

These relations are interpreted as meaning:

If we are in state S and event E occurs, we are to perform actions A and make a transition to state S'.

For an FSM implemented using the gen_fsm behaviour, the state transition rules are written as a number of Erlang functions, which conform to the following convention:

StateName(Event, StateData) ->
    .. code for actions here ...
    {next_state, StateName', StateData'}

3.2  Example

A door with a code lock can be viewed as an FSM. Initially, the door is locked. Anytime someone presses a button, this generates an event. Depending on what buttons have been pressed before, the sequence so far can be correct, incomplete, or wrong.

If it is correct, the door is unlocked for 30 seconds (30,000 ms). If it is incomplete, we wait for another button to be pressed. If it is is wrong, we start all over, waiting for a new button sequence.

Implementing the code lock FSM using gen_fsm results in the following callback module:

-module(code_lock).
-behaviour(gen_fsm).

-export([start_link/1]).
-export([button/1]).
-export([init/1, locked/2, open/2]).

start_link(Code) ->
    gen_fsm:start_link({local, code_lock}, code_lock, lists:reverse(Code), []).

button(Digit) ->
    gen_fsm:send_event(code_lock, {button, Digit}).

init(Code) ->
    {ok, locked, {[], Code}}.

locked({button, Digit}, {SoFar, Code}) ->
    case [Digit|SoFar] of
        Code ->
            do_unlock(),
            {next_state, open, {[], Code}, 30000};
        Incomplete when length(Incomplete)<length(Code) ->
            {next_state, locked, {Incomplete, Code}};
        _Wrong ->
            {next_state, locked, {[], Code}}
    end.

open(timeout, State) ->
    do_lock(),
    {next_state, locked, State}.

The code is explained in the next sections.

3.3  Starting gen_fsm

In the example in the previous section, the gen_fsm is started by calling code_lock:start_link(Code):

start_link(Code) ->
    gen_fsm:start_link({local, code_lock}, code_lock, lists:reverse(Code), []).
    

start_link calls the function gen_fsm:start_link/4, which spawns and links to a new process, a gen_fsm.

  • The first argument, {local, code_lock}, specifies the name. In this case, the gen_fsm is locally registered as code_lock.

    If the name is omitted, the gen_fsm is not registered. Instead its pid must be used. The name can also be given as {global, Name}, in which case the gen_fsm is registered using global:register_name/2.

  • The second argument, code_lock, is the name of the callback module, that is, the module where the callback functions are located.

    The interface functions (start_link and button) are then located in the same module as the callback functions (init, locked, and open). This is normally good programming practice, to have the code corresponding to one process contained in one module.

  • The third argument, Code, is a list of digits that which is passed reversed to the callback function init. Here, init gets the correct code for the lock as indata.

  • The fourth argument, [], is a list of options. See the gen_fsm(3) manual page for available options.

If name registration succeeds, the new gen_fsm process calls the callback function code_lock:init(Code). This function is expected to return {ok, StateName, StateData}, where StateName is the name of the initial state of the gen_fsm. In this case locked, assuming the door is locked to begin with. StateData is the internal state of the gen_fsm. (For gen_fsm, the internal state is often referred to 'state data' to distinguish it from the state as in states of a state machine.) In this case, the state data is the button sequence so far (empty to begin with) and the correct code of the lock.

init(Code) ->
    {ok, locked, {[], Code}}.

gen_fsm:start_link is synchronous. It does not return until the gen_fsm has been initialized and is ready to receive notifications.

gen_fsm:start_link must be used if the gen_fsm is part of a supervision tree, that is, started by a supervisor. There is another function, gen_fsm:start, to start a standalone gen_fsm, that is, a gen_fsm that is not part of a supervision tree.

3.4  Notifying about Events

The function notifying the code lock about a button event is implemented using gen_fsm:send_event/2:

button(Digit) ->
    gen_fsm:send_event(code_lock, {button, Digit}).

code_lock is the name of the gen_fsm and must agree with the name used to start it. {button, Digit} is the actual event.

The event is made into a message and sent to the gen_fsm. When the event is received, the gen_fsm calls StateName(Event, StateData), which is expected to return a tuple {next_state,StateName1,StateData1}. StateName is the name of the current state and StateName1 is the name of the next state to go to. StateData1 is a new value for the state data of the gen_fsm.

locked({button, Digit}, {SoFar, Code}) ->
    case [Digit|SoFar] of
        Code ->
            do_unlock(),
            {next_state, open, {[], Code}, 30000};
        Incomplete when length(Incomplete)<length(Code) ->
            {next_state, locked, {Incomplete, Code}};
        _Wrong ->
            {next_state, locked, {[], Code}};
    end.

open(timeout, State) ->
    do_lock(),
    {next_state, locked, State}.

If the door is locked and a button is pressed, the complete button sequence so far is compared with the correct code for the lock and, depending on the result, the door is either unlocked and the gen_fsm goes to state open, or the door remains in state locked.

3.5  Time-Outs

When a correct code has been given, the door is unlocked and the following tuple is returned from locked/2:

{next_state, open, {[], Code}, 30000};

30,000 is a time-out value in milliseconds. After this time, that is, 30 seconds, a time-out occurs. Then, StateName(timeout, StateData) is called. The time-out then occurs when the door has been in state open for 30 seconds. After that the door is locked again:

open(timeout, State) ->
    do_lock(),
    {next_state, locked, State}.

3.6  All State Events

Sometimes an event can arrive at any state of the gen_fsm. Instead of sending the message with gen_fsm:send_event/2 and writing one clause handling the event for each state function, the message can be sent with gen_fsm:send_all_state_event/2 and handled with Module:handle_event/3:

-module(code_lock).
...
-export([stop/0]).
...

stop() ->
    gen_fsm:send_all_state_event(code_lock, stop).

...

handle_event(stop, _StateName, StateData) ->
    {stop, normal, StateData}.

3.7  Stopping

In a Supervision Tree

If the gen_fsm is part of a supervision tree, no stop function is needed. The gen_fsm is automatically terminated by its supervisor. Exactly how this is done is defined by a shutdown strategy set in the supervisor.

If it is necessary to clean up before termination, the shutdown strategy must be a time-out value and the gen_fsm must be set to trap exit signals in the init function. When ordered to shutdown, the gen_fsm then calls the callback function terminate(shutdown, StateName, StateData):

init(Args) ->
    ...,
    process_flag(trap_exit, true),
    ...,
    {ok, StateName, StateData}.

...

terminate(shutdown, StateName, StateData) ->
    ..code for cleaning up here..
    ok.

Standalone gen_fsm

If the gen_fsm is not part of a supervision tree, a stop function can be useful, for example:

...
-export([stop/0]).
...

stop() ->
    gen_fsm:send_all_state_event(code_lock, stop).
...

handle_event(stop, _StateName, StateData) ->
    {stop, normal, StateData}.

...

terminate(normal, _StateName, _StateData) ->
    ok.

The callback function handling the stop event returns a tuple, {stop,normal,StateData1}, where normal specifies that it is a normal termination and StateData1 is a new value for the state data of the gen_fsm. This causes the gen_fsm to call terminate(normal,StateName,StateData1) and then it terminates gracefully:

3.8  Handling Other Messages

If the gen_fsm is to be able to receive other messages than events, the callback function handle_info(Info, StateName, StateData) must be implemented to handle them. Examples of other messages are exit messages, if the gen_fsm is linked to other processes (than the supervisor) and trapping exit signals.

handle_info({'EXIT', Pid, Reason}, StateName, StateData) ->
    ..code to handle exits here..
    {next_state, StateName1, StateData1}.

The code_change method must also be implemented.

code_change(OldVsn, StateName, StateData, Extra) ->
    ..code to convert state (and more) during code change
    {ok, NextStateName, NewStateData}