A note on nomenclature

Confusingly, POE::NFA is a class for implementing actual event-driven state machines. Its documentation uses "state" in the proper sense.

POE::Session's Calling Convention

  sub handle_event {
    my ($kernel, $heap, $parameter) = @_[KERNEL, HEAP, ARG0];
    ...;
  }

Or the use of $_[KERNEL], $_[HEAP] and $_[ARG0] inline, as is done in most examples.

What's going on here is rather basic. Perl passes parameters into subroutines or methods using the @_ array. "KERNEL", "HEAP", "ARG0" and others are constants exported by POE::Session (which is included for free when a program uses POE).

So $_[KERNEL] is an event handler's KERNELth parameter. @_[HEAP, ARG0] is a slice of @_ containing the HEAPth and ARG0th parameters.

While this looks odd, it's perfectly plain and legal Perl syntax. POE uses it for a few reasons:

1.

In the common case, passing parameters in @_ is faster than passing hash or array references and then dereferencing them in the handler.

2.

Typos in hash-based parameter lists are either subtle run-time errors or requires constant run-time checking. Constants are either known at compile time, or are clear compile-time errors.

3.

Referencing @_ offsets by constants allows parameters to move in the future without breaking application code.

4.

Most event handlers don't need all of @_. Slices allow handlers to use only the parameters they're interested in.

POE::Session Parameters

$_[OBJECT]

$_[OBJECT] is $self for event handlers that are an object method. It is the class (package) name for class-based event handlers. It is undef for plain coderef callbacks, which have no special $self-ish value.

"OBJECT" is always zero, since $_[0] is always $self or $class in object and class methods. Coderef handlers are called with an "undef" placeholder in $_[0] so that the other offsets remain valid.

It's often useful for method-based event handlers to call other methods in the same object. $_[OBJECT] helps this happen.

  sub ui_update_everything {
    my $self = $_[OBJECT];
    $self->update_menu();
    $self->update_main_window();
    $self->update_status_line();
  }

You may also use method inheritance. Here we invoke $self->a_method(@_). Since Perl's "->" operator unshifts $self onto the beginning of @_, we must first shift a copy off to maintain POE's parameter offsets:

  sub a_method {
    my $self = shift;
    $self->SUPER::a_method( @_ );
    # ... more work ...
  }

$_[SESSION]

$_[SESSION] is a reference to the current session object. This lets event handlers access their session's methods. Programs may also compare $_[SESSION] to $_[SENDER] to verify that intra-session events did not come from other sessions.

$_[SESSION] may also be used as the destination for intra-session post() and call(). yield() is marginally more convenient and efficient than "post($_[SESSION], ...)" however.

It is bad form to access another session directly. The recommended approach is to manipulate a session through an event handler.

  sub enable_trace {
    my $previous_trace = $_[SESSION]->option( trace => 1 );
    my $id = $_[SESSION]->ID;
    if ($previous_trace) {
      print "Session $id: dispatch trace is still on.\n";
    }
    else {
      print "Session $id: dispatch trace has been enabled.\n";
    }
  }

$_[KERNEL]

The KERNELth parameter is always a reference to the application's singleton POE::Kernel instance. It is most often used to call POE::Kernel methods from event handlers.

  # Set a 10-second timer.
  $_[KERNEL]->delay( time_is_up => 10 );

$_[HEAP]

Every POE::Session object contains its own variable namespace known as the session's "HEAP". It is modeled and named after process memory heaps (not priority heaps). Heaps are by default anonymous hash references, but they may be initialized in create() to be almost anything. POE::Session itself never uses $_[HEAP], although some POE components do.

Heaps do not overlap between sessions, although create()'s "heap" parameter can be used to make this happen.

These two handlers time the lifespan of a session:

  sub _start_handler {
    $_[HEAP]{ts_start} = time();
  }

  sub _stop_handler {
    my $time_elapsed = time() - $_[HEAP]{ts_start};
    print "Session ", $_[SESSION]->ID, " elapsed seconds: $elapsed\n";
  }

$_[STATE]

The STATEth handler parameter contains the name of the event being dispatched in the current callback. This can be important since the event and handler names may significantly differ. Also, a single handler may be assigned to more than one event.

  POE::Session->create(
    inline_states => {
      one => \&some_handler,
      two => \&some_handler,
      six => \&some_handler,
      ten => \&some_handler,
      _start => sub {
        $_[KERNEL]->yield($_) for qw(one two six ten);
      }
    }
  );

  sub some_handler {
    print(
      "Session ", $_[SESSION]->ID,
      ": some_handler() handled event $_[STATE]\n"
    );
  }

It should be noted however that having event names and handlers names match will make your code easier to navigate.

$_[SENDER]

Events must come from somewhere. $_[SENDER] contains the currently dispatched event's source.

$_[SENDER] is commonly used as a return address for responses. It may also be compared against $_[KERNEL] to verify that timers and other POE::Kernel-generated events were not spoofed.

This "echo_handler()" responds to the sender with an "echo" event that contains all the parameters it received. It avoids a feedback loop by ensuring the sender session and event (STATE) are not identical to the current ones.

  sub echo_handler {
    return if $_[SENDER] == $_[SESSION] and $_[STATE] eq "echo";
    $_[KERNEL]->post( $_[SENDER], "echo", @_[ARG0..$#_] );
  }

$_[CALLER_FILE], $_[CALLER_LINE] and $_[CALLER_STATE]

These parameters are a form of caller(), but they describe where the currently dispatched event originated. CALLER_FILE and CALLER_LINE are fairly plain. CALLER_STATE contains the name of the event that was being handled when the event was created, or when the event watcher that ultimately created the event was registered.

@_[ARG0..ARG9] or @_[ARG0..$#_]

Parameters $_[ARG0] through the end of @_ contain parameters provided by application code, event watchers, or higher-level libraries. These parameters are guaranteed to be at the end of @_ so that @_[ARG0..$#_] will always catch them all.

$#_ is the index of the last value in @_. Blame Perl if it looks odd. It's merely the $#array syntax where the array name is an underscore.

Consider

  $_[KERNEL]->yield( ev_whatever => qw( zero one two three ) );

The handler for ev_whatever will be called with "zero" in $_[ARG0], "one" in $_[ARG1], and so on. @_[ARG0..$#_] will contain all four words.

  sub ev_whatever {
    $_[OBJECT]->whatever( @_[ARG0..$#_] );
  }

Using POE::Session With Objects

  POE::Session->create(
    object_states => [
      $object_1 => { event_1a => "method_1a" },
      $object_2 => { event_2a => "method_2a" },
    ],
    inline_states => {
      event_3 => \&piece_of_code,
    },
  );

However only one handler may be assigned to a given event name. Duplicates will overwrite earlier ones.

event_1a is handled by calling "$object_1->method_1a(...)". $_[OBJECT] is $object_1 in this case. $_[HEAP] belongs to the session, which means anything stored there will be available to any other event handler regardless of the object.

event_2a is handled by calling "$object_2->method_2a(...)". In this case $_[OBJECT] is $object_2. $_[HEAP] is the same anonymous hashref that was passed to the event_1a handler, though. The methods are resolved when the event is handled (late-binding).

event_3 is handled by calling "piece_of_code(...)". $_[OBJECT] is "undef" here because there's no object. And once again, $_[HEAP] is the same shared hashref that the handlers for event_1a and event_2a saw.

Interestingly, there's no technical reason that a single object can't handle events from more than one session:

  for (1..2) {
    POE::Session->create(
      object_states => [
        $object_4 => { event_4 => "method_4" },
      ]
    );
  }

Now "$object_4->method_4(...)" may be called to handle events from one of two sessions. In both cases, $_[OBJECT] will be $object_4, but $_[HEAP] will hold data for a particular session.

The same goes for inline states. One subroutine may handle events from many sessions. $_[SESSION] and $_[HEAP] can be used within the handler to easily access the context of the session in which the event is being handled.

create LOTS_OF_STUFF

"create()" invokes the newly started session's _start event handler before returning.

"create()" also passes the new POE::Session object to POE::Kernel. POE's kernel holds onto the object in order to dispatch events to it. POE::Kernel will release the object when it detects the object has become moribund. This should cause Perl to destroy the object if application code has not saved a copy of it.

"create()" accepts several named parameters, most of which are optional. Note however that the parameters are not part of a hashref.

args => ARRAYREF

The "args" parameter accepts a reference to a list of parameters that will be passed to the session's _start event handler in @_ positions "ARG0" through $#_ (the end of @_).

This example would print "arg0 arg1 etc.":

  POE::Session->create(
    inline_states => {
      _start => sub {
        print "Session started with arguments: @_[ARG0..$#_]\n";
      },
    },
    args => [ 'arg0', 'arg1', 'etc.' ],
  );

heap => ANYTHING

The "heap" parameter allows a session's heap to be initialized differently at instantiation time. Heaps are usually anonymous hashrefs, but "heap" may set them to be array references or even objects.

This example prints "tree":

  POE::Session->create(
    inline_states => {
      _start => sub {
        print "Slot 0 = $_[HEAP][0]\n";
      },
    },
    heap => [ 'tree', 'bear' ],
  );

Be careful when initializing the heap to be something that doesn't behave like a hashref. Some libraries assume hashref heap semantics, and they will fail if the heap doesn't work that way.

inline_states => HASHREF

"inline_states" maps events names to the subroutines that will handle them. Its value is a hashref that maps event names to the coderefs of their corresponding handlers:

  POE::Session->create(
    inline_states => {
      _start => sub {
        print "arg0=$_[ARG0], arg1=$_[ARG1], etc.=$_[ARG2]\n";
      },
      _stop  => \&stop_handler,
    },
    args => [qw( arg0 arg1 etc. )],
  );

The term "inline" comes from the fact that coderefs can be inlined anonymous subroutines.

Be very careful with closures, however. "Beware circular references".

object_states => ARRAYREF

"object_states" associates one or more objects to a session and maps event names to the object methods that will handle them. It's value is an "ARRAYREF"; "HASHREFs" would stringify the objects, ruining them for method invocation.

Here _start is handled by "$object->_session_start()" and _stop triggers "$object->_session_stop()":

  POE::Session->create(
    object_states => [
      $object => {
        _start => '_session_start',
        _stop  => '_session_stop',
      }
    ]
  );

POE::Session also supports a short form where the event and method names are identical. Here _start invokes $object->_start(), and _stop triggers $object->_stop():

  POE::Session->create(
    object_states => [
      $object => [ '_start', '_stop' ],
    ]
  );

Methods are verified when the session is created, but also resolved when the handler is called (late binding). Most of the time, a method won't change. But in some circumstance, such as dynamic inheritance, a method could resolve to a different subroutine.

options => HASHREF

POE::Session sessions support a small number of options, which may be initially set with the "option" constructor parameter and changed at run time with the "option()|/option" method.

"option" takes a hashref with option => value pairs:

  POE::Session->create(
    ... set up handlers ...,
    options => { trace => 1, debug => 1 },
  );

This is equivalent to the previous example:

  POE::Session->create(
    ... set up handlers ...,
  )->option( trace => 1, debug => 1 );

The supported options and values are documented with the "option()|/option" method.

package_states => ARRAYREF

"package_states" associates one or more classes to a session and maps event names to the class methods that will handle them. Its function is analogous to "object_states", but package names are specified rather than objects.

In fact, the following documentation is a copy of the "object_states" description with some word substitutions.

The value for "package_states" is an ARRAYREF to be consistent with "object_states", even though class names (also known as package names) are already strings, so it's not necessary to avoid stringifying them.

Here _start is handled by "$class_name->_session_start()" and _stop triggers "$class_name->_session_stop()":

  POE::Session->create(
    package_states => [
      $class_name => {
        _start => '_session_start',
        _stop  => '_session_stop',
      }
    ]
  );

POE::Session also supports a short form where the event and method names are identical. Here _start invokes "$class_name->_start()", and _stop triggers "$class_name->_stop()":

  POE::Session->create(
    package_states => [
      $class_name => [ '_start', '_stop' ],
    ]
  );

ID

It's theoretically possible that a session ID will not be unique, but this requires at least 4.29 billion sessions to be created within a program's lifespan. POE guarantees that no two sessions will have the same ID at the same time, however; your computer doesn't have enough memory to store 4.29 billion session objects.

A session's ID is unique within a running process, but multiple processes are likely to have the same session IDs. If a global ID is required, it will need to include both "$_[KERNEL]->ID" and "$_[SESSION]->ID".

option OPTION_NAME [, OPTION_VALUE [, OPTION_NAME, OPTION_VALUE]... ]

It may be called with a single OPTION_NAME to retrieve the value of that option.

  my $trace_value = $_[SESSION]->option('trace');

"option()" sets an option's value when called with a single OPTION_NAME, OPTION_VALUE pair. In this case, "option()" returns the option's previous value.

  my $previous_trace = $_[SESSION]->option(trace => 1);

"option()" may also be used to set the values of multiple options at once. In this case, "option()" returns all the specified options' previous values in an anonymous hashref:

  my $previous_values = $_[SESSION]->option(
    trace => 1,
    debug => 1,
  );

  print "Previous option values:\n";
  while (my ($option, $old_value) = each %$previous_values) {
    print "  $option = $old_value\n";
  }

POE::Session currently supports three options:

The "debug" option.

The "debug" option is intended to enable additional warnings when strange things are afoot within POE::Session. At this time, there is only one additional warning:

•

Redefining an event handler does not usually cause a warning, but it will when the "debug" option is set.

The "default" option.

Enabling the "default" option causes unknown events to become warnings, if there is no _default handler to catch them.

The class-level "POE::Session::ASSERT_STATES" flag is implemented by enabling the "default" option on all new sessions.

The "trace" option.

Turn on the "trace" option to dump a log of all the events dispatched to a particular session. This is a session-specific trace option that allows individual sessions to be debugged.

Session-level tracing also indicates when events are redirected to _default. This can be used to discover event naming errors.

User-defined options.

"option()" does not verify whether OPTION_NAMEs are known, so "option()" may be used to store and retrieve user-defined information.

Choose option names with caution. There is no established convention to avoid namespace collisions between user-defined options and future internal options.

postback EVENT_NAME, EVENT_PARAMETERS

In other words, ARG0 allows the postback's creator to pass context through the postback. ARG1 allows the caller to return information.

This example creates a coderef that when called posts "ok_button" to $some_session with ARG0 containing "[ 8, 6, 7 ]".

  my $postback = $some_session->postback( "ok_button", 8, 6, 7 );

Here's an example event handler for "ok_button".

  sub handle_ok_button {
    my ($creation_args, $called_args) = @_[ARG0, ARG1];
    print "Postback created with (@$creation_args).\n";
    print "Postback called with (@$called_args).\n";
  }

Calling $postback->(5, 3, 0, 9) would perform the equivalent of...

  $poe_kernel->post(
    $some_session, "ok_button",
    [ 8, 6, 7 ],
    [ 5, 3, 0, 9 ]
  );

This would be displayed when "ok_button" was dispatched to handle_ok_button():

  Postback created with (8 6 7).
  Postback called with (5 3 0 9).

Postbacks hold references to their target sessions. Therefore sessions with outstanding postbacks will remain active. Under every event loop except Tk, postbacks are blessed so that DESTROY may be called when their users are done. This triggers a decrement on their reference counts, allowing sessions to stop.

Postbacks have one method, weaken(), which may be used to reduce their reference counts upon demand. weaken() returns the postback, so you can do:

  my $postback = $session->postback("foo")->weaken();

Postbacks were created as a thin adapter between callback libraries and POE. The problem at hand was how to turn callbacks from the Tk graphical toolkit's widgets into POE events without subclassing several Tk classes. The solution was to provide Tk with plain old callbacks that posted POE events.

Since "postback()" and "callback()" are Session methods, they may be called on $_[SESSION] or $_[SENDER], depending on particular needs. There are usually better ways to interact between sessions than abusing postbacks, however.

Here's a brief example of attaching a Gtk2 button to a POE event handler:

  my $btn = Gtk2::Button->new("Clear");
  $btn->signal_connect( "clicked", $_[SESSION]->postback("ev_clear") );

Points to remember: The session will remain alive as long as $btn exists and holds a copy of $_[SESSION]'s postback. Any parameters passed by the Gtk2 button will be in ARG1.

callback EVENT_NAME, EVENT_PARAMETERS

callback() was created to avoid race conditions that arise when external libraries assume callbacks will execute synchronously. File::Find is an obvious (but not necessarily appropriate) example. It provides a lot of information in local variables that stop being valid after the callback. The information would be unavailable by the time a post()ed event was dispatched.

get_heap

That is, it prevents the need for:

  sub some_helper_function {
    my ($kernel, $session, $heap, @specific_parameters) = @_;
    ...;
  }

Rather, helper functions may use:

  use POE::Kernel; # exports $poe_kernel
  sub some_helper_function {
    my (@specific_parameters) = @_;
    my $session = $poe_kernel->get_active_session();
    my $heap = $session->get_heap();
  }

This isn't very convenient for people writing libraries, but it makes the libraries much more convenient to use.

Using "get_heap()" to break another session's encapsulation is strongly discouraged.

instantiate CREATE_PARAMETERS

Subclasses may override "instantiate()" to alter the underlying session's structure. They may extend "instantiate()" to add new parameters to "create()".

Any parameters not recognized by "create()" must be removed from the CREATE_PARAMETERS before "instantiate()" returns. "create()" will croak if it discovers unknown parameters.

Be sure to return $self from instantiate.

  sub instantiate {
    my ($class, $create_params) = @_;

    # Have the base class instantiate the new session.
    my $self = $class->SUPER::instantiate($create_parameters);

    # Extend the parameters recognized by create().
    my $new_option = delete $create_parameters->{new_option};
    if (defined $new_option) {
      # ... customize $self here ...
    }

    return $self;
  }

try_alloc START_ARGS

It is a subclassing hook for late session customization prior to "create()" returning. It may also affect the contents of @_[ARG0..$#_] that are passed to the session's _start handler.

  sub try_alloc {
    my ($self, @start_args) = @_;

    # Perform late initialization.
    # ...

    # Give $self to POE::Kernel.
    return $self->SUPER::try_alloc(@args);
  }

_default

If there's no _default handler, POE::Session will silently drop the event unless the "default" option is set.

To preserve the original information, the original event is slightly changed before being redirected to the _default handler: The original event parameters are moved to an array reference in ARG1, and the original event name is passed to _default in ARG0.

  sub handle_default {
    my ($event, $args) = @_[ARG0, ARG1];
    print(
      "Session ", $_[SESSION]->ID,
      " caught unhandled event $event with (@$args).\n"
    );
  }

_default is quite flexible. It may be used for debugging, or to handle dynamically generated event names without pre-defining their handlers. In the latter sense, _default performs analogously to Perl's "AUTOLOAD".

_default may also be used as the default or "otherwise" clause of a switch statement. Consider an input handler that throws events based on a command name:

  sub parse_command {
    my ($command, @parameters) = split /\s+/, $_[ARG0];
    $_[KERNEL]->post( "cmd_$command", @parameters );
  }

A _default handler may be used to emit errors for unknown commands:

  sub handle_default {
    my $event = $_[ARG0];
    return unless $event =~ /^cmd_(\S+)/;
    warn "Unknown command: $1\n";
  }

The _default behavior is implemented in POE::Session, so it may be different for other session types.

POE::Session's Debugging Features

ASSERT_STATES

Setting ASSERT_STATES to true causes every Session to warn when they are asked to handle unknown events. Session.pm implements the guts of ASSERT_STATES by defaulting the "default" option to true instead of false. See the option() method earlier in this document for details about the "default" option.

Beware circular references

Here a POE::Session is created that refers to itself via an external scalar. The event handlers import $session via closures which are in turn stored within $session. Even if this session stops, the circular references will remain.

  my $session;
  $session = POE::Session->create(
    inline_states => {
      _start => sub {
        $_[HEAP]->{todo} = [ qw( step1 step2 step2a ) ],
        $_[KERNEL]->post( $session, 'next' );
      },
      next => sub {
        my $next = shift @{ $_[HEAP]->{todo} };
        return unless $next;
        $_[KERNEL]->post( $session, $next );
      }
      # ....
    }
  );

Reduced to its essence:

  my %event_handlers;
  $event_handler{_start} = sub { \%event_handlers };

Note also that an anonymous sub creates a closure on all lexical variables in the scope it was defined in, even if it doesn't reference them. $session is still being held in a circular reference here:

  my $self = $package->new;
  my $session;
  $session = POE::Session->create(
    inline_state => {
      _start => sub { $self->_start( @_[ARG0..$#_] ) }
    }
  );

To avoid this, a session may set an alias for itself. Other parts of the program may then refer to it by alias. In this case, one needn't keep track of the session themselves (POE::Kernel will do it anyway).

  POE::Session->create(
    inline_states => {
      _start => sub {
        $_[HEAP]->{todo} = [ qw( step1 step2 step2a ) ],
        $_[KERNEL]->alias_set('step_doer');
        $_[KERNEL]->post( 'step_doer', 'next' );
      },
      next => sub {
        my $next = shift @{ $_[HEAP]->{todo} };
        return unless $next;
        $_[KERNEL]->post( 'step_doer', $next );
      }
      # ....
    }
  );

Aliases aren't even needed in the previous example because the session refers to itself. One could instead use POE::Kernel's yield() method to post the event back to the current session:

  next => sub {
    my $next = shift @{ $_[HEAP]->{todo} };
    return unless $next;
    $_[KERNEL]->yield( $next );
  }

Or the "$_[SESSION]" parameter passed to every event handler, but yield() is more efficient.

  next => sub {
    my $next = shift @{ $_[HEAP]->{todo} };
    return unless $next;
    $_[KERNEL]->post( $_[SESSION], $next );
  }

Along the same lines as "$_[SESSION]", a session can respond back to the sender of an event by posting to "$_[SENDER]". This is great for responding to requests.

If a program must hold onto some kind of dynamic session reference, it's recommended to use the session's numeric ID rather than the object itself. A session ID may be converted back into its object, but post() accepts session IDs as well as objects and aliases:

  my $session_id;
  $session_id = POE::Session->create(
    inline_states => {
      _start => sub {
        $_[HEAP]->{todo} = [ qw( step1 step2 step2a ) ],
        $_[KERNEL]->post( $session_id, 'next' );
      },
      # ....
    }
  )->ID;