Similar to this question about iterating over subroutine references, and as a result of answering this question about a OO dispatch table, I was wondering how to call a method reference inside a reference, without removing it first, or if it was even possible.
For example:
package Class::Foo;
use 5.012; #Yay autostrict!
use warnings;
# a basic constructor for illustration purposes....
sub new {
my $class = shift;
return bless {#_}, $class;
}
# some subroutines for flavor...
sub sub1 { say 'in sub 1'; return shift->{a} }
sub sub2 { say 'in sub 2'; return shift->{b} }
sub sub3 { say 'in sub 3'; return shift->{c} }
# and a way to dynamically load the tests we're running...
sub sublist {
my $self = shift;
return [
$self->can('sub1'),
$self->can('sub3'),
$self->can('sub2'),
];
}
package main;
sub get_index { ... } # details of how we get the index not important
my $instance = Class::Foo->new(a => 1, b => 2, c => 3);
my $subs = $instance->sublist();
my $index = get_index();
# <-- HERE
So, at HERE, we could do:
my $ref = $subs->[$index];
$instance->$ref();
but how would we do this, without removing the reference first?
Edit:
Changed code example so people don't get hung up on implementation details (sigh, tried my best). The important difference between this and the first link I gave was that the function should be invoked as a method, not as a straight subroutine.
Edit 2:
See the discussion in the linked comment about the technical details, and why the longer way (storing the subref to a variable, then calling it) is probably preferable.
As written, you can get away with
$tests->[$index]();
because the methods in your question aren't using $self.
You could pass $instance explicitly, but that's clunky. Better would be to simulate delegates with closures:
sub sublist {
my $self = shift;
my $sublist;
for (qw/ sub1 sub3 sub2 /) {
my $meth = $_;
push #$sublist => sub { $self->$meth() };
}
return $sublist;
}
If you prefer to be concise, use
sub sublist {
my $self = shift;
return [ map { my $meth = $_; sub { $self->$meth() } }
qw/ sub1 sub3 sub2 / ];
}
Calling one at random is still
$tests->[$index]();
but now the methods get invocants.
Update
Grabbing subrefs via can appears to be unnecessary complexity. If a runtime-determined list of names of methods to call will do, then you can simplify your code greatly:
sub sublist {
my $self = shift;
return [ qw/ sub1 sub3 sub2 / ];
}
Below, we call them all for testing purposes, but you can also see how to call only one:
foreach my $method (#$subs) {
my $x = $instance->$method();
say "$method returned $x";
}
Output:
in sub 1
sub1 returned 1
in sub 3
sub3 returned 3
in sub 2
sub2 returned 2
(Temporary placeholder here until/unless the original poster of the answer returns):
The trick is adding a dereference:
$instance->${\$sublist->[$index]}(#args);
thus you can also do:
$instance->${\$instance->sublist->[$index]}(#args);
otherwise it thinks it's a scalar to dereference. (eg, Not a SCALAR reference at script.pl, line XX).
Related
Is there any way to achieve partial application in Perl?
Suppose, I want to do something like:
sub each_file($arr, $op) {
$op->($_) for #{$arr};
...
}
sub each_line($op, $file) {
...
}
each_file($arr, each_line($op));
I want to partially apply each_line() to only $op, so it'll become a new function can be passed to $each_file, how do I express this in idiomatic Perl?
You can do this in Perl with two approaches combined:
A function which returns a function reference
Closures
Example:
sub each_file {
my ($arr, $line_fn) = #_;
$line_fn->($_) for #{$arr};
...
}
sub each_line {
my ($op, $file) = #_;
...
}
sub make_line_processor {
my ( $op ) = #_;
# This is closed over $op, which effectively becomes
# a constant for the returned function
my $fn = sub {
return each_line( $op, #_ );
};
return $fn;
}
# To call it:
each_file( $arr, make_line_processor($op) );
This can be an even more useful technique in cases where you don't want $op directly, but some expensive-to-fetch derivation of it. In which case you would calculate the derived value just once (in the make_line_processor function) and close over that instead.
# given some $op as implied by your code snippet
each_file($arr, sub { each_line($op, shift) });
# shift op will be applied when anonymous sub { … } is called
(Your code snippet doesn't make it entirely clear what you intend $op to be when you make the call to each_line. It's usually better to present small working programs.)
You can roll this functionality up into a class. Then you can overload the subroutine dereference operator to make it look like your class is really a code reference.
package Partial;
use overload '&{}' => \&call;
sub new {
my $class = shift;
my $code = shift;
bless {code => $code, args => \#_}, $class;
}
sub call {
my ($self) = #_;
return sub{ $self->{code}->(#{$self->{args}}, #_) }
}
You can then use it like this:
sub printArgs {
print join ", ", #_;
print "\n";
}
my $partial = Partial->new(\&printArgs, 'foo', 'bar');
$partial->('baz', 'bat');
# prints foo, bar, baz, bat
I have a dispatch table that I wish to initialize only once, and is only intended to be used by one function. I was hoping to move the dispatch table outside of the subroutine and into the same anonymous block, but since the dispatch table uses closures to call methods for the object passed into the function, moving the table outside the function separates it from access to the object. What other options do I have for this dispatch table?
I'm using Perl 5.8, so unfortunately I'm unable to use state variables.
sub foo {
my ($self, $var) = #_;
my %funcs = (
a => sub { $self->_a() },
b => sub { $self->_b() },
...
);
return $funcs{$var}->();
}
Your functions in the dispatch table are closures over $self. If you pass in the $self as a parameter, you can get around that. Note that state variables are not true closures over $self, and would require an explicit parameter as well.
my %funcs = (
a => sub { shift->_a }, # these are like anonymous methods
b => sub { shift->_b },
);
sub foo {
my ($self, $var) = #_;
my $meth = $funcs{$var} || die "There is no entry $var";
return $self->$meth(); # sugary syntax
}
Here is a demonstration why state would be a bad idea:
use 5.010;
package Foo;
sub new { my ($c, $v) = #_; bless \$v, $c }
sub foo {
my ($self) = #_;
state $cb = sub { say $$self };
$cb->();
}
Foo->new($_)->foo for 1..3;
Output:
1
1
1
The inner sub is a closure, but the initialization of $cb is only performed once. Thus the closed over $self is the first one.
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.
How do you either:
have a sub return a sub
or
execute text as code
in Perl?
Also, how do I have an anonymous function store state?
A sub returns a sub as a coderef:
# example 1: return a sub that is defined inline.
sub foo
{
return sub {
my $this = shift;
my #other_params = #_;
do_stuff();
return $some_value;
};
}
# example 2: return a sub that is defined elsewhere.
sub bar
{
return \&foo;
}
Arbitrary text can be executed with the eval function: see the documentation at perldoc -f eval:
eval q{print "hello world!\n"};
Note that this is very dangerous if you are evaluating anything extracted from user input, and is generally a poor practice anyway as you can generally define your code in a coderef as in the earlier examples above.
You can store state with a state variable (new in perl5.10), or with a variable scoped higher than the sub itself, as a closure:
use feature 'state';
sub baz
{
state $x;
return ++$x;
}
# create a new scope so that $y is not visible to other functions in this package
{
my $y;
sub quux
{
return ++$y;
}
}
Return a subroutine reference.
Here's a simple example that creates sub refs closed over a value:
my $add_5_to = add_x_to(5);
print $add_5_to->(7), "\n";
sub add_x_to {
my $x = shift;
return sub { my $value = shift; return $x + $value; };
}
You can also work with named subs like this:
sub op {
my $name = shift;
return $op eq 'add' ? \&add : sub {};
}
sub add {
my $l = shift;
my $r = shift;
return $l + $r;
}
You can use eval with an arbitrary string, but don't do it. The code is hard to read and it restarts compilation, which slows everything down. There are a small number of cases where string eval is the best tool for the job. Any time string eval seems like a good idea, you are almost certainly better off with another approach.
Almost anything you would like to do with string eval can be achieved with closures.
Returning subs is easy by using the sub keyword. The returned sub closes over the lexical variables it uses:
#!/usr/bin/perl
use strict; use warnings;
sub mk_count_from_to {
my ($from, $to) = #_;
return sub {
return if $from > $to;
return $from ++;
};
}
my $c = mk_count_from_to(-5, 5);
while ( defined( my $n = $c->() ) ) {
print "$n\n";
}
5.10 introduced state variables.
Executing text as Perl is accomplished using eval EXPR:
the return value of EXPR is parsed and executed as if it were a little Perl program. The value of the expression (which is itself determined within scalar context) is first parsed, and if there weren't any errors, executed in the lexical context of the current Perl program, so that any variable settings or subroutine and format definitions remain afterwards. Note that the value is parsed every time the eval executes
Executing arbitrary strings will open up huge gaping security holes.
You can create anonymous subroutines and access them via a reference; this reference can of course be assigned to a scalar:
my $subref = sub { ... code ... }
or returned from another subroutine
return sub { ... code ... }
If you need to store states, you can create closures with lexical variables defined in an outer scope like:
sub create_func {
my $state;
return sub { ... code that can refer to $state ... }
}
You can run code with eval
I often find it useful to be able to schedule code to be executed upon leaving the current scope. In my previous life in TCL, a friend created a function we called defer.
It enabled code like:
set fp [open "x"]
defer("close $fp");
which was invoked when the current scope exited. The main benefit is that it's always invoked no matter how/where I leave scope.
So I implemented something similar in Perl but it seems there'd be an easier way. Comments critiques welcome.
The way I did it in Perl:
create a global, tied variable which holds an array of subs to be executed.
whenever I want to schedule a fn to be invoked on exit, I use local to change the array.
when I leave the current scope, Perl changes the global to the previous value
because the global is tied, I know when this value change happens and can invoke the subs in the list.
The actual code is below.
Is there a better way to do this? Seems this would be a commonly needed capability.
use strict;
package tiescalar;
sub TIESCALAR {
my $class = shift;
my $self = {};
bless $self, $class;
return $self;
}
sub FETCH {
my $self = shift;
return $self->{VAL};
}
sub STORE {
my $self = shift;
my $value = shift;
if (defined($self->{VAL}) && defined($value)) {
foreach my $s (#{$self->{VAL}}) { &$s; }
}
$self->{VAL} = $value;
}
1;
package main;
our $h;
tie($h, 'tiescalar');
$h = [];
printf "1\n";
printf "2\n";
sub main {
printf "3\n";
local $h = [sub{printf "9\n"}];
push(#$h, sub {printf "10\n";});
printf "4\n";
{
local $h = [sub {printf "8\n"; }];
mysub();
printf "7\n";
return;
}
}
sub mysub {
local $h = [sub {printf "6\n"; }];
print "5\n";
}
main();
printf "11\n";
Well, your specific case is already handled if you use lexical filehandles (as opposed to the old style bareword filehandles). For other cases, you could always use the DESTROY method of an object guaranteed to go to zero references when it goes out of scope:
#!/usr/bin/perl
use strict;
use warnings;
for my $i (1 .. 5) {
my $defer = Defer::Sub->new(sub { print "end\n" });
print "start\n$i\n";
}
package Defer::Sub;
use Carp;
sub new {
my $class = shift;
croak "$class requires a function to call\n" unless #_;
my $self = {
func => shift,
};
return bless $self, $class;
}
sub DESTROY {
my $self = shift;
$self->{func}();
}
ETA: I like brian's name better, Scope::OnExit is a much more descriptive name.
Instead of using tie for this, I think I'd just create an object. You can also avoid the local that way too.
{
my $defer = Scope::OnExit->new( #subs );
$defer->push( $other_sub ); # and pop, shift, etc
...
}
When the variable goes out of scope, you have a chance to do things in the DESTROY method.
Also, in the example you posted, you need to check that the values you store are code references, and it's probably a good idea to check that the VAL value is an array reference:
sub TIESCALAR { bless { VAL => [] }, $_[0] }
sub STORE {
my( $self, $value ) = #_;
carp "Can only store array references!" unless ref $value eq ref [];
foreach { #$value } {
carp "There should only be code refs in the array"
unless ref $_ eq ref sub {}
}
foreach ( #{ $self->{VAL}} ) { $_->() }
$self->{VAL} = $value;
}
You may want to try out B::Hooks::EndOfScope
I Believe this works:
use B::Hooks::EndOfScope;
sub foo {
on_scope_end {
$codehere;
};
$morecode
return 1; # scope end code executes.
}
foo();
I think you want something like Scope::Guard, but it can't be pushed. Hmmm.
Thanks.
Trivially,
sub OnLeavingScope::DESTROY { ${$_[0]}->() }
used like:
{
...
my $onleavingscope = bless \sub { ... }, 'OnLeavingScope';
my $onleavingscope2 = bless \\&whatever, 'OnLeavingScope';
...
}
(The extra level of having a reference to a reference to a sub is necessary only to work around an optimization (that's arguably a bug) when using a non-closure anonymous sub.)