NAME

Object::Pad - a simple syntax for lexical slot-based objects

SYNOPSIS

use Object::Pad;

class Point {
   has $x = 0;
   has $y = 0;

   BUILD {
     ($x, $y) = @_;
   }

   method move ($dX, $dY) {
      $x += $dX;
      $y += $dY;
   }

   method describe {
      print "A point at ($x, $y)\n";
   }
}

Point->new(5,10)->describe;

DESCRIPTION

WARNING This is an experimental proof-of-concept. Please don't actually use this in production unless you are crazy :)

This module provides a simple syntax for creating object classes, which uses private variables that look like lexicals as object member fields.

Automatic Construction

Classes are automatically provided with a constructor method, called new, which helps create the object instances.

As part of the construction process, the BUILD block of every component class will be invoked, passing in the list of arguments the constructor was invoked with. Each class should perform its required setup behaviour, but does not need to chain to the SUPER class first; this is handled automatically.

If the class provides a BUILDARGS class method, that is used to mangle the list of arguments before the BUILD blocks are called. Note this must be a class method not an instance method (and so implemented using sub). It should perform any SUPER chaining as may be required.

@args = $class->BUILDARGS( @_ )

KEYWORDS

class

class Name :ATTRS... {
   ...
}

class Name :ATTRS...;

Behaves similarly to the package keyword, but provides a package that defines a new class. Such a class provides an automatic constructor method called new.

As with package, an optional block may be provided. If so, the contents of that block define the new class and the preceding package continues afterwards. If not, it sets the class as the package context of following keywords and definitions.

As with package, an optional version declaration may be given. If so, this sets the value of the package's $VERSION variable.

class Name VERSION { ... }

class Name VERSION;

A single superclass is supported by the keyword extends

class Name extends BASECLASS {
   ...
}

class Name extends BASECLASS BASEVER {
   ...
}

If a package providing the superclass does not exist, an attempt is made to load it by code equivalent to

require Animal ();

and thus it must either already exist, or be locatable via the usual @INC mechanisms.

The superclass may or may not itself be implemented by Object::Pad, but if it is not then see "SUBCLASSING CLASSIC PERL CLASSES" for further detail on the semantics of how this operates.

An optional version check can also be supplied; it performs the equivalent of

BaseClass->VERSION( $ver )

An optional list of attributes may be supplied in similar syntax as for subs or lexical variables. (These are annotations about the class itself; the concept should not be confused with per-object-instance data, which here is called "slots").

The following class attributes are supported:

:repr(TYPE)

Sets the representation type for instances of this class. Must be one of the following values:

:repr(native)

The native representation. This is an opaque representation type whose contents are not specified. It only works for classes whose entire inheritence hierarchy is built only from classes based on Object::Pad.

:repr(HASH)

The representation will be a blessed hash reference. The instance data will be stored in an array referenced by a key called Object::Pad/slots, which is fairly unlikely to clash with existing storage on the instance. No other keys will be used; they are available for implementions and subclasses to use. The exact format of the value stored here is not specified and may change between module versions, though it can be relied on to be well-behaved as some kind of perl data structure for purposes of modules like Data::Dumper or serialisation into things like YAML or JSON.

This representation type may be useful when converting existing classes into using Object::Pad where there may be existing subclasses of it that presume a blessed hash for their own use.

:repr(magic)

The representation will use MAGIC to apply the instance data in a way that is invisible at the Perl level, and shouldn't get in the way of other things the instance is doing even in XS modules.

This representation type is the only one that will work for subclassing existing classes that do not use blessed hashes.

:repr(autoselect), :repr(default)

Since version 0.23.

This representation will select one of the representations above depending on what is best for the situation. Classes not derived from a non-Object::Pad base class will pick native, and classes derived from non-Object::Pad bases will pick either the HASH or magic forms depending on whether the instance is a blessed hash reference or some other kind.

This achieves the best combination of DWIM while still allowing the common forms of hash reference to be inspected by Data::Dumper, etc. This is the default representation type, and does not have to be specifically requested.

has

has $var;
has $var = EXPR;
has @var;
has %var;

has $var :ATTR ATTR...;

Declares that the instances of the class have a member field of the given name. This member field (called a "slot") will be accessible as a lexical variable within any method declarations in the class.

Array and hash members are permitted and behave as expected; you do not need to store references to anonymous arrays or hashes.

Member fields are private to a class. They are not visible to users of the class, nor to subclasses. In order to provide access to them a class may wish to use "method" to create an accessor.

A scalar slot may provide a expression that gives an initialisation value, which will be assigned into the slot of every instance during the constructor before the BUILD blocks are invoked. Since version 0.29 this expression does not have to be a compiletime constant, though it is evaluated exactly once, at runtime, after the class definition has been parsed. It is not evaluated individually for every object instance of that class.

The following slot attributes are supported:

:reader, :reader(NAME)

Since version 0.27.

Generates a reader method to return the current value of the slot. Currently these are only permitted for scalar slots. If no name is given, the name of the slot is used. A single prefix character _ will be removed if present.

has $slot :reader;

# equivalent to
has $slot;  method slot { return $slot }

:writer, :writer(NAME)

Since version 0.27.

Generates a writer method to set a new value of the slot from its first argument. Currently these are only permitted for scalar slots. If no name is given, the name of the slot is used prefixed by set_. A single prefix character _ will be removed if present.

has $slot :writer;

# equivalent to
has $slot;  method set_slot { $slot = shift; return $self }

Since version 0.28 a generated writer method will return the object invocant itself, allowing a chaining style.

$obj->set_x("x")
   ->set_y("y")
   ->set_z("z");

:mutator, :mutator(NAME)

Since version 0.27.

Generates an lvalue mutator method to return or set the value of the slot. These are only permitted for scalar slots. If no name is given, the name of the slot is used. A single prefix character _ will be removed if present.

has $slot :mutator;

# equivalent to
has $slot;  method slot :lvalue { $slot }

Since version 0.28 all of these generated accessor methods will include argument checking similar to that used by subroutine signatures, to ensure the correct number of arguments are passed - usually zero, but exactly one in the case of a :writer method.

method

method NAME {
   ...
}

method NAME (SIGNATURE) {
   ...
}

method NAME :ATTRS... {
   ...
}

Declares a new named method. This behaves similarly to the sub keyword, except that within the body of the method all of the member fields ("slots") are also accessible. In addition, the method body will have a lexical called $self which contains the invocant object directly; it will already have been shifted from the @_ array.

The signatures feature is automatically enabled for method declarations. In this case the signature does not have to account for the invocant instance; that is handled directly.

method m ($one, $two) {
   say "$self invokes method on one=$one two=$two";
}

...
$obj->m(1, 2);

A list of attributes may be supplied as for sub. The most useful of these is :lvalue, allowing easy creation of read-write accessors for slots (but see also the :reader, :writer and :mutator slot attributes).

class Counter {
   has $count;

   method count :lvalue { $count }
}

my $c = Counter->new;
$c->count++;

Every method automatically gets the :method attribute applied, which suppresses warnings about ambiguous calls resolved to core functions if the name of a method matches a core function.

The following additional attributes are recognised by Object::Pad directly:

:override

Since version 0.29.

Marks that this method expects to override another of the same name from a superclass. It is an error at compiletime if the superclass does not provide such a method.

BUILD

BUILD {
   ...
}

BUILD (SIGNATURE) {
   ...
}

Since version 0.27.

Declares the builder block for this component class. A builder block may use subroutine signature syntax, as for methods, to assist in unpacking its arguments. A build block is not a subroutine and thus is not permitted to use subroutine attributes (for example :lvalue).

Currently attempts to create a method named BUILD (i.e. with syntax method BUILD {...}) will create a builder block instead. As of version 0.31 such attempts will print a warning at compiletime, and a later version may remove this altogether.

IMPLIED PRAGMATA

In order to encourage users to write clean, modern code, the body of the class block acts as if the following pragmata are in effect:

use strict;
use warnings;
no indirect ':fatal';  # or  no feature 'indirect' on perl 5.32 onwards
use feature 'signatures';

This list may be extended in subsequent versions to add further restrictions and should not be considered exhaustive.

Further additions will only be ones that remove "discouraged" or deprecated language features with the overall goal of enforcing a more clean modern style within the body. As long as you write code that is in a clean, modern style (and I fully accept that this wording is vague and subjective) you should not find any new restrictions to be majorly problematic. Either the code will continue to run unaffected, or you may have to make some small alterations to bring it into a conforming style.

SUBCLASSING CLASSIC PERL CLASSES

There are a number of details specific to the case of deriving an Object::Pad class from an existing classic Perl class that is not implemented using Object::Pad.

Storage of Instance Data

Instances will pick either the :repr(HASH) or :repr(magic) storage type.

Object State During Methods Invoked By Superclass Constructor

It is common in classic Perl OO style to invoke methods on $self during the constructor. This is supported here since Object::Pad version 0.19. Note however that any methods invoked by the superclass constructor may not see the object in a fully consistent state. (This fact is not specific to using Object::Pad and would happen in classic Perl OO as well). The slot initialisers will have been invoked but the BUILD blocks will not.

For example; in the following

package ClassicPerlBaseClass {
   sub new {
      my $self = bless {}, shift;
      say "Value seen by superconstructor is ", $self->get_value;
      return $self;
   }
   sub get_value { return "A" }
}

class DerivedClass extends ClassicPerlBaseClass {
   has $_value = "B";
   BUILD {
      $_value = "C";
   }
   method get_value { return $_value }
}

my $obj = DerivedClass->new;
say "Value seen by user is ", $obj->get_value;

Until the ClassicPerlBaseClass::new superconstructor has returned the BUILD block will not have been invoked. The $_value slot will still exist, but its value will be B during the superconstructor. After the superconstructor, the BUILD blocks are invoked before the completed object is returned to the user. The result will therefore be:

Value seen by superconstructor is B
Value seen by user is C

STYLE SUGGESTIONS

While in no way required, the following suggestions of code style should be noted in order to establish a set of best practices, and encourage consistency of code which uses this module.

$VERSION declaration

While it would be nice for CPAN and other toolchain modules to parse the embedded version declarations in class statements, the current state at time of writing (June 2020) is that none of them actually do. As such, it will still be necessary to make a once-per-file $VERSION declaration in syntax those modules can parse.

Further note that these modules will also not parse the class declaration, so you will have to duplicate this with a package declaration as well as a class keyword. This does involve repeating the package name, so is slightly undesirable.

It is hoped that eventually upstream toolchain modules will be adapted to accept the class syntax as being sufficient to declare a package and set its version.

See also

File Layout

Begin the file with a use Object::Pad line; ideally including a minimum-required version. This should be followed by the toplevel package and class declarations for the file. As it is at toplevel there is no need to use the block notation; it can be a unit class.

There is no need to use strict or apply other usual pragmata; these will be implied by the class keyword.

use Object::Pad 0.16;

package My::Classname 1.23;
class My::Classname;

# other use statements

# has, methods, etc.. can go here

Slot Names

Slot names should follow similar rules to regular lexical variables in code - lowercase, name components separated by underscores. For tiny examples such as "dumb record" structures this may be sufficient.

class Tag {
   has $name  :mutator;
   has $value :mutator;
}

In larger examples with lots of non-trivial method bodies, it can get confusing to remember where the slot variables come from (because we no longer have the $self->{ ... } visual clue). In these cases it is suggested to prefix the slot names with a leading underscore, to make them more visually distinct.

class Spudger {
   has $_grapefruit;

   ...

   method mangle {
      $_grapefruit->peel; # The leading underscore reminds us this is a slot
   }
}

WITH OTHER MODULES

Syntax::Keyword::Dynamically

A cross-module integration test asserts that dynamically works correctly on object instance slots:

use Object::Pad;
use Syntax::Keyword::Dynamically;

class Container {
   has $value = 1;

   method example {
      dynamically $value = 2;
      ,..
      # value is restored to 1 on return from this method
   }
}

Future::AsyncAwait

As of Future::AsyncAwait version 0.38 and Object::Pad version 0.15, both modules now use XS::Parse::Sublike to parse blocks of code. Because of this the two modules can operate together and allow class methods to be written as async subs which await expressions:

use Future::AsyncAwait;
use Object::Pad;

class Example
{
   async method perform ($block)
   {
      say "$self is performing code";
      await $block->();
      say "code finished";
   }
}

These three modules combine; there is additionally a cross-module test to ensure that object instance slots can be dynamically set during a suspended async method.

DESIGN TODOs

The following points are details about the design of pad slot-based object systems in general:

  • Is multiple inheritence actually required, if role composition is implemented including giving roles the ability to use private slots?

  • Consider the visibility of superclass slots to subclasses. Do subclasses even need to be able to see their superclass's slots, or are accessor methods always appropriate?

    Concrete example: The $self->{split_at} access that Tickit::Widget::HSplit makes of its parent class Tickit::Widget::LinearSplit.

IMPLEMENTATION TODOs

These points are more about this particular module's implementation:

  • Implement roles, including required method checking and the ability to have private slots.

  • Consider multiple inheritence of subclassing, if that is still considered useful after adding roles.

  • Work out why no indirect doesn't appear to work properly before perl 5.20.

  • Work out why we don't get a Subroutine new redefined at ... warning if we

    sub new { ... }
  • The local modifier does not work on slot variables, because they appear to be regular lexicals to the parser at that point. A workaround is to use Syntax::Keyword::Dynamically instead:

    use Syntax::Keyword::Dynamically;
    
    has $loglevel;
    
    method quietly {
       dynamically $loglevel = LOG_ERROR;
       ...
    }

AUTHOR

Paul Evans <leonerd@leonerd.org.uk>