man Prima::Object (3): Prima toolkit base classes

SYNOPSIS

if ( $obj-> isa('Prima::Component')) {
# set and get a property
my $name = $obj-> name;
$obj->name( 'an object' );
# set a notification callback
$obj-> onPostMessage( sub {
shift;
print "hey! I've received this: @_\n";
});
# can set multiple properties. note, that 'name' and 'owner',
# replace the old values, while onPostMessage are aggregated.
$obj-> set(
name => 'AnObject',
owner => $new_owner,
onPostMessage => sub {
shift;
print "hey! me too!\n";
},
);
# de-reference by name
$new_owner-> AnObject-> post_message(1,2);
}

DESCRIPTION

Prima::Object and Prima::Component are the root objects of the Prima toolkit hierarchy. All the other objects are derived from the Component class, which in turn is the only descendant of Object class. Both of these classes are never used for spawning their instances, although this is possible using

   Prima::Component-> create( .. parameters ... );

call. This document describes the basic concepts of the OO programming with Prima toolkit. Although Component has wider functionality than Object, all examples will be explained on Component, since Object has no descendant classes and all the functionality of Object is present in Component. Some of the information here can be found in Prima::internals as well, the difference is that Prima::internals considers the coding tasks from a C programmer's view, whereas this document is wholly about perl programming.

Object base features

Creation

Object creation has fixed syntax:

   $new_object = Class-> create(
     parameter => value,
     parameter => value,
     ...
   );

Parameters and values form a hash, which is passed to the create() method. This hash is applied to a default parameter-value hash ( a profile ), specific to every Prima class. The object creation is performed in several stages.

create

create() calls profile_default() method that returns ( as its name states ) the default profile, a hash with the appropriate default values assigned to its keys. The Component class defaults are ( see Classes.pm ):

     name        => ref $_[ 0],
     owner       => $::application,
     delegations => undef,

While the exact meaning of these parameters is described later, in ``Properties'', the idea is that a newly created object will have 'owner' parameter set to '$::application' and 'delegations' to undef etc etc - unless these parameters are explicitly passed to create(). Example:

     $a1 = Prima::Component-> create();

$a1's owner will be $::application

     $a2 = Prima::Component-> create( owner => $a1);

$a2's owner will be $a1. The actual merging of the default and the parameter hashes is performed on the next stage, in profile_check_in() method which is called inside profile_add() method.

profile_check_in

A profile_check_in() method merges the default and the parameter profiles. By default all specified parameters have the ultimate precedence over the default ones, but in case the specification is incomplete or ambiguous, the profile_check_in()'s task is to determine actual parameter values. In case of Component, this method maintains a simple automatic naming of the newly created objects. If the object name was not specified with create(), it is assigned to a concatenated class name with an integer - Component1, Component2 etc.

Another example can be taken from Prima::Widget::profile_check_in(). Prima::Widget horizontal position can be specified by using basic "left" and "width" parameters, and as well by auxiliary "right", "size" and "rect". The default of both "left" and "width" is 100. But if only "right" parameter, for example, was passed to create() it is profile_check_in() job to determine "left" value, given that "width" is still 100.

After profiles gets merged, the resulting hash is passed to the third stage, init().

init

init() duty is to map the profile content into object, e.g., assign "name" property to "name" parameter value, and so on - for all relevant parameters. After that, it has to return the profile in order the overridden subsequent init() methods can perform same actions. This stage along with the previous is exemplified in almost all Prima modules.

Note: usually init() attaches the object to its owner in order to keep the newly-created object instance from being deleted by garbage-collection mechanisms. More on that later ( see ``Links between objects'').

After init() finishes, create() calls setup() method

setup

setup() method is a convenience function, it is used when some post-init actions must be taken. It is seldom overloaded, primarily because the Component::setup() method calls "onCreate" notification, which is more convenient to overload than setup().

As can be noticed from the code pieces above, a successful create() call returns a newly created object. If an error condition occurred, undef is returned. It must be noted, that only errors that were generated via die() during init() stage result in undef. Other errors raise an exception instead. It is not recommended to frame create() calls in an "eval{}" block, because the error conditions can only occur in two situations. The first is a system error, either inside perl or Prima guts, and not much can be done here, since that error can very probably lead to an unstable program and almost always signals an implementation bug. The second reason is a caller's error, when an unexistent parameter key or invalid value is passed; such conditions are not subject to a runtime error handling as are not the syntax errors.

After create(), the object is subject to the event flow. As "onCreate" event is the first event the object receives, only after that stage other events can be circulated.

Destruction

Object destruction can be caused by many conditions, but all execution flow is finally passed through destroy() method. destroy(), as well as create() performs several finalizing steps:

cleanup: The first method called inside destroy() is cleanup(). cleanup() is the pair to setup(), as destroy() is the pair to create(). cleanup() generates "onDestroy" event, which can be overridden more easily than cleanup() itself.
"onDestroy" is the last event the object sees. After cleanup() no events are allowed to circulate.
done: done() method is the pair to init(), and is the place where all object resources are freed. Although it is as safe to overload done() as init(), it almost never gets overloaded, primarily because overloading "onDestroy" is easier.

The typical conditions that lead to object destructions are direct destroy() call, garbage collections mechanisms, user-initiated window close ( on "Prima::Window" only ), and exception during init() stage. Thus, one must be careful implementing done() which is called after init() throws an exception.

Methods

The class methods are declared and used with perl OO syntax, which allow both method of object referencing:

  $object-> method();

and

  method( $object);

The actual code is a sub, located under the object class package. The overloaded methods that call their ancestor code use

  $object-> SUPER::method();

syntax. Most Prima methods have fixed number of parameters.

Properties

Properties are methods that combine functionality of two ephemeral ``get'' and ``set'' methods. The idea behind properties is that many object parameters require two independent methods, one that returns some internal state and another that changes it. For example, for managing the object name, set_name() and get_name() methods are needed. Indeed, the early Prima implementation dealt with large amount of these get's and set's, but later these method pairs were deprecated in the favor of properties. Currently, there is only one method name() ( referred as "::name" later in the documentation ).

The property returns a value if no parameters ( except the object) are passed, and changes the internal data to the passed parameters otherwise. Here's a sketch code for "::name" property implementation:

 sub name
 {
    return $_[0]-> {name} unless $#_;
    $_[0]->{name} = $_[1];
 }

There are many examples of properties throughout the toolkit. Not all properties deal with scalar values, some accept arrays or hashes as well. The properties can be set-called not only by name like

  $object-> name( "new name");

but also with set() method. The set() method accepts a hash, that is much like to create(), and assigns the values to the corresponding properties. For example, the code

  $object-> name( "new name");
  $object-> owner( $owner);

can be rewritten as

  $object-> set(
     name  => "new name",
     owner => $owner
  );

A minor positive effect of a possible speed-up is gained by eliminating C-to-perl and perl-to-C calls, especially if the code called is implemented in C. The negative effect of such technique is that the order in which the properties are set, is undefined. Therefore, the usage of set() is recommended either when the property order is irrelevant, or it is known beforehand that such a call speeds up the code, or is an only way to achieve the result. An example of the latter case from Prima::internals shows that Prima::Image calls

    $image-> type( $a);
    $image-> palette( $b);

and

    $image-> palette( $b);
    $image-> type( $a);

produce different results. It is indeed the only solution to call for such a change using

    $image-> set(
       type => $a,
       palette => $b
    );

when it is known beforehand that "Prima::Image::set" is aware of such combinations and calls neither "::type" nor "::palette" but performs another image conversion instead.

Some properties are read-only and some are write-only. Some methods that might be declared as properties are not; these are declared as plain methods with get_ or set_ name prefix. There is not much certainty about what methods are better off being declared as properties and vice versa.

However, if get_ or set_ methods cannot be used in correspondingly write or read fashion, the R/O and W/O properties can. They raise an exception on an attempt to do so.

Links between objects

Prima::Component descendants can be used as containers, as objects that are on a higher hierarchy level than the others. This scheme is implemented in a child-owner relationship. The 'children' objects have the "::owner" property value assigned to a reference to a 'owner' object, while the 'owner' object conducts the list of its children. It is a one-to-many hierarchy scheme, as a 'child' object can have only one owner, but an 'owner' object can have many children. The same object can be an owner and a child at the same time, so the owner-child hierarchy can be viewed as a tree-like structure.

Prima::Component::owner property maintains this relation, and is writable - the object can change its owner dynamically. There is no corresponding property that manages children objects, but is a method get_components(), that returns an array of the child references.

The owner-child relationship is used in several ways in the toolkit. For example, the widgets that are children of another widget appear ( usually, but not always ) in the geometrical interior of the owner widget. Some events ( keyboard events, for example ) are propagated automatically up and/or down the object tree. Another important feature is that when an object gets destroyed, its children are destroyed first. In a typical program the whole object tree roots in a Prima::Application object instance. When the application finishes, this feature helps cleaning up the widgets and quitting gracefully.

Implementation note: name 'owner' was taken instead of initial 'parent', because the 'parent' is a fixed term for widget hierarchy relationship description. Prima::Widget relationship between owner and child is not the same as GUI's parent-to-child. The parent is the widget for the children widgets located in and clipped by its inferior. The owner widget is more than that, its children can be located outside its owner boundaries.

The special convenience variety of create(), the insert() method is used to explicitly select owner of the newly created object. insert() can be considered a 'constructor' in OO-terms. It makes the construct

   $obj = Class-> create( owner => $owner, name => 'name);

Events

Prima::Component descendants employ a well-developed event propagation mechanism, which allows handling events using several different schemes. An event is a condition, caused by the system or the user, or an explicit notify() call. The formerly described events onCreate and onDestroy are triggered after a new object is created or before it gets destroyed. These two events, and the described below onPostMessage are present in namespaces of all Prima objects. New classes can register their own events and define their execution flow, using notification_types() method. This method returns all available information about the events registered in a class.

Prima defines also a non-object event dispatching and filtering mechanism, available through ``event_hook'' static method.

Propagation

The event propagation mechanism has three layers of user-defined callback registration, that are called in different order and contexts when an event is triggered. The examples below show the usage of these layers. It is assumed that an implicit

  $obj-> notify("PostMessage", $data1, $data2);

call is issued for all these examples.

Direct methods

As it is usual in OO programming, event callback routines are declared as methods. 'Direct methods' employ such a paradigm, so if a class method with name "on_postmessage" is present, it will be called as a method ( i.e., in the object context ) when "onPostMessage" event is triggered. Example:

 sub on_postmessage 
 {
    my ( $self, $data1, $data2) = @_;
    ...
 }

The callback name is a modified lower-case event name: the name for Create event is on_create, PostMessage - on_postmessage etc. These methods can be overloaded in the object's class descendants. The only note on declaring these methods in the first instance is that no "::SUPER" call is needed, because these methods are not defined by default.

Usually the direct methods are used for the internal object book-keeping, reacting on the events that are not designed to be passed higher. For example, a Prima::Button class catches mouse and keyboard events in such a fashion, because usually the only notification that is interesting for the code that employs push-buttons is "Click". This scheme is convenient when an event handling routine serves the internal, implementation-specific needs.

Delegated methods

The delegated methods are used when objects ( mostly widgets ) include other dependent objects, and the functionality requires interaction between these. The callback functions here are the same methods as direct methods, except that they get called in context of two, not one, objects. If, for example, a $obj's owner, $owner would be interested in $obj's PostMessage event, it would register the notification callback by

   $obj-> delegations([ $owner, 'PostMessage']);

where the actual callback sub will be

 sub Obj_PostMessage 
 {
    my ( $self, $obj, $data1, $data2) = @_;
 }

Note that the naming style is different - the callback name is constructed from object name ( let assume that $obj's name is 'Obj') and the event name. ( This is one of the reasons why Component::profile_check_in() performs automatic naming of newly created onbjects). Note also that context objects are $self ( that equals $owner ) and $obj.

The delegated methods can be used not only for the owner-child relations. Every Prima object is free to add a delegation method to every other object. However, if the objects are in other than owner-child relation, it is a good practice to add Destroy notification to the object which events are of interest, so if it gets destroyed, the partner object gets a message about that.

Anonymous subroutines

The two previous callback types are more relevant when a separate class is developed, but it is not necessary to declare a new class every time the event handling is needed. It is possible to use the third and the most powerful event hook method using perl anonymous subroutines ( subs ) for the easy customization.

Contrary to the usual OO event implementations, when only one routine per class dispatches an event, and calls inherited handlers when it is appropriate, Prima event handling mechanism can accept many event handlers for one object ( it is greatly facilitated by the fact that perl has anonymous subs, however).

All the callback routines are called when an event is triggered, one by one in turn. If the direct and delegated methods can only be multiplexed by the usual OO inheritance, the anonymous subs are allowed to be multiple by the design. There are three syntaxes for setting such a event hook; the example below sets a hook on $obj using each syntax for a different situation:

- during create():

   $obj = Class-> create(
    ... 
    onPostMessage => sub {
       my ( $self, $data1, $data2) = @_;
    },
    ...
    );

- after create using set()

   $obj-> set( onPostMessage => sub {
       my ( $self, $data1, $data2) = @_;
   });

- after create using event name:

   $obj-> onPostMessage( sub {
       my ( $self, $data1, $data2) = @_;
   });

As was noted in Prima, the events can be addressed as properties, with the exception that they are not substitutive but additive. The additivity is that when the latter type of syntax is used, the subs already registered do not get overwritten or discarded but stack in queue. Thus,

   $obj-> onPostMessage( sub { print "1" });
   $obj-> onPostMessage( sub { print "2" });
   $obj-> notify( "PostMessage", 0, 0);

code block would print

as the execution result.

This, it is a distinctive feature of a toolkit is that two objects of same class may have different set of event handlers.

Flow

When there is more than one handler of a particular event type present on an object, a question is risen about what are callbacks call priorities and when does the event processing stop. One of ways to regulate the event flow is based on prototyping events, by using notification_types() event type description. This function returns a hash, where keys are the event names and the values are the constants that describe the event flow. The constant can be a bitwise OR combination of several basic flow constants, that control the three aspects of the event flow.

Order

If both anonymous subs and direct/delegated methods are present, it must be decided which callback class must be called first. Both 'orders' are useful: for example, if it is designed that a class's default action is to be overridden, it is better to call the custom actions first. If, on the contrary, the class action is primary, and the others are supplementary, the reverse order is preferred. One of two "nt::PrivateFirst" and "nt::CustomFirst" constants defines the order.

Direction

Almost the same as order, but for finer granulation of event flow, the direction constants "nt::FluxNormal" and "nt::FluxReverse" are used. The 'normal flux' defines FIFO ( first in first out ) direction. That means, that the sooner the callback is registered, the greater priority it would possess during the execution. The code block shown above

   $obj-> onPostMessage( sub { print "1" });
   $obj-> onPostMessage( sub { print "2" });
   $obj-> notify( "PostMessage", 0, 0);

results in 21, not 12 because PostMessage event type is prototyped "nt::FluxReverse".

Execution control

It was stated above that the events are additive, - the callback storage is never discarded when 'set'-syntax is used. However, the event can be told to behave like a substitutive property, e.g. to call one and only one callback. This functionality is governed by "nt::Single" bit in execution control constant set, which consists of the following constants:

  nt::Single
  nt::Multiple
  nt::Event

These constants are mutually exclusive, and may not appear together in an event type declaration. A "nt::Single"-prototyped notification calls only the first ( or the last - depending on order and direction bits ) callback. The usage of this constant is somewhat limited.

In contrary of "nt::Single", the "nt::Multiple" constant sets the execution control to call all the available callbacks, with respect to direction and order bits.

The third constant, "nt::Event", is the impact as "nt::Multiple", except that the event flow can be stopped at any time by calling clear_event() method.

Although there are 12 possible event type combinations, a half of them are not viable. Another half were assigned to unique more-less intelligible names:

  nt::Default       ( PrivateFirst | Multiple | FluxReverse)
  nt::Property      ( PrivateFirst | Single   | FluxNormal )
  nt::Request       ( PrivateFirst | Event    | FluxNormal )
  nt::Notification  ( CustomFirst  | Multiple | FluxReverse )
  nt::Action        ( CustomFirst  | Single   | FluxReverse )
  nt::Command       ( CustomFirst  | Event    | FluxReverse )

Success state

Events do not return values, although the event generator, the notify() method does - it returns either 1 or 0, which is the value of event success state. The 0 and 1 results in general do not mean either success or failure, they simply reflect the fact whether clear_event() method was called during the processing - 1 if it was not, 0 otherwise. The state is kept during the whole processing stage, and can be accessed from Component::eventFlag property. Since it is allowed to call notify() inside event callbacks, the object maintains a stack for those states. Component::eventFlags always works with the topmost one, and fails if is called from outside the event processing stage. Actually, clear_event() is an alias for ::eventFlag(0) call. The state stack is operated by push_event() and pop_event() methods.

Implementation note: a call of clear_event() inside a "nt::Event"-prototyped event call does not automatically stops the execution. The execution stops if the state value equals to 0 after the callback is finished. A ::eventFlag(1) call thus cancels the effect of clear_event().

A particular coding style is used when the event is "nt::Single"-prototyped and is called many times in a row, so overheads of calling notify() become a burden. Although notify() logic is somewhat complicated, it is rather simple with "nt::Single" case. The helper function get_notify_sub() returns the context of callback to-be-called, so it can be used to emulate notify() behavior. Example:

  for ( ... ) {
     $result = $obj-> notify( "Measure", @parms);
  }

can be expressed in more cumbersome, but efficient code if "nt::Single"-prototyped event is used:

   my ( $notifier, @notifyParms) = $obj-> get_notify_sub( "Measure" );
   $obj-> push_event;
   for ( ... ) {
       $notifier-> ( @notifyParms, @parms);
       # $result = $obj-> eventFlag; # this is optional
   }
   $result = $obj-> pop_event;

API

Prima::Object methods

alive

Returns the object 'vitality' state - true if the object is alive and usable, false otherwise. This method can be used as a general checkout if the scalar passed is a Prima object, and if it is usable. The true return value can be 1 for normal and operational object state, and 2 if the object is alive but in its init() stage. Example:

  print $obj-> name if Prima::Object::alive( $obj);

can NAME, CACHE = 1

Checks if an object namespace contains a NAME method. Returns the code reference to it, if found, and undef if not. If CACHE is true, caches the result to speed-up subsequent calls.

cleanup

Called right after destroy() started. Used to initiate "cmDestroy" event. Is never called directly.

create CLASS, %PARAMETERS

Creates a new object instance of a given CLASS and sets its properties corresponding to the passed parameter hash. Examples:

   $obj = Class-> create( PARAMETERS);
   $obj = Prima::Object::create( "class" , PARAMETERS);

Is never called in an object context.

Alias: new()

destroy

Initiates the object destruction. Perform in turn cleanup() and done() calls. destroy() can be called several times and is the only Prima re-entrant function, therefore may not be overloaded.

done

Called by destroy() after cleanup() is finished. Used to free the object resources, as a finalization stage. During done() no events are allowed to circulate, and alive() returns 0. The object is not usable after done() finishes. Is never called directly.

Note: the eventual child objects are destroyed inside done() call.

get @PARAMETERS

Returns hash where keys are @PARAMETERS and values are the corresponding object properties.

init %PARAMETERS

The most important stage of object creation process. %PARAMETERS is the modified hash that was passed to create(). The modification consists of merging with the result of profile_default() class method inside profile_check_in() method. init() is responsible for applying the relevant data into PARAMETERS to the object properties. Is never called directly.

insert CLASS, %PARAMETERS

A convenience wrapper for create(), that explicitly sets the owner property for a newly created object.

   $obj = $owner-> insert( 'Class', name => 'name');

is adequate to

   $obj = Class-> create( owner => $owner, name => 'name);

code. insert() has another syntax that allows simultaneous creation of several objects:

   @objects = $owner-> insert( 
     [ 'Class', %parameters],
     [ 'Class', %parameters],
     ...
   );

With such syntax, all newly created objects would have $owner set to their 'owner' properties.

new CLASS, %PARAMETERS

Same as create.

profile_add PROFILE

The first stage of object creation process. PROFILE is a reference to a PARAMETERS hash, passed to create(). It is merged with profile_default() after passing both to profile_check_in(). The merge result is stored back in PROFILE. Is never called directly.

profile_check_in CUSTOM_PROFILE, DEFAULT_PROFILE

The second stage of object creation process. Resolves eventual ambiguities in CUSTOM_PROFILE, which is the reference to PARAMETERS passed to create(), by comparing to and using default values from DEFAULT_PROFILE, which is the result of profile_default() method. Is never called directly.

profile_default

Returns hash of the appropriate default values for all properties of a class. In object creation process serves as a provider of fall-back values, and is called implicitly. This method can be used directly, contrary to the other creation process-related functions.

Can be called in a context of class.

raise_ro TEXT

Throws an exception with text TEXT when a read-only property is called in a set- context.

raise_wo TEXT

Throws an exception with text TEXT when a write-only property is called in a get- context.

set %PARAMETERS

The default behavior is an equivalent to

  sub set
  {
     my $obj = shift;
     my %PARAMETERS = @_;
     $obj-> $_( $PARAMETERS{$_}) for keys %PARAMETERS;
  }

code. Assigns object properties correspondingly to PARAMETERS hash. Many Prima::Component descendants overload set() to make it more efficient for particular parameter key patterns.

As the code above, raises an exception if the key in PARAMETERS has no correspondent object property.

setup

The last stage of object creation process. Called after init() finishes. Used to initiate "cmCreate" event. Is never called directly.

Prima::Component methods

add_notification NAME, SUB, REFERER = undef, INDEX = -1

Adds SUB to the list of notification of event NAME. REFERER is the object reference, which is used to create a context to SUB and is passed as a parameter to it when called. If REFERER is undef ( or not specified ), the same object is assumed. REFERER also gets implicitly attached to the object, - the implementation frees the link between objects when one of these gets destroyed.

INDEX is a desired insert position in the notification list. By default it is -1, what means 'in the start'. If the notification type contains nt::FluxNormal bit set, the newly inserted SUB will be called first. If it has nt::FluxReverse, it is called last, correspondingly.

Returns positive integer value on success, 0 on failure. This value can be later used to refer to the SUB in remove_notification().

See also: "remove_notification", "get_notification".

attach OBJECT

Inserts OBJECT to the attached objects list and increases OBJECT's reference count. The list can not hold more than one reference to the same object. The warning is issued on such an attempt.

Prima::Component properties

eventFlag STATE

Provides access to the last event processing state in the object event state stack.

See also: ``Success state'', "clear_event", ``Events''.

delegations [ <REFERER>, NAME, <NAME>, < <REFERER>, NAME, ... > ]

Accepts an anonymous array in set- context, which consists of a list of event NAMEs, that a REFERER object ( the caller object by default ) is interested in. Registers notification entries for routines if subs with naming scheme REFERER_NAME are present on REFERER name space. The example code

   $obj-> name("Obj");
   $obj-> delegations([ $owner, 'PostMessage']);

registers Obj_PostMessage callback if it is present in $owner namespace.

In get- context returns an array reference that reflects the object's delegated events list content.

Prima::Component events

ChangeOwner OLD_OWNER: Called at runtime when the object changes its owner.
ChildEnter CHILD: Triggered when a child object is attached, either as a new instance or as a result of runtime owner change.
ChildLeave CHILD: Triggered when a child object is detached, either because it is getting destroyed or as a result of runtime owner change.
Create: The first event an event sees. Called automatically after init() is finished. Is never called directly.
Destroy: The last event an event sees. Called automatically before done() is started. Is never called directly.
PostMessage SCALAR1, SCALAR2: Called after post_message() call is issued, not inside post_message() but at the next idle event loop. SCALAR1 and SCALAR2 are the data passed to post_message().

AUTHOR

Dmitry Karasik, <[email protected]>.