Beefy Boxes and Bandwidth Generously Provided by pair Networks
Welcome to the Monastery

RFC - Class::LazyObject

by QwertyD (Pilgrim)
on Oct 08, 2003 at 03:29 UTC ( #297466=perlmeditation: print w/replies, xml ) Need Help??

Class::LazyObject allows you to create lazy objects. A lazy object holds the place of another object, (Called the "inflated object"). The lazy object turns into the inflated object ("inflates") only after a method is called on the lazy object. After that, any variables holding the lazy object will hold the inflated object.

I would like to submit this module to CPAN, after hearing your comments about it.

The main things I'd especially appreciate your reactions to are:

  • Namespace: I chose this namespace because Lazy Objects seem to follow kind of the same principle as lazy lists, though I realize that a lazy list is a completely different data structure. Class::LazyObject seems to be consistent with what I know about lazy evaluation. However, since I haven't studied this part of computer science (yet), I'm not sure if this is or is not the right name for this module. What would you suggest?

  • CPAN worthiness: Is this module (as part of a module distribution, with tests) worthy of CPAN? This is the first module I've considered submitting to CPAN.

I'd also like comments on:

  • The documentation: Is it clear? Does it make sense?

  • The code: Is it readable? Is it too sloppy? Do I overcomment?

  • Problems with my technique: Is there anything that I've overlooked in implementing lazy objects? Are there any weird cases (other than those listed in the CAVEATS or BUGS sections of the module's documentation) in which a Lazy Object couldn't be used in the place of an inflated object?

An HTML version of the module's POD is available at my website. If you want, I can also make a downloadable version of this module available.

Again, thank you tremendously for your input.

package Class::LazyObject; use strict; use warnings; use Carp qw(); our $VERSION = 0.09_030; use vars '$AUTOLOAD'; #We want to inflate calls to methods defined in UNIVERSAL, but we impl +icitly inherit from UNIVERSAL. #As long as we predeclare the methods here, they will override those i +n UNIVERSAL, but since they are never defined, AUTOLOAD will be calle +d: use subs grep {defined UNIVERSAL::can(__PACKAGE__, $_)} keys %UNIVERSA +L::; sub AUTOLOAD { my $self = $_[0]; #don't shift it, since we will need to access th +is directly later. $AUTOLOAD =~ /.*::(\w+)/; my $method = $1; my $class_method = (ref($self) || $self).'::Methods';#call all cla +ss methods on this. if (($method eq 'new') && !ref($self)) { #new was called on a class, rather than an object, so we shoul +d actually construct ourselves, rather than passing this to whatever +we're proxying. return $class_method->new(@_); } ref($self) or return $class_method->$method(@_[ 1 .. $#_ ]); #If t +his was called as a class method, pass it on to ::Methods but don't p +ass OUR package name. print "Lazy...\n" if $class_method->get_classdata('debug'); my $object; #this is the object we will eventually replace ourselv +es with. if ( ref($$self) && UNIVERSAL::isa($$self, $class_method->get_clas +sdata('class')) ) { $object = $$self; } else { $object = $class_method->get_classdata('inflate')->($class_met +hod->get_classdata('class'), $$self); $$self = $object; #don't create this again. } $_[0] = $object; #replace ourselves with the object. goto ( UNIVERSAL::can($_[0], $method) || $class_method->_prepareAUTOLOADRef($_[0], ref($_[0]).'::' +.$method) || #UNIVERSAL::can can't detect if a method is AUTOLOADed, +so we have to. Carp::croak(sprintf qq{Can\'t locate object method "%s" v +ia package "%s" }, $method, ref $_[0] )#Error message stolen from Cla +ss::WhiteHole ); } sub DESTROY { #You won't AUTOLOAD this! Muahahaha! } #class method to see whether something is lazy? #class methods for original isa and can #--------- package Class::LazyObject::Methods; #stick all of our class methods here so we don't pollute Class::LazyOb +ject's namespace. #everything in this class should be called as class methods, NOT objec +t methods. use Carp::Clan '^Class::LazyObject(::|$)'; use Class::Data::TIN qw(get_classdata); use Class::ISA; sub _findAUTOLOADPackage { #Takes 1 argument, either an object or the name of a package. #Returns the name of the package containing the sub AUTOLOAD that +would be called when $first_arg->AUTOLOAD was called #In other words, it traverses the inheritance hierarchy the same w +ay Perl does until it finds an AUTOLOAD, and returns the name of the +package containing the AUTOLOAD. #Returns undef if AUTOLOAD is not in the inheritance hierarchy. shift;#Don't care about our package name. my $object_or_package = shift; my $orig_class = ref($object_or_package) || $object_or_package; return undef unless UNIVERSAL::can($orig_class, 'AUTOLOAD'); my @classes = (Class::ISA::self_and_super_path($orig_class), 'UNIV +ERSAL'); my $package; foreach my $class (@classes) { no strict 'refs';#Symbol table munging ahead $package = $class; last if defined(*{$package.'::AUTOLOAD';}{CODE}); } return $package; } sub _prepareAUTOLOADRef { #Takes 2 arguments: # either an object or the name of a package # the fully qualified method name to make AUTOLOAD think it was + called as a result of #Sets the appropriate package's $AUTOLOAD so that when the AUTOLOA +D method is called on the first argument, it will think it was called + as a result of a call to the method specified by the second argument +. #Returns the result of (UNIVERSAL::can($first_arg, 'AUTOLOAD')); my $class = shift; my ($object, $method) = @_; if (UNIVERSAL::can($object, 'AUTOLOAD'))#no point in doing any of +this if it can't AUTOLOAD. { my $package = $class->_findAUTOLOADPackage($object); { no strict 'refs'; *{$package.'::AUTOLOAD'} = \$method; } } return UNIVERSAL::can($object, 'AUTOLOAD'); } #defaults, these are overridable when someone calls ->inherit Class::Data::TIN->new(__PACKAGE__, inflate => sub {return $_[0]->new_inflate($_[1]);}, debug => 1, ); sub inherit { #calls to Class::LazyObject->inherit are forwarded here. my $class = shift; #don't care about our own package name. my %params = @_; my @required_params = qw(inflated_class deflated_class); foreach my $param (@required_params) { carp "You did not pass '$param', which is a required parameter +." unless exists $params{$param}; } my %param_map = ( #keys are key names in the parameters passed to +this function. Values are corrisponding class data names. inflated_class => 'class' ); my %class_data = %params; delete @class_data{keys %param_map, 'deflated_class'};#we'll stick + these in with their appropriate names: @class_data{values %param_map} = @params{keys %param_map};#pass th +e parameters whose names have changed my $method_package = $params{deflated_class}.'::Methods'; { no strict 'refs'; #more symbol table munging push(@{$method_package.'::ISA'}, __PACKAGE__); #Create a packa +ge to hold all the methods, that inherits from THIS class, or add thi +s class to its inheritance if it does exist. #Should this be $class +instead of __PACKAGE__ #^Push is used instead of unshift so that someone can override + their ::Methods package with its own inheritence hierarchy, and meth +ods will be called here only AFTER Perl finds they don't exist in the + overridden ISA. } Class::Data::TIN->new($method_package, %class_data); } sub new { my ($own_package, $class, $id) = @_;#first argument is this method +'s, class, not the lazy object's if (ref($id) && UNIVERSAL::isa($id, $own_package->get_classdata('c +lass'))) { croak "A Lazy Object's ID cannot be a an object of same class +(or of a class inherited from the one) it is proxying!"; } return bless \$id, $class; } 1; #LAUNDRY LIST: #LAZYNESS, impatience, hubris #should we document the $AUTOLOAD persistence thingy as a caveat? #CALLING AUTOLOAD on inflate #CAVEAT: can't distinguish between no id and an id of undef. # -solve by storing a Class::LazyObject::NoID object instead of unde +f? #Does goto propogate scalar/list context? #Lvalue subs? __END__ =head1 NAME Class::LazyObject - Deferred object construction =head1 SYNOPSIS use Class::LazyObject; package Bob::Class::LazyObject; our @ISA = 'Class::LazyObject'; Class::LazyObject->inherit( deflated_class => __PACKAGE__, inflated_class => 'Bob' inflate => sub { my ($class, $id) = @_; return $class->new($id); } ); package main; my @bobs; foreach (0..10_000)#make 10 thousand lazy Bobs { push @bobs, Bob::Class::LazyObject->new($_); } # @bobs now contains lazy objects, not real Bobs. # No Bob objects have been constructed yet. my $single = $bobs[rand @bobs]; #rand returned 10 $single->string;#returns 10. #Single is now an actual Bob object. Only one #Bob object has been constructed. package Bob; #It's really expensive to create Bob objects. sub string { #return the scalar passed to ->new() } #other Bob methods here =head1 DESCRIPTION Class::LazyObject allows you to create lazy objects. A lazy object hol +ds the place of another object, (Called the "inflated object"). The lazy obje +ct turns into the inflated object ("inflates") only after a method is called on + the lazy object. After that, any variables holding the lazy object will hold th +e inflated object. In other words, you can treat a lazy object just like the object it's +holding the place of, and it won't turn into a real object until necessary. Th +is also means that the real object won't be constructed until necessary. A lazy object takes up less memory than most other objects (it's even +smaller than a blessed empty hash). Constructing a lazy object is also likely +to be computationally cheaper than constructing an inflated object (especial +ly if a database is involved). A lazy object can hold a scalar (called the "ID") that is passed to th +e constructor for the inflated object. Note that I believe I've coined the term "lazy object". =head1 WHY When would you want to use lazy objects? Any time you have a large num +ber of objects, but you will only need to use some of them and throw the rest + of them away. =head2 Example For example, say you have a class C<Word>. A Word has a name, a part +of speech, and a definition. Word's constructor is passed a name, and then it fet +ches the other information about the word from a database (which is a dictionar +y and so has thousands of words). C<$word_object-E<gt>others_with_this_pos()> r +eturns an array of all Words in the database with the same part of speech as $wo +rd_object. If you only want to pick 4 words at random that have the same part of +speech as $word_object, hundreds of unnecessary Word objects might be created by C<others_with_this_pos()>. Each of them would require information to b +e retrieved from the database, and stored in memory, only to be destroye +d when the array goes out of scope. It would be much more efficient if C<others_with_this_pos()> returned +an array of lazy objects, whose IDs were word names. Lazy objects take up less +memory than Word objects and do not require a trip to the database when they +are constructed. The 4 lazy objects that are actually used would turn into + Word objects automatically when necessary. =head2 But wait! "But wait," you say, "that example doesn't make any sense! C<others_with_this_pos()> should just return an array of word names. J +ust pass these word names to C<Word>'s constructor!" Well, I don't know about you, but I use object orientation because I w +ant to be able to ignore implementation details. So if I ask for words, I want W +ord objects, not something else representing Word objects. I want to be ab +le to call methods on those Word objects. C<Class::LazyObject> lets you have objects that are almost as small as + scalars holding the word names. These objects can be treated exactly like Word + objects. Once you call a method on any one of them, it suddenly B<is> a word ob +ject. Better yet, you don't have to know about any of this to use the lazy W +ord objects. As far as you know, they B<are> word objects. =head1 SETUP You need to create a lazy object class for each regular class you want + to inflate to. =over 4 =item 1. Create a class to hold lazy objects that inflate to a particu +lar class. package Bob::Class::LazyObject; Note that a package whose name is your package name with ::Methods app +ended (C<Bob::Class::LazyObject::Methods> for this example) is also automati +cally created by Class::LazyObject, so don't use a package with that name fo +r anything. =item 2. Make the class inherit from Class::LazyObject. package Bob::Class::LazyObject; our @ISA = 'Class::LazyObject'; =item 3. Do some configuration Call C<Class::LazyObject-E<gt>inherit()>. It takes a series of named parameters (a hash). The only two required +parameters are C<deflated_class> and C<inflated_class>. See L<"inherit"> for more information. package Bob::Class::LazyObject; our @ISA = 'Class::LazyObject'; Class::LazyObject->inherit( deflated_class => __PACKAGE__, inflated_class => 'Bob' ); When you call C<Class::LazyObject-E<gt>inherit()>, Class::LazyObject s +ets some class data in your lazy object class. =item 4. Create an inflation constructor in the inflated class In the class that the lazy object will inflate to, define a class meth +od C<new_inflate>. This is called with a single parameter, the ID passed + to C<Class::LazyObject-E<gt>new> when this particular lazy object was cre +ated. (If no ID was passed, C<undef> is passed to C<new_inflate>.) This method s +hould be a constructor for your class. It must return an object of the inflated c +lass, or of a class that inherits from the inflated class. (Unless the object i +sa the inflated class, bad things will happen.) If you wish to have the inflation constructor be named something other + than C<new_inflate>, or want it to be called in different way, see L<"THE INFLATE SUB">. The reason C<new_inflate> is called by default rather than just C<new> + is so that you can write C<new> to return lazy objects, unbeknownst to its c +aller. =back That's all it takes to set up a lazy object class. =head1 CLASS METHODS Now that you've set up a lazy object class (if you haven't, see L<"SET +UP">), how do you actually make use of it? The methods here are all class methods, and they must all be called on + a class inherited from C<Class::LazyObject>. If you want to know about object +methods instead, look at L<"OBJECT METHODS">. =head2 C<new> new(ID) new() C<Class::LazyObject-E<gt>new> takes one optional scalar parameter, the + object's ID. This ID is passed to the inflation constructor when the lazy objec +t inflates. Note that the ID I<cannot> be an object of the same class (or any clas +s that inherits from the class) that the lazy object inflates to. =head2 C<inherit> inherit(deflated_class => __PACKAGE__, inflated_class => CLASS) inherit(deflated_class => __PACKAGE__, inflated_class => CLASS, inflate => CODE); #Optional C<Class::LazyObject-E<gt>inherit> should only be called by any class t +hat inherits from Class::LazyObject. It takes a hash of named arguments. O +nly the C<deflated_class> and C<inflated_class> arguments are required. The ar +guments are: =over 4 =item deflated_class B<Required>. The package the lazy object should be in before inflating +, in other words, the class that's calling C<inherit>. You should almost always j +ust set this to C<__PACKAGE__>. =item inflated_class B<Required>. The package the lazy object should inflate into. =item inflate B<Optional>. Takes a reference to a subroutine. This subroutine will b +e called when the lazy object inflates. See L<"THE INFLATE SUB"> for more infor +mation. This allows you to override the default inflation behavior. By default +, when a lazy object inflates, C<Inflated::Class-E<gt>new_inflate> is called an +d passed the lazy object's ID as an argument. =back =head1 OBJECT METHODS None, except an AUTOLOAD that catches calls to any other methods. Calling any method on a lazy object will inflate it and call that meth +od on the inflated object. =head1 THE INFLATE SUB You should pass a reference to a sub as the value of the C<inflate> pa +rameter of L<the C<inherit> class method|"inherit">. This sub is called when the +lazy object needs to be inflated. The inflate sub is passed two parameters: the name of the class to inf +late into, and the ID passed to the lazy object's constructor. The inflate sub should return a newly constructed object. If you supply an inflate parameter to inherit, you override the defaul +t inflate sub, which is: sub {my ($class, $id) = @_; return $class->new_inflate($id);} But you could define your inflate sub to do whatever you want. =head1 IMPLEMENTATION A lazy object is a blessed scalar containing the ID to be passed to th +e inflation constructor. AUTOLOAD is used to intercept calls to methods. + When a method is called on a lazy object, it calls the inflation constructor +on the neccesary class, and sets $_[0] to the newly created object, replacing + the lazy object with the full object. The full object is also stored in the ble +ssed scalar, so that if any other variables hold references to the lazy obj +ect, they can be given the already created full object when they call a method o +n the lazy object. Additional chicanery takes place so that calls to methods inherited fr +om C<UNIVERSAL> are intercepted, and so that C<AUTOLOAD>ed methods of the + inflated object are called correctly. =head1 CAVEATS =head2 The ID cannot be an object of the same class as the inflated cl +ass or any class that inherits from the inflated class. =head2 The C<DESTROY> method does not cause inflation. There's no way (either that, or it's very difficult) to tell whether t +he C<DESTROY> method has been explicitly invoked on a lazy object, or whe +ther Perl is just trying to destroy the object. It is, however, unlikely that yo +u would need to explicitly call C<DESTROY> on any of your objects anyway. I ma +y later add capability to change this behavior. =head2 C<use Class::LazyObject> B<after> C<use>ing any module that put +s subs in C<UNIVERSAL>. C<Class::LazyObject> has to do extra work to handle calls on lazy obje +cts to methods defined in C<UNIVERSAL>. It does this work when you C<use Class::LazyObject>. Therefore, if you add any subs to C<UNIVERSAL> (wi +th C<UNIVERSAL::exports>, C<UNIVERSAL::moniker>, or C<UNIVERSAL::require> +, for example), only C<use Class::LazyObject> B<afterwards>. =head2 Explicitly calling C<AUTOLOAD> on a lazy object may not do what + you expect. If you never explictly call C<$a_lazy_object-E<gt>AUTOLOAD>, this cave +at does not apply to you. (Calling C<AUTOLOAD>ed methods, on the other hand, i +s fine.) If you set $AUTOLOAD in a package with a hardcoded value (because you +think you know in which package the AUTOLOAD sub is defined for a particular cla +ss) and then call C<$a_lazy_object-E<gt>AUTOLOAD>, the object will inflate, bu +t a different method will be called on the inflated object than you intend +ed. If you're trying to spoof calls to AUTOLOAD, you should really be searchi +ng through the inheritance heirarchy of the object (with the help of something li +ke Class::ISA) until you find the package that the object's AUTOLOAD meth +od is defined in, and then set that package's $AUTOLOAD. (In fact, Class::La +zyObject does this kind of AUTOLOAD search itself.) I will most likely revise this caveat to make more sense. =head1 BUGS (The difference between bugs and L<caveats|"CAVEATS"> is that I plan t +o fix the bugs.) =head2 Objects with C<overload>ed operators Currently, lazy objects will not intercept overloaded operators. This +means that if your inflated object uses overloaded operators, you cannot use a la +zy object in its place. This may be fixed in future versions by using a combinat +ion of C<nomethod> and C<overload::Method>. See L<overload> to learn more abo +ut overloaded operators. =head2 C<UNIVERSAL::isa> and C<UNIVERSAL::can> Currently, C<UNIVERSAL::isa($a_lazy_object, 'Class::The::Lazy::Object::Inflates::To')> is false, though C<$a_lazy_object-E<gt>isa> will do the right thing. Similarly, C<UNIVERSAL::can($a_lazy_object, 'method')> won't work like it's suppo +sed to, but C<$a_lazy_object-E<gt>can> I<will> work correctly. This may be fix +ed in a future release. =head2 Objects implementing C<tie>d datatypes C<Class::LazyObject> has not yet been tested with objects that implime +nt C<tie>d datatypes. It may very well work, and then again, it may not. Explicit + support may be added in a future release. See L<perltie> to learn more about C +<tie>s. =head1 AUTHOR Daniel C. Axelrod, =head1 SEE ALSO =head2 Fergal Daly had the idea for lazy objects before I did. Note that I ha +d the idea independently, but subsequently discovered his posting. =head2 L<Class::Data::TIN> =head2 L<Class::ISA> =head1 COPYRIGHT Copyright (c) 2003, Daniel C. Axelrod. All Rights Reserved. This program is free software; you can redistribute it and/or modify i +t under the same terms as Perl itself. =cut

Edit s/create objects/create lazy objects/;

Once it's Turing complete, everything else is just syntactic sugar.

Replies are listed 'Best First'.
Re: RFC - Class::LazyObject
by lachoy (Parson) on Oct 08, 2003 at 03:34 UTC
    I hate to be a stinker after you've put in what's obviously a good bit of work, but this appears to do pretty much the same thing as Object::Realize::Later. Is there a difference?

    M-x auto-bs-mode

      *mumble grumble mumble*
      I searched CPAN for hours and didn't find that!

      After quickly looking at it, it looks like Object::Realize::Later impliments the same sort of thing, but goes about it in a very different way, and with a different philosophy.

      Class::LazyObject is all about invisiblity. An assumption I made in Class::LazyObject is that the Lazy Object could be used invisibly in the place of an inflated object, meaning it could be easily grafted on to a program even after the program had been designed. Object::Realize::Later seems to require you to have planned to use it, restricting which method names you can use in your inflated objects. (It looks like it pollutes its lazy objects' namespaces with methods named "becomes", "believe_caller", "realize", "source_module", "warn_realization", "forceRealize", "willRealize" and "warn_realize_again ", this could be problematic if the inflated class uses methods with these names.)

      An even greater major difference is in how copies of lazy objects are handled. (Since objects are really just refs, after $object2 = $object1, both variables refer to the same object, and any changes to $object1 will be reflected in $object2). This means that, after $lazy_object2 = $lazy_object1, if $lazy_object1 is inflated, $lazy_object2 should be a reference to the inflated object. While Object::Realize::Later leaves the user to worry about how to go about implimenting this (thought it does give suggestions in the documenation), Class::LazyObject specifically handles this for the user, ensuring that the correct behavior always occurrs. Additionally, it does this without reblessing the lazy object.

      Another thing that O::R::L doesn't appear to be able to do is handle AUTOLOADed methods in the inflated class. Class::LazyObject can handle them.

      Additionally, O::R::L doesn't appear to have support for overloaded operators on lazy objects. I am already working on automatic operator overloading of lazy objects, which should be in a future release.

      While it will be worth contacting the author of Object::Realize::Later, it seems like our two modules are headed in slightly different paths. I can think of several instances in which Class::LazyObject would fit but Object::Realize::Later would not. I still think Class::LazyObject warrants its own place on CPAN.

      Thank you for finding Object::Realize::Later, though.

      Edit:Added more information.

      Once it's Turing complete, everything else is just syntactic sugar.
        If you decide to upload to CPAN, you should definately include a compare/contrast to Object::Realize::Later in the POD/README. Contact the author first to make sure what you are saying is accurate. This will help people decide which module is right to use in what situation.

        Cheers - L~R

        That's cool -- it's just a sad thing when you can't find something on CPAN that could have an impact on the work you're doing, quirky naming schemes and all. Best of luck!

        M-x auto-bs-mode

        Hi, I'm the author of O::R::L. You may want to have a look at the presentations I gave on various YAPCs as well: may help. ORL is also invisible. Even isa() and can() are fake. The use of 'use XYZ keyword' does NOT add 'keyword' to your namespace: it only calls XYZ::import(). When Exporter is used in XYZ, then the name-space is poluted, but not in the ORL case. About copy: my docs must be unclear. It explains that there is the danger that there are two references to the UNinflated object. When one handle gets inflated, the other may not. Your 'inflation' method must take care about it. Not handling AUTOLOAD? Why not? It should do that. Overloading is a nice idea, not in ORL... however overloading is quite nasty. You can use overloading on the lazy objects, but not automatically simulate the overloading of the inflated. Be aware that ORL has already proven itself in various applications: it works. For instance, my Mail::Box module uses all corners of its power. As far as I can see, both modules try to do the same. But I don't mind an alternative implementation.
Re: RFC - Class::LazyObject
by diotalevi (Canon) on Oct 09, 2003 at 16:01 UTC

    This reminds of nothing so much as E's promises. Neat idea actually. Just for inspiration I'd like to point you to the (short) E in a Walnut document.

Log In?

What's my password?
Create A New User
Node Status?
node history
Node Type: perlmeditation [id://297466]
Approved by virtualsue
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others examining the Monastery: (3)
As of 2020-06-02 18:05 GMT
Find Nodes?
    Voting Booth?
    Do you really want to know if there is extraterrestrial life?

    Results (19 votes). Check out past polls.