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

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

<chapter>
  <header>
    <copyright>
      <year>1997</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>sys and proc_lib</title>
    <prepared></prepared>
    <docno></docno>
    <date></date>
    <rev></rev>
    <file>spec_proc.xml</file>
  </header>
  <marker id="sys and proc_lib"></marker>
    <p>The <c>sys</c> module has functions for simple debugging of
      processes implemented using behaviours. It also has functions that,
      together with functions in the <c>proc_lib</c> module, can be used
      to implement a <em>special process</em> that complies to the OTP
      design principles without using a standard behaviour. These
      functions can also be used to implement user-defined (non-standard)
      behaviours.</p>
    <p>Both <c>sys</c> and <c>proc_lib</c> belong to the STDLIB
      application.</p>

  <section>
    <title>Simple Debugging</title>
    <p>The <c>sys</c> module has functions for simple debugging of
    processes implemented using behaviours. The <c>code_lock</c>
    example from
    <seealso marker="statem#Example">gen_statem Behaviour</seealso>
    is used to illustrate this:</p>
    <pre>
Erlang/OTP 20 [DEVELOPMENT] [erts-9.0] [source-5ace45e] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:10] [hipe] [kernel-poll:false]

Eshell V9.0  (abort with ^G)
1>  code_lock:start_link([1,2,3,4]).
Lock
{ok,&lt;0.63.0>}
2> sys:statistics(code_lock, true).
ok
3>  sys:trace(code_lock, true).
ok
4>  code_lock:button(1).
*DBG* code_lock receive cast {button,1} in state locked
ok
*DBG* code_lock consume cast {button,1} in state locked
5>  code_lock:button(2).
*DBG* code_lock receive cast {button,2} in state locked
ok
*DBG* code_lock consume cast {button,2} in state locked
6>  code_lock:button(3).
*DBG* code_lock receive cast {button,3} in state locked
ok
*DBG* code_lock consume cast {button,3} in state locked
7>  code_lock:button(4).
*DBG* code_lock receive cast {button,4} in state locked
ok
Unlock
*DBG* code_lock consume cast {button,4} in state locked
*DBG* code_lock receive state_timeout lock in state open
Lock
*DBG* code_lock consume state_timeout lock in state open
8> sys:statistics(code_lock, get).
{ok,[{start_time,{{2017,4,21},{16,8,7}}},
     {current_time,{{2017,4,21},{16,9,42}}},
     {reductions,2973},
     {messages_in,5},
     {messages_out,0}]}
9> sys:statistics(code_lock, false).
ok
10> sys:trace(code_lock, false).
ok
11> sys:get_status(code_lock).
{status,&lt;0.63.0>,
        {module,gen_statem},
        [[{'$initial_call',{code_lock,init,1}},
          {'$ancestors',[&lt;0.61.0>]}],
         running,&lt;0.61.0>,[],
         [{header,"Status for state machine code_lock"},
          {data,[{"Status",running},
                 {"Parent",&lt;0.61.0>},
                 {"Logged Events",[]},
                 {"Postponed",[]}]},
          {data,[{"State",
                  {locked,#{code => [1,2,3,4],remaining => [1,2,3,4]}}}]}]]}
    </pre>
  </section>

  <section>
    <title>Special Processes</title>
    <p>This section describes how to write a process that complies to
    the OTP design principles, without using a standard behaviour.
    Such a process is to:</p>
    <list type="bulleted">
      <item>Be started in a way that makes the process fit into a
      supervision tree</item>
      <item>Support the <c>sys</c>
      <seealso marker="#debug">debug facilities</seealso></item>
      <item>Take care of
      <seealso marker="#msg">system messages</seealso>.</item>
    </list>
    <p>System messages are messages with a special meaning, used in
      the supervision tree. Typical system messages are requests for
      trace output, and requests to suspend or resume process execution
      (used during release handling). Processes implemented using
      standard behaviours automatically understand these messages.</p>

    <section>
      <title>Example</title>
      <p>The simple server from
      <seealso marker="des_princ#ch1">Overview</seealso>,
      implemented using <c>sys</c> and <c>proc_lib</c> so it fits into a
      supervision tree:</p>
      <marker id="ex"></marker>
      <pre>
-module(ch4).
-export([start_link/0]).
-export([alloc/0, free/1]).
-export([init/1]).
-export([system_continue/3, system_terminate/4,
         write_debug/3,
         system_get_state/1, system_replace_state/2]).

start_link() ->
    proc_lib:start_link(ch4, init, [self()]).

alloc() ->
    ch4 ! {self(), alloc},
    receive
        {ch4, Res} ->
            Res
    end.

free(Ch) ->
    ch4 ! {free, Ch},
    ok.

init(Parent) ->
    register(ch4, self()),
    Chs = channels(),
    Deb = sys:debug_options([]),
    proc_lib:init_ack(Parent, {ok, self()}),
    loop(Chs, Parent, Deb).

loop(Chs, Parent, Deb) ->
    receive
        {From, alloc} ->
            Deb2 = sys:handle_debug(Deb, fun ch4:write_debug/3,
                                    ch4, {in, alloc, From}),
            {Ch, Chs2} = alloc(Chs),
            From ! {ch4, Ch},
            Deb3 = sys:handle_debug(Deb2, fun ch4:write_debug/3,
                                    ch4, {out, {ch4, Ch}, From}),
            loop(Chs2, Parent, Deb3);
        {free, Ch} ->
            Deb2 = sys:handle_debug(Deb, fun ch4:write_debug/3,
                                    ch4, {in, {free, Ch}}),
            Chs2 = free(Ch, Chs),
            loop(Chs2, Parent, Deb2);

        {system, From, Request} ->
            sys:handle_system_msg(Request, From, Parent,
                                  ch4, Deb, Chs)
    end.

system_continue(Parent, Deb, Chs) ->
    loop(Chs, Parent, Deb).

system_terminate(Reason, _Parent, _Deb, _Chs) ->
    exit(Reason).

system_get_state(Chs) ->
    {ok, Chs}.

system_replace_state(StateFun, Chs) ->
    NChs = StateFun(Chs),
    {ok, NChs, NChs}.

write_debug(Dev, Event, Name) ->
    io:format(Dev, "~p event = ~p~n", [Name, Event]).</pre>
      <p>Example on how the simple debugging functions in the <c>sys</c>
      module can also be used for <c>ch4</c>:</p>
      <pre>
% <input>erl</input>
Erlang (BEAM) emulator version 5.2.3.6 [hipe] [threads:0]

Eshell V5.2.3.6  (abort with ^G)
1> <input>ch4:start_link().</input>
{ok,&lt;0.30.0>}
2> <input>sys:statistics(ch4, true).</input>
ok
3> <input>sys:trace(ch4, true).</input>
ok
4> <input>ch4:alloc().</input>
ch4 event = {in,alloc,&lt;0.25.0>}
ch4 event = {out,{ch4,ch1},&lt;0.25.0>}
ch1
5> <input>ch4:free(ch1).</input>
ch4 event = {in,{free,ch1}}
ok
6> <input>sys:statistics(ch4, get).</input>
{ok,[{start_time,{{2003,6,13},{9,47,5}}},
     {current_time,{{2003,6,13},{9,47,56}}},
     {reductions,109},
     {messages_in,2},
     {messages_out,1}]}
7> <input>sys:statistics(ch4, false).</input>
ok
8> <input>sys:trace(ch4, false).</input>
ok
9> <input>sys:get_status(ch4).</input>
{status,&lt;0.30.0>,
        {module,ch4},
        [[{'$ancestors',[&lt;0.25.0>]},{'$initial_call',{ch4,init,[&lt;0.25.0>]}}],
         running,&lt;0.25.0>,[],
         [ch1,ch2,ch3]]}</pre>
    </section>

    <section>
      <title>Starting the Process</title>
      <p>A function in the <c>proc_lib</c> module is to be used to
        start the process. Several functions are available, for
        example, <c>spawn_link/3,4</c> for asynchronous start and
        <c>start_link/3,4,5</c> for synchronous start.</p>
      <p>A process started using one of these functions stores
        information (for example, about the ancestors and initial call)
	that is needed for a process in a supervision tree.</p>
      <p>If the process terminates with another reason than
        <c>normal</c> or <c>shutdown</c>, a crash report is generated.
	For more information about the crash report, see the SASL
        User's Guide.</p>
      <p>In the example, synchronous start is used. The process
        starts by calling <c>ch4:start_link()</c>:</p>
      <code type="none">
start_link() ->
    proc_lib:start_link(ch4, init, [self()]).</code>
      <p><c>ch4:start_link</c> calls the function
        <c>proc_lib:start_link</c>. This function takes a module name,
        a function name, and an argument list as arguments, spawns,
        and links to a new process. The new process starts by executing
        the given function, here <c>ch4:init(Pid)</c>, where
        <c>Pid</c> is the pid (<c>self()</c>) of the first process,
	which is the parent process.</p>
      <p>All initialization, including name registration, is done in
      <c>init</c>. The new process must also acknowledge that it has
      been started to the parent:</p>
      <code type="none">
init(Parent) ->
    ...
    proc_lib:init_ack(Parent, {ok, self()}),
    loop(...).</code>
      <p><c>proc_lib:start_link</c> is synchronous and does not return
        until <c>proc_lib:init_ack</c> has been called.</p>
    </section>

    <section>
      <marker id="debug"></marker>
      <title>Debugging</title>
      <p>To support the debug facilites in <c>sys</c>, a
        <em>debug structure</em> is needed. The <c>Deb</c> term is
        initialized using <c>sys:debug_options/1</c>:</p>
      <code type="none">
init(Parent) ->
    ...
    Deb = sys:debug_options([]),
    ...
    loop(Chs, Parent, Deb).</code>
      <p><c>sys:debug_options/1</c> takes a list of options as argument.
        Here the list is empty, which means no debugging is enabled
        initially. For information about the possible options, see the
        <c>sys(3)</c> manual page in STDLIB.</p>
      <p>Then, for each <em>system event</em> to be logged
        or traced, the following function is to be called.</p>
      <code type="none">
sys:handle_debug(Deb, Func, Info, Event) => Deb1</code>
<p>Here:</p>
      <list type="bulleted">
        <item><c>Deb</c> is the debug structure.</item>
        <item><c>Func</c> is a fun specifying
            a (user-defined) function used to format
            trace output. For each system event, the format function is
            called as <c>Func(Dev, Event, Info)</c>, where:
          <list type="bulleted">
            <item><c>Dev</c> is the I/O device to which the output is to
	      be printed. See the <c>io(3)</c> manual page in
	      STDLIB.</item>
            <item><c>Event</c> and <c>Info</c> are passed as is from
                <c>handle_debug</c>.</item>
          </list>
        </item>
        <item><c>Info</c> is used to pass more information to
            <c>Func</c>. It can be any term and is passed as is.</item>
        <item><c>Event</c> is the system event. It is up to the user to
            define what a system event is and how it is to be
            represented. Typically at least incoming and outgoing
            messages are considered system events and represented by
            the tuples <c>{in,Msg[,From]}</c> and <c>{out,Msg,To[,State]}</c>,
            respectively.</item>
      </list>
      <p><c>handle_debug</c> returns an updated debug structure
        <c>Deb1</c>.</p>
      <p>In the example, <c>handle_debug</c> is called for each incoming
        and outgoing message. The format function <c>Func</c> is
        the function <c>ch4:write_debug/3</c>, which prints the message
        using <c>io:format/3</c>.</p>
      <code type="none">
loop(Chs, Parent, Deb) ->
    receive
        {From, alloc} ->
            Deb2 = sys:handle_debug(Deb, fun ch4:write_debug/3,
                                    ch4, {in, alloc, From}),
            {Ch, Chs2} = alloc(Chs),
            From ! {ch4, Ch},
            Deb3 = sys:handle_debug(Deb2, fun ch4:write_debug/3,
                                    ch4, {out, {ch4, Ch}, From}),
            loop(Chs2, Parent, Deb3);
        {free, Ch} ->
            Deb2 = sys:handle_debug(Deb, fun ch4:write_debug/3,
                                    ch4, {in, {free, Ch}}),
            Chs2 = free(Ch, Chs),
            loop(Chs2, Parent, Deb2);
        ...
    end.

write_debug(Dev, Event, Name) ->
    io:format(Dev, "~p event = ~p~n", [Name, Event]).</code>
    </section>

    <section>
      <marker id="msg"></marker>
      <title>Handling System Messages</title>
      <p><em>System messages</em> are received as:</p>
      <code type="none">
{system, From, Request}</code>
      <p>The content and meaning of these messages do not need to be
        interpreted by the process. Instead the following function
        is to be called:</p>
      <code type="none">
sys:handle_system_msg(Request, From, Parent, Module, Deb, State)</code>
      <p>This function does not return. It handles the system
      message and then either calls the following if process execution is
      to continue:</p>
      <code type="none">
Module:system_continue(Parent, Deb, State)</code>
      <p>Or calls the following if the process is to terminate:</p>
      <code type="none">
Module:system_terminate(Reason, Parent, Deb, State)</code>
      <p>A process in a supervision tree is expected to terminate with
      the same reason as its parent.</p>
      <list type="bulleted">
        <item><c>Request</c> and <c>From</c> are to be passed as is from
	the system message to the call to <c>handle_system_msg</c>.</item>
        <item><c>Parent</c> is the pid of the parent.</item>
        <item><c>Module</c> is the name of the module.</item>
        <item><c>Deb</c> is the debug structure.</item>
        <item><c>State</c> is a term describing the internal state and
         is passed to <c>system_continue</c>/<c>system_terminate</c>/
         <c>system_get_state</c>/<c>system_replace_state</c>.</item>
      </list>
      <p>If the process is to return its state, <c>handle_system_msg</c>
      calls:</p>
      <code type="none">
Module:system_get_state(State)</code>
      <p>If the process is to replace its state using the fun <c>StateFun</c>,
      <c>handle_system_msg</c> calls:</p>
      <code type="none">
Module:system_replace_state(StateFun, State)</code>
      <p>In the example:</p>
      <code type="none">
loop(Chs, Parent, Deb) ->
    receive
        ...

        {system, From, Request} ->
            sys:handle_system_msg(Request, From, Parent,
                                  ch4, Deb, Chs)
    end.

system_continue(Parent, Deb, Chs) ->
    loop(Chs, Parent, Deb).

system_terminate(Reason, Parent, Deb, Chs) ->
    exit(Reason).

system_get_state(Chs) ->
    {ok, Chs, Chs}.

system_replace_state(StateFun, Chs) ->
    NChs = StateFun(Chs),
    {ok, NChs, NChs}.
</code>
      <p>If the special process is set to trap exits and if the parent
      process terminates, the expected behavior is to terminate with
      the same reason:</p>
      <code type="none">
init(...) ->
    ...,
    process_flag(trap_exit, true),
    ...,
    loop(...).

loop(...) ->
    receive
        ...

        {'EXIT', Parent, Reason} ->
            ..maybe some cleaning up here..
            exit(Reason);
        ...
    end.</code>
    </section>
  </section>

  <section>
    <title>User-Defined Behaviours</title>
    <p><marker id="behaviours"/>To implement a user-defined behaviour,
      write code similar to
      code for a special process, but call functions in a callback
      module for handling specific tasks.</p>
    <p>If the compiler is to warn for missing callback functions, as it
    does for the OTP behaviours, add <c>-callback</c> attributes in the
    behaviour module to describe the expected callbacks:</p>
    <code type="none">
-callback Name1(Arg1_1, Arg1_2, ..., Arg1_N1) -> Res1.
-callback Name2(Arg2_1, Arg2_2, ..., Arg2_N2) -> Res2.
...
-callback NameM(ArgM_1, ArgM_2, ..., ArgM_NM) -> ResM.</code>
    <p><c>NameX</c> are the names of the expected callbacks.
      <c>ArgX_Y</c> and <c>ResX</c> are types as they are described in
      <seealso marker="../reference_manual/typespec">Types and
      Function Specifications</seealso>. The whole syntax of the <c>-spec</c>
      attribute is supported by the <c>-callback</c> attribute.</p>
    <p>Callback functions that are optional for the user of the
      behaviour to implement are specified by use of the
      <c>-optional_callbacks</c> attribute:</p>

<code type="none">
-optional_callbacks([OptName1/OptArity1, ..., OptNameK/OptArityK]).</code>

    <p>where each <c>OptName/OptArity</c> specifies the name and arity
      of a callback function. Note that the <c>-optional_callbacks</c>
      attribute is to be used together with the <c>-callback</c>
      attribute; it cannot be combined with the
      <c>behaviour_info()</c> function described below.</p>
    <p>Tools that need to know about optional callback functions can
      call <c>Behaviour:behaviour_info(optional_callbacks)</c> to get
      a list of all optional callback functions.</p>

    <note><p>We recommend using the <c>-callback</c> attribute rather
      than the <c>behaviour_info()</c> function. The reason is that
      the extra type information can be used by tools to produce
      documentation or find discrepancies.</p></note>

    <p>As an alternative to the <c>-callback</c> and
      <c>-optional_callbacks</c> attributes you may directly implement
      and export <c>behaviour_info()</c>:</p>

    <code type="none">
behaviour_info(callbacks) ->
    [{Name1, Arity1},...,{NameN, ArityN}].</code>

    <p>where each <c>{Name, Arity}</c> specifies the name and arity of
      a callback function. This function is otherwise automatically
      generated by the compiler using the <c>-callback</c>
      attributes.</p>
    <p>When the compiler encounters the module attribute
      <c>-behaviour(Behaviour).</c> in a module <c>Mod</c>, it
      calls <c>Behaviour:behaviour_info(callbacks)</c> and compares the
      result with the set of functions actually exported from
      <c>Mod</c>, and issues a warning if any callback function is
      missing.</p>
    <p>Example:</p>
    <code type="none">
%% User-defined behaviour module
-module(simple_server).
-export([start_link/2, init/3, ...]).

-callback init(State :: term()) -> 'ok'.
-callback handle_req(Req :: term(), State :: term()) -> {'ok', Reply :: term()}.
-callback terminate() -> 'ok'.
-callback format_state(State :: term()) -> term().

-optional_callbacks([format_state/1]).

%% Alternatively you may define:
%%
%% -export([behaviour_info/1]).
%% behaviour_info(callbacks) ->
%%     [{init,1},
%%      {handle_req,2},
%%      {terminate,0}].

start_link(Name, Module) ->
    proc_lib:start_link(?MODULE, init, [self(), Name, Module]).

init(Parent, Name, Module) ->
    register(Name, self()),
    ...,
    Dbg = sys:debug_options([]),
    proc_lib:init_ack(Parent, {ok, self()}),
    loop(Parent, Module, Deb, ...).

...</code>
    <p>In a callback module:</p>
    <code type="none">
-module(db).
-behaviour(simple_server).

-export([init/1, handle_req/2, terminate/0]).

...</code>

    <p>The contracts specified with <c>-callback</c> attributes in
    behaviour modules can be further refined by adding <c>-spec</c>
    attributes in callback modules. This can be useful as
    <c>-callback</c> contracts are usually generic. The same callback
    module with contracts for the callbacks:</p>
    
    <code type="none">
-module(db).
-behaviour(simple_server).

-export([init/1, handle_req/2, terminate/0]).

-record(state, {field1 :: [atom()], field2 :: integer()}).

-type state()   :: #state{}.
-type request() :: {'store', term(), term()};
                   {'lookup', term()}.

...

-spec handle_req(request(), state()) -> {'ok', term()}.

...</code>

    <p>Each <c>-spec</c> contract is to be a subtype of the respective
    <c>-callback</c> contract.</p>
    
  </section>
</chapter>