What is the default parameter for AUTOLOAD in Perl? - perl

I've been playing with AUTOLOAD to create my accessors in Perl and I have encountered this confusion (I have searched google and perldoc already).
I have this code:
package Class;
sub new {
..code for constructor here.
}
sub AUTOLOAD {
my $name= shift;
print $name;
}
But when I do something like : my $a=Class->new; The autoload subroutine still executes, and prints Class=HASH(some weird number);
I thought AUTOLOAD only runs when there is an undefined method or subroutine?
And also I did this:
my $class = our $AUTOLOAD;
print $class #prints ::DESTROY
Am I right when I assumed that DESTROY is the value of $AUTOLOAD when no undefined function is passed?

Using Autoload is inherently difficult. If you want a solid object system that makes accessors for you then please use Moose, Mouse, Moo, or just loop over your fields and install the accessors yourself:
BEGIN {
my #fields = qw/foo bar baz/;
for my $field (#fields) {
no strict 'refs';
# install a closure in the package stash.
*{ __PACKAGE__ . "::" . $field } = sub {
my $self = shift;
$self->{$field} = shift if #_;
return $self->{$field};
};
}
}
If a class that can AUTOLOAD encounters an undefined method, the AUTOLOAD sub is called with the arguments of the missing sub. The fully qualified name of the requested sub is passed in the $AUTOLOAD package variable.
A typical Autoload sub would look like:
use Carp;
my %fields_allowed = map {$_ => 1} qw/foo bar baz/;
sub AUTOLOAD {
my $field = our $AUTOLOAD;
$field =~ s/.*:://; # strip the package name
$fields_allowed{$field}
or croak qq(Can't locate object method $field via package "#{[__PACKAGE__]}");
my $self = shift;
$self->{$field} = shift if #_;
return $self->{$field};
}
There remain two problems:
When the reference count of an object drops to zero, or when a thread terminates, the DESTROY method is called on the object if it provides one. We can prevent autoloading of DESTROY by providing an empty implementation: sub DESTROY {}.
We can ask any object if it can perform a certain method, like say "Good dog" if $dog->can("roll"). Therefore, we have to override can to support our autoloading. The can method is useful for safe duck typing. Every object inherits from UNIVERSAL, which provides default implementations for can and isa.
The contract of can is that it takes the name of a method. It will return undef when the object cannot perform the method, or a code reference to that method if it can. A suitable implementation would be
sub can {
my ($self, $name) = #_;
# check if it's a field of ours
if ($fields_allowed{$name}) {
return sub {
my $self = shift;
$self->{$name} = shift if #_;
return $self->{$name};
};
}
# Ask SUPER implementation of can if we can do $name
if (my $meth = $self->SUPER::can($name)) {
return $meth;
}
return; # no method found
}
We can now simplify AUTOLOAD to
sub AUTOLOAD {
my $field = our $AUTOLOAD;
$field =~ s/.*:://; # strip the package name
my $code = $self->can($field)
or croak qq(Can't locate object method $field via package "#{[__PACKAGE__]}");
goto &$code; # tail call; invisible via `caller()`.
}
This is a lot of complexity to get right. Verdict: Don't use Autoload because you think it might be less work. It never is. It is quite useful for implementing a proxy pattern, but that is a bit advanced.
I urge you to dabble around with OO basics, and the Moose object system, before diving deep into Perl's unique and strange features.

Related

Catching undefined value accessing in Perl

How can I catch access to member variables?
$Class1->{Class2}
If the Class2 field doesn't exist, is is possible to catch this from an internal function?
You can, but you probably shouldn't. The problem here is - if you access a variable within a class directly... then you just can. You can prevent this with a couple of workarounds - and this is where things like Moose come in.
And there's a couple of slightly hacky tricks like inside-out objects (which I think aren't common practice any more - Perl Best Practice advocated them some years back) or using anonymous hashes to hold state.
But failing that - why not use an accessor, and auto-generate one using 'AUTOLOAD'.
#!/usr/bin/env perl
package MyClass;
use strict;
use warnings;
use vars '$AUTOLOAD';
sub AUTOLOAD {
my ( $self ) = #_;
my $subname = $AUTOLOAD =~ s/.*:://r;
if ( $self -> {$subname} ) {
return $self -> {$subname};
}
warn "Sub called $subname was called\n";
return "$subname";
}
sub new {
my ( $class ) = #_;
my $self = {};
bless $self, $class;
}
package main;
use strict;
use warnings;
my $object = MyClass -> new;
$object -> {var} = "fleeg";
print "Undef fiddle was: ", $object -> fiddle,"\n";
print "But 'var' was: ", $object -> var,"\n";
This has the same problem, in that changing method names might cause things to break. However it has the advantage that you can handle 'invalid' method calls however you like.
But really - explicit 'get' and 'set' methods are better choices for most use-cases.
You do this by providing proper getter/setter methods that wrap around your class/instance variables. The internals should never be accessed directly, particularly from outside of the class itself (it's wise to not do so within the class either, except for the actual method that maintains that specific attribute. Here's a very basic example:
use warnings;
use strict;
package A;
sub new {
my ($class, %args) = #_;
my $self = bless {}, $class;
$self->x($args{x});
$self->y($args{y});
return $self;
}
sub x {
my ($self, $x) = #_;
$self->{x} = $x if defined $x;
return $self->{x} // 1;
}
sub y {
my ($self, $y) = #_;
$self->{y} = $y if defined $y;
return $self->{y} // 2;
}
package main;
my $obj = A->new(x => 5, y => 3);
print $obj->x ."\n";
print $obj->y ."\n";
Now, you could just as easily do print $obj->{x}, but that's where your problem is. What happens when the code is much more complicated than this, and for some reason you want to change the x attribute name to foo, but retain the x() method? $obj->{x} will now be undef as its never set.
Always use the provided methods for accessing attributes of a class/object. Encapsulation such as this is a staple of OO programming.

Using Perl's Method::Signatures, why can't I invoke methods on an object instance?

I followed what friedo said here.
Now, when I try to call the method testScript I get the error global symbol $obj requires explicit package name and it fails to call testScriptTwo.
use strict;
use warnings;
package Test;
use Method::Signatures;
method new {
my $obj = bless {}, $self;
return $obj;
}
method testScript {
$obj->testScriptTwo(); # Error happens here
}
method testScriptTwo { ... }
Test script:
use Test;
my $class = Test->new();
$class->testScript();
How do I make use of $obj to call methods within the package itself?
Use this instead:
method testScript {
$self->testScriptTwo();
}
The first argument is in the variable $self, not $obj
Your questions seem to indicate you do not understand the basics of scope, and how plain Perl objects work.
In Perl, when you use the ->method syntax on a package name or blessed reference, the subroutine method in that package is invoked. The first argument to the subroutine is the thing on which you invoked method.
So, if you do
My::Friend->new('Alfred');
the new subroutine in the package My::Friend receives two arguments. My::Friend and Alfred.
In a new method, it is customary to refer to the first argument as $class, but that is completely up to you. You could use $basket_case if you were so inclined:
sub new {
my $basket_case = shift;
my $basket = shift;
my $obj = bless { name => $basket } => $basket_case;
return $obj;
}
If you then invoke a method on the returned reference, that method will receive said reference as its first argument, allowing you to access data stored in that reference:
sub blurb {
my $schmorp = shift;
print $schmorp->{name}, "\n";
return;
}
Putting it all together:
#!/usr/bin/env perl
package My::Package;
use strict;
use warnings;
sub new {
my $basket_case = shift;
my $basket = shift;
my $obj = bless { name => $basket } => $basket_case;
return $obj;
}
sub blurb {
my $schmorp = shift;
print $schmorp->{name}, "\n";
return;
}
sub derp {
my $herp = shift;
printf "%s derp derp\n", $herp->{name};
return;
}
package main;
my $x = My::Package->new('Alfred');
$x->blurb;
$x->derp;
Output:
Alfred
Alfred derp derp
You need to understand these basics. Trying to put another layer of abstraction on top of the basics before understanding what is underneath will not make things any easier.
Now, if you are using Method::Signatures, it, by convention, puts that implicit first argument in a lexically scoped variable which, by default, it calls $self.
You can override that name in specific methods, and doing so in new might be a good idea to convey the fact that it doesn't expect an object instance; instead it returns a new instance.
Whatever you called that lexically scoped instance variable in one sub does not affect what it is called in another sub. For example:
#!/usr/bin/env perl
use strict;
use warnings;
sub a_number {
my $number = int(rand(10));
return $number;
}
sub square_that_number {
my $x = shift;
return $x * $x;
}
my $bzzzt = a_number();
my $trrrp = square_that_number($bzzzt);
print $trrrp, "\n";
Output:
$ ./zt.pl
36
OK, you need to backtrack a bit - you're new method is broken in the first place, which indicates that you don't really understand what's going on with OO perl.
A very simple object looks like this:
package Foo;
sub new {
#when Foo -> new is called, then 'Foo' is passed in as the class name
my ( $class ) = #_;
#create an empty hash reference - can be anything, but $self is the convention
my $self = {};
#tell perl that $self is a 'Foo' object
bless ( $self, $class );
#return the reference to your `Foo` object
return $self;
}
sub set_name {
my ( $self, $new_name ) = #_;
$self -> {name} = $new_name;
}
sub get_name {
my ( $self ) = #_;
return $self -> {name};
}
When you call this in your code:
use Foo;
my $new_instance = Foo -> new();
The class is passed into the new method, which you then use bless to create an instantiated object.
Then you can 'do stuff' with it - when you 'call' a method using -> then the first argument into the subroutine is the object reference.
So
$new_instance -> set_name ( "myname" );
print $new_instance -> get_name();
Is equivalent to:
Foo::set_name($new_instance, "myname" );
print Foo::get_name($new_instance);
You act on $new_instance which is a sort of magic hash that allows you to include code.
Method::Signatures is largely irrelevant until you understand the basics of OO. But what that does is 'simply' expand the functions within a module, such that you don't have to extract self/class etc.
By default, a method defined as method provides $self automatically. no $obj like you're using. That's a variable that's local to you new method, and simply doesn't exist outside that.

Is it possible to get all valid methods for a particular Perl class?

Is it possible to get all valid methods for a particular Perl class?
I am trying to manipulate the symbol table of a class and get all of its methods. I found I can separate out the subroutines from the non-subroutines via the $obj->can($method), but that doesn't do exactly what I think it does.
The following returns:
subroutine, Property, croak, Group, confess, carp, File
However, subroutine isn't a method, (just a subroutine), and croak, confess, and carp were all imported into my package.
What I really want to print out is:
Property,Group, File
But I'll take:
subroutine, Property,Group, File
Below is my program:
#! /usr/bin/env perl
use strict;
use warnings;
use feature qw(say);
my $sections = Section_group->new;
say join ", ", $sections->Sections;
package Section_group;
use Carp;
sub new {
return bless {}, shift;
}
sub Add {
my $self = shift;
my $section = shift;
}
sub Sections {
my $self = shift;
my #sections;
for my $symbol ( keys %Section_group:: ) {
next if $symbol eq "new"; # This is a constructor
next if $symbol eq "Add"; # Not interested in this method
next if $symbol eq "Sections"; # This is it's own method
push #sections, $symbol if $self->can($symbol);
}
return wantarray ? #sections : \#sections;
}
sub subroutine {
my $param1 = shift;
my $param2 = shift;
}
sub Group {
my $self = shift;
my $section = shift;
}
sub File {
my $self = shift;
my $section = shift;
}
sub Property {
my $self = shift;
my $section = shift;
}
This is fairly trivial. We only want to keep those sub names that were originally defined in our package. Every CV (code value) has a pointer to the package where it was defined. Thanks to B, we can examine that:
use B ();
...
if (my $coderef = $self->can($symbol)) {
my $cv = B::svref_2object $coderef;
push #sections, $symbol if $cv->STASH->NAME eq __PACKAGE__;
}
# Output as wanted
That is, we perform introspection using svref_2object. This returns a Perl object representing an internal perl data structure.
If we look into a coderef, we get a B::CV object, which represents the internal CV. The STASH field in a CV points to the Stash where it was defined. As you know, a Stash is just a special hash (internally represented as a HV), so $cv->STASH returns a B::HV. The NAME field of a HV contains the fully qualified package name of the Stash if the HV is a Stash, and not a regular hash.
Now we have all the info we need, and can compare the wanted package name to the name of the stash of the coderef.
Of course, this is simplified, and you will want to recurse through #ISA for general classes.
Nobody likes polluted namespaces. Thankfully, there are modules that remove foreign symbols from the Stash, e.g. namespace::clean. This is no problem when the CVs of all subs you are calling are known at compile time.
What are you trying to do? Why does it matter how a class defined or implements a method it responds to?
Perl is a dynamic language, so that means that methods don't have to exist at all. With AUTOLOAD, a method might be perfectly fine and callable, but never show up in the symbol table. A good interface would make can work in those cases, but there might be cases where a class or an object decides to respond to that with false.
The Package::Stash module can help you find defined subroutines in a particular namespace, but as you say, they might not be defined in the same file. The methods in a class might come from an inherited class. If you care about where they come from, you're probably doing it wrong.

How to create (or not) class instance methods at construction time based on inputs?

How would I create my class such that some methods will exist in the instance only if certain values were passed to the constructor?
Perhaps a more generic way of asking is: How can I add a method to an existing class instance?
You can attach an anonymous sub to an object based on flags:
use strict;
use warnings;
package Object;
sub new {
my $class = shift;
my $self = bless {}, $class;
my %args = #_;
if ($args{method}) {
$self->{method} = sub { print "hello\n" }
}
return $self;
}
sub method {
my $self = shift;
if (not defined $self->{method}) {
warn "Not bound\n";
return;
}
$self->{method}->();
}
1;
to use:
use Object;
my $obj1 = Object->new(method=>1);
$obj1->method();
my $obj2 = Object->new();
$obj2->method();
You can extend this to a number of methods through the same interface.
You can use Moose to apply a role at runtime.
package My::Class;
use Moose;
has foo => ( isa => 'Str', is => 'ro', required => 1 );
sub BUILD {
my $self = shift;
if ($self->foo eq 'bar') {
My::Class::Role->meta->apply($self);
}
}
no Moose;
package My::Class::Role;
use Moose::Role;
sub frobnicate {
my $self = shift;
print "Frobnicated!\n";
}
no Moose;
my $something = My::Class->new( foo => 'bar' );
print $something, "\n";
$something->frobnicate;
my $something_else = My::Class->new( foo => 'baz' );
print $something_else, "\n";
$something_else->frobnicate;
Gives:
Moose::Meta::Class::__ANON__::SERIAL::1=HASH(0x2fd5a10)
Frobnicated!
My::Class=HASH(0x2fd2c08)
Can't locate object method "frobnicate" via package "My::Class" at testmoose.pl line 32.
use AUTOLOAD to define the function. As a example method foo is called if $self->{foo} exists
sub AUTOLOAD {
my $methodname = $AUTOLOAD;
if ($methodname eq "foo" && exists($_[0]->{foo})){
goto &fooimplementationsub;
}
return;
}
An alternative technique is to use globs to define a new method at runtime
*PACKAGE::method = sub {
#code here
};
This has the disadvantage that the method is now visible to all instances of the class so is not quite what you want.
A third and possibly more risky/inefficient method is to use string eval
eval <<EOF
sub foo {
#code here
};
EOF
Again this has the disadvantage that the method is now visible to all instances of the class so is not quite what you want.
Methods are just subroutines in a package, and a package is just a hash holding typeglobs. And hashes can be modified at runtime.
So you could, in theory, add or remove methods given values in a constructor.
package WeirdClass;
sub new {
my ($class, $name, $code) = #_;
if ($name) {
no strict;
*{__PACKAGE__ . "::$name"} = $code;
}
bless {} => $class;
}
And then use it like:
my $object = WeirdClass->new(foo => sub {say "foo"});
$object->foo(); # prints "foo\n";
However, this method is available for all objects of that class:
my $another_object = WeirdClass->new();
$another_object->foo; # works too.
Using autoload, one can mock arbitrary methods:
package BetterClass;
sub new {
my ($class, %args) = #_;
bless \%args => $class;
}
# destructor will be called at cleanup, catch with empty implementation
sub DESTROY {};
sub AUTOLOAD {
my $self = shift;
(my $method = our $AUTOLOAD) =~ s/.*://; # $AUTOLOAD is like "BetterClass::foo"
# check if method is allowed
die "forbidden method $method" unless $self->{can}{$method};
# mock implementations
given ($method) {
say "foo" when "foo";
say "bar" when "bar";
when ("add") {
my ($x, $y) = #_;
return $x + $y;
}
default { die "unknown method $method" }
}
}
Then:
my $o = BetterClass->new(can => { foo => 1, bar => 0});
$o->foo;
my $p = BetterClass->new(can => {bar => 1, add => 1});
$p->bar;
say $p->add(5, 6);
Of course, these techniques can be combined freely.
Edit: can()
To make the AUTOLOAD work with can, the protected methods should be moved into a data structure:
my %methods;
BEGIN {
%methods = (
foo => sub {say "foo"},
bar => sub {say "bar"},
add => sub {
my ($self, $x, $y) = #_;
$x + $y;
},
);
}
Then override the can method:
# save a reference to the origional `can` before we override
my $orig_can;
BEGIN{ $orig_can = __PACKAGE__->can("can") }
sub can {
my ($self, $meth) = #_;
# check if we have a special method
my $code = $methods{$meth} if ref $self and $self->{can}{$meth};
return $code if $code;
# check if we have a normal method
return $self->$orig_can($meth);
}
And AUTOLOAD would change to
my ($self) = #_; # do not `shift`
(my $method = our $AUTOLOAD) =~ s/.*://;
my $code = $self->can($method) or die "unknown method $method";
goto &$code; # special goto. This is a AUTOLOAD idiom, and avoids extra call stack frames
Don't do too much magic. I've gotten away from AUTOLOAD because it causes maintenance issues where mysterious methods suddenly appear and disappear.
One way to handle what you want is to define all the methods you need, and if a particular object is of the wrong type, simply cause that method to croak:
sub Foo {
my $self = shift;
my $parameter = shift;
if ( $self->Class_type ne "Foo" ) {
croak qq(Invalid method 'Foo' on object #{[ref $self]});
}
print "here be dragons\";
return "Method 'Foo' successfully called";
}
The above will not allow method Foo to be called unless the class type is Foo.
If your objects won't change (or you don't want them to change) once an object is created, you can define that object as a sub-class.
Before you bless a newly created object, check that special value and decide whether or not you need to create a specific sub-class instead.
package My_class;
sub new {
my $class = shift;
my $class_type = shift;
my $self = shift;
if ( $class_type eq "Foo" ) {
bless $self, "My_class::Foo";
}
else {
bless $self, $class;
}
package My_class::Foo;
use base qw(My_class);
sub Foo {
my $self = shift;
return "Foo Method successfully called!";
}
Notice that my class My_class::Foo is a sub-class of My_class via the use base pragma. That means all methods for My_class are valid with objects of My_class::Foo. However, only objects of My_class::Foo can call the Foo method.
When I create my object (via the new subroutine), I look at the $class_type parameter. If it's a type Foo, I bless the class as My_class::Foo.
Here's an example where I use sub-classes to do what you want.
Every object is a class type of Question. You can see my constructor on line 1129. I pass in a question type as one of the parameters to my constructor.
In line 1174 to 1176, I create my object, but then append the question type to the class, and then bless the question as that sub-class type. All of my subclasses are a type Question (see my use base qw(Question); below each package declaration. However, only questions of sub-class Question::Date and Question::Regex have a method Format. And, only objects of type Question::Words have a method Force.
Hope this helps.
None of the answers so far given actually handle the question actually asked.
Adding methods to an instance in Perl is not directly supported. Object instances are always instances of some class, and that class is the thing that actually has methods. You cannot add a method to a single instance of a class, without making that method also available on every other instance of the same class.
For your problem you have two basic solutions:
Provide the methods always, but test a flag to see whether the method should apply to the given instance or not. This is by far the simplest.
Bless each object into subclasses depending on the flags. Subclass the main class to provide those methods as appropriate.
If you truely want to add methods on individual instances, then what you'll have to do is arrange that every instance is a single instance of a newly-derived class for every object. This gets harder to arrange for, doubly-so if you want to avoid leaking memory and cleaning up the classes once the objects are DESTROYed. This would however allow truely per-instance methods.
Since it is highly unlikely you'll truely need this third option it is far better to go with one of the first.

Perl encapsulate class variable?

I'm pretty new to perl, and I'm getting stuck on a homework problem. I have an object with a class variable that counts the number of instances created. Then I have a subclass with an instance variable.
My first question is, how do I make the class variable hidden from the user? I tried using closures but couldn't figure out how to make inheritance work with that. And the fact that it's a class variable made it worse because the code that increments it executed twice and it said I had two instances when I had one. Not exactly sure why it happened but it makes sense. I tried using scalars but the variable again wasn't incrementing correctly. Haven't tried "inside-out objects" yet and I'm not sure I want to, it seems way over my head. I'm getting the feeling that encapsulating class variables is different than encapsulating instance variables, but I can't find anything that explains how to do it.
My second questions is, as I mentioned, I can't get encapsulation to work with inheritance. With closures when you call the super constructor from the subclass you get a reference to the subroutine right, so there's no way (that I know of) to add the instance variables to that.
Here's my base class:
#!/usr/bin/perl -w
use strict;
package Base;
my $count = 1;
sub new {
my $class = shift;
my $self = {
_Count => $count # not hidden
};
$count++; # increment count
bless $self, $class;
return $self;
}
sub Count { # getter
my $self = shift;
return $self->{_Count};
}
1;
Here's my subclass:
#!/usr/bin/perl -w
use strict;
package Sub;
use Base;
our #ISA = qw(Base);
sub new {
my $class = shift;
my $self = $class->SUPER::New();
$self->{_Name} = undef; # not hidden
return $self;
}
sub Name { #getter/setter
my($self, $name) = #_;
$self->{_Name} = $name if defined($name);
return $self->{_Name};
}
1;
If you are using bare Perl 5 (rather than employing an OO framework), the usual way to do class variables is as a lexical visible only to the accessor:
{
my $count = 0;
sub Count {
my ($self, $new_count) = #_;
if (defined $new_count) { # NB only works if undef is not a legit value
$count = $new_count;
}
return $count;
}
}
$count is only visible in the enclosing block; not even other methods on the same class can see it. But anyone can manipulate it with either $base_obj->Count or Base->Count, and any such manipulation will affect the shared variable.
You can also employ closure to provide really-hidden instance variables. This is not worth doing unless you are fulfilling the arbitrary rules of a homework assignment.
package Base;
sub new {
my ($class, $name) = #_;
die "Need name!" unless defined $name;
my $age;
return bless sub {
my ($attribute, #args) = #_;
if ($attribute eq 'name') {
if (#args) {
die "Attempt to set read-only attribute!";
}
return $name;
}
if ($attribute eq 'age') {
if (#args) {
($age) = #args;
}
return $age;
}
die "Unknown attribute $attribute";
} => $class;
}
sub name {
my ($self, #args) = #_;
return $self->(name => #args);
}
sub age {
my ($self, #args) = #_;
return $self->(age => #args);
}
What happens here is that the blessed sub returned by new closes over two lexicals, $name and $age. When new returns, those lexicals go out of scope and the only way to access them from that point forward is through the closure. The closure can inspect its arguments to permit or deny access to the values it holds. So long as it never returns a reference, it can be sure that it has the only direct access to those variables.
This works with inheritance, too, without too much added subtlety:
package Derived;
use base 'Base';
sub new {
my ($class, $name, $color) = #_;
my $base_instance = $class->SUPER::new($name);
return bless sub {
my ($attribute, #args) = #_;
if ($attribute eq 'color') {
if (#args) {
($color) = #args;
}
return $color;
}
# base class handles anything we don't, possibly by dying
return $base_instance->($attribute, #args);
} => $class;
}
This emulates what languages with distinct storage for base- and derived-class instance data do, either handling the request locally or passing it on to the base class instance, which has been added to the closure. Deeper inheritance trees will result in closures that close over closures that close over closures, each of them optionally also closing over instance variables needed by that particular class.
This is a pretty big mess to produce and really hard to inspect and debug, which is why I'm going to emphasize one more time that you should never do this. But it is very useful to understand, to which end I refer you to SICP.
As a module-local my variable, $count is already hidden from users of the module/class. It appears as if you're using instance variable _Count as a "current ID" type variable, so that each object (instance) created gets a new ID starting from 1. (If instead it is meant to track the number of active instances, then you need to decrement it in DESTROY and there's no need to store a copy in the object.) If your test code is only creating one instance then its Count() method should return 1 but $count will be 2, since it started as 1 and was incremented after storing the old value in the object.
It is typical in perl to store instance variables in the $self hash as you are doing, without hiding them, although sometimes a prefix is used to avoid collisions. They are protected more by convention (it's not safe to rely on implementation details because they might change) than language features.
Take a look at the Moose suite of modules if you want higher-level control over perl classes.
To quote perldoc perlmodlib, "Perl does not enforce private and public parts of its modules as you may have been used to in other languages like C++, Ada, or Modula-17. Perl doesn't have an infatuation with enforced privacy. It would prefer that you stayed out of its living room because you weren't invited, not because it has a shotgun."
The standard convention in Perl is to put everything into the $self hash and use an underscore prefix to indicate which items should be treated as private... and then trust users of the class to respect that indication. The same convention is also applied to methods. If you use one of my modules and you choose to peek under the covers and modify the contents of $self directly or call $obj->_some_private_method, then you're going off into the woods and may break something, or what works fine in this version may break when you upgrade to the next version; if that happens, you get to keep both pieces.
If you're going to insist on making data inaccessible to anyone outside the class itself, there are ways to do that, but a) they add complexity which is, in almost all cases, unnecessary and b) as you've already seen, they have a tendency to make inheritance a lot more of a hassle to work with.
My question to you, then, is what are you actually attempting to accomplish and why do you feel the need to make your object data Sooper-Sekret and completely inaccessible? What benefit will you gain by doing so which isn't provided by simply marking things that you think should be treated as private, then trusting others to leave them alone (unless they have good reason to do otherwise)?
In Perl, fields are not usually hidden by enforcing this through the semantics of the language, but rather through a contract in the form of documentation. However, fields can be hidden through the use of closures. It is also worth noting that Perl does not semantically differentiate between class methods and instance methods.
One of the standard ways to implement objects is a blessed hash, like you do. This hash contains all instance variables / fields. It is customary to start "private" fields with an underscore. Usually, the contract (the documentation) will not state how these fields are stored, but will require the user of the class to go through various method calls.
Class variables should not be stored with the instance. It is better to use global variables, or lexical variables. In the code you gave, $count is just a counter, but you never access it as a class variable. Instead, you assign each instance an unique ID. To use it as a class variable, provide an appropriate accessor (I stripped out unneccessary stuff like returns):
{
package Base;
my $count = 0;
sub new {
my ($class) = #_;
my $self = {
ID => $count++,
};
bless $self, $class;
}
sub Count { $count }
sub ID { my ($self) = #_; $self->{ID} }
sub report { my ($self) = #_; "I am the Base object ".($self->ID)."." }
}
=head1 Base
A generic base class
=head2 Base->Count
Return the object count.
=head2 $base->ID
Give the unique ID of this object.
=head2 $base->report
Returns a string containing a short description.
=cut
The subclass has no business meddling with the count. This is enforced by the scope of the variable $count above, denoted via the outer curly braces. The subs are closures over this variable.
{
package Sub;
use parent -norequire, qw(Base); # remove `-norequire` if Base in different file
sub new {
my ($class) = #_;
my $self = $class->SUPER::new;
$self->{Name} = undef;
$self;
}
sub Name :lvalue {
my ($self) = #_;
$self->{Name};
}
sub report {
my ($self) = #_;
"I am the Sub object ".($self->ID)." called ".($self->Name).".";
}
}
=head1 Sub
A generic subclass. It subclasses Base.
=head2 $sub->Name [= SCALAR]
Gets or sets the name of $sub.
my $oldname = $sub->Name;
$sub->name = "new name";
=cut
As you can see, the Sub constructor calls the Base initializer, then adds a new field. It has no class methods or class variables. The class has no access to the $count variable, except via the accessor class method. The contract is stated via POD documentation.
(In the Name method, I use an :lvalue annotation. This allows me to simply assign to the appropriate field in the object. However, this disallows argument checking.)
The testcase
my $base1 = Base->new; my $base2 = Base->new;
print "There are now " . Base->Count . " Base objects\n";
my $sub1 = Sub->new; my $sub2 = Sub->new;
print "There are now " . Base->Count . " Base objects\n";
$sub2->Name = "Fred";
print $_->report . "\n" for ($base1, $sub1, $base2, $sub2);
prints
There are now 2 Base objects
There are now 4 Base objects
I am the Base object 0.
I am the Sub object 2 called .
I am the Base object 1.
I am the Sub object 3 called Fred.
Beautiful, isn't it? (Except $sub1, that object is missing its name.)
The documentation can be viewed with perldoc -F FILENAME, and would output something like
Base
A generic base class
Base->Count
Return the object count.
$base->ID
Give the unique ID of this object.
$base->report
Returns a string containing a short description.
Sub
A generic subclass. It subclasses Base.
$sub->Name [= SCALAR]
Gets or sets the name of $sub.
my $oldname = $sub->Name;
$sub->name = "new name";
only typeset more nicely, if you are on a *nix system.
Tested under v5.12.4.
Edit: Inside-out objects
While inside-out objects provide better encapulation, they are a bad idea: difficult to understand, difficult to debug, and difficult to inherit they provide more problems than solutions.
{
package Base;
my $count = 0;
sub new { bless \do{my $o = $count++}, shift }
sub Count { $count }
sub ID { ${+shift} }
sub report { my ($self) = #_; "I am the Base object ".($self->ID)."." }
}
{
package Sub;
my #_obj = ();
my $count = 0;
sub new {
my ($class) = #_;
$count++;
$_obj[$count - 1] = +{
parent => Base->new(),
Name => undef,
};
bless \do{my $o = $count - 1}, shift;
}
sub Name :lvalue { $_obj[${+shift}]{Name} }
sub AUTOLOAD {
my $self = shift;
my $package = __PACKAGE__ . "::";
(my $meth = $AUTOLOAD) =~ s/^$package//;
$_obj[$$self]{parent}->$meth(#_)
}
sub report {
my ($self) = #_;
"I am the Sub object ".($self->ID)." called ".($self->Name).".";
}
}
This implementation has the exact same interface, and completes the test case with the same output. This solution is far from optimal, supports only single inheritance, does some intermediate stuff (autoloading, dynamic method calls), but it does suprisingly work. Each object is actually just a reference to an ID that can be used to look up the actual hash containing the fields. The array holding the hashes is not accessible from the outside. The Base class has no fields, therefore no object array had to be created.
Edit2: Objects as coderefs
Yet another bad idea, but it is fun to code:
{
package Base;
my $count = 0;
sub new {
my ($class) = #_;
my $id = $count++;
bless sub {
my ($field) = #_;
die "Undefined field name" unless defined $field;
if ($field eq "ID") { return $id }
else { die "Unrecognised name $field" }
}, $class;
}
sub Count { $count }
sub ID { my ($self) = #_; $self->("ID") }
sub report { my ($self) = #_; "I am the Base object " . $self->ID . "." }
}
{
package Sub;
use parent -norequire, qw(Base);
sub new {
my ($class) = #_;
my $name = undef;
my $super = $class->SUPER::new;
bless sub {
my ($field, $val ) = #_;
die "Undefined field name" unless defined $field;
if ($field eq "Name") { defined $val ? $name = $val : $name }
else { $super->(#_) }
}, $class;
}
sub Name { my $self = shift; $self->("Name", #_) }
sub report {
my ($self) = #_;
"I am the Sub object ".($self->ID)." called ".($self->Name).".";
}
}
The test case has to be adapted to $sub2->Name("Fred"), and the documentation updated accordingly, as we cannot use an lvalue annotation here safely.
First, I'm not sure exactly what you mean by "hidden from the user", but it looks like you may be looking for package scoped variables (our) vs. instance scoped.
package MyBaseClass;
use warnings;
use strict;
our $counter = 0;
sub new {
my $class = shift;
$counter++;
return bless {}, $class;
}
sub howManyInstances {
return $counter;
}
1;
On your second question, I'm not sure what closures have to do with inheritance.
Here's a simple subclass:
package MySubClass;
use warnings;
use strict;
use parent 'MyBaseClass'; # use parent schema, don't mess with #ISA
sub new {
my $class = shift;
my $self = $class->SUPER::new(#_);
$self->{_name} = undef;
return $self;
}
# Your setter/getter looks ok as is, though lowercase is tradional for methods/subs
1;
Now, if this were real code you would not do it like this - you would use Moo or Moose.