I am trying to use MooseX::Method::Signatures and MooseX::Declare in an application, my need is to inject custom code at the beginning of each method at compile time not at run time:
instead of this:
use MooseX::Declare;
method check ($value) {
$return $value;
}
I want to inject a code at the beginning of each method at compile time to be like that:
method check ($value) {
my ($value) = $self->validate($value);
$return $value;
}
now I want the code
my ($value) = $self->validate($value);
to be injected automatically at the beginning of all methods in the package using the MooseX::Decalre module at compile time and not at run time, I mean not using the Moose method modifiers before, after, around etc.
This needs a modification of these module but I need someone to tell me where to start.
I was able to modify the module Method::Signatures::Simple to do this exactly and emailed the author for the modification but did not get a reply. The reason I can not use this even with modification because it does not support type checking and defaults like MooseX::Declare.
The modified version of the module Method::Signatures::Simple below for reference and I use it as follows:
use Method::Signatures::Simple (method => 'method,action', function =>
'function', invocant=>'$this', 'inject'=>'my ($me) = $this->me;');
now in all methods, I get the code my ($me) = $this->me; injected and I just can use it like that:
method check ($value) {
say $me
}
Here is the modified Method::Signatures::Simple module.
package Method::Signatures::Simple;
{
$Method::Signatures::Simple::VERSION = '1.07';
}
use warnings;
use strict;
=head1 NAME
Method::Signatures::Simple - Basic method declarations with signatures, without source filters
=head1 VERSION
version 1.07
=cut
use base 'Devel::Declare::MethodInstaller::Simple';
our $inject_code;
sub import {
my $class = shift;
my %opts = #_;
$opts{into} ||= caller;
my $meth = delete $opts{name} || delete $opts{method};
my $func = delete $opts{function};
my $invocant = delete $opts{invocant} || '$self';
$inject_code = delete $opts{inject};
$inject_code .= ";" if ($inject_code && $inject_code !~ /\;$/);
# if no options are provided at all, then we supply defaults
unless (defined $meth || defined $func) {
$meth = 'method';
$func = 'func';
}
my #meth = split /\s*\,+\s*/, $meth;
# we only install keywords that are requested
foreach $meth (#meth) {
if (defined $meth) {
$class->install_methodhandler(
name => $meth,
invocant => $invocant,
%opts,
);
}
}
if (defined $func) {
$class->install_methodhandler(
name => $func,
%opts,
invocant => undef,
);
}
}
sub strip_proto {
my $self = shift;
my ($proto) = $self->SUPER::strip_proto()
or return '';
# we strip comments and newlines here, and stash the number of newlines.
# we will re-inject the newlines in strip_attrs(), because DD does not
# like it when you inject them into the following code block. it does not
# object to tacking on newlines to the code attribute spec though.
# (see the call to inject_if_block() in DD::MethodInstaller::Simple->parser)
$proto =~ s/\s*#.*$//mg;
$self->{__nls} = $proto =~ s/[\r\n]//g;
$proto;
}
sub strip_attrs {
my $self = shift;
my ($attrs) = $self->SUPER::strip_attrs();
$attrs ||= '';
$attrs .= $/ x $self->{__nls} if $self->{__nls};
$attrs;
}
sub parse_proto {
my $self = shift;
my ($proto) = #_;
$proto ||= '';
$proto =~ s/\s*#.*$//mg;
$proto =~ s/^\s+//mg;
$proto =~ s/\s+$//mg;
$proto =~ s/[\r\n]//g;
my $invocant = $self->{invocant};
$invocant = $1 if $proto =~ s{(\$\w+)\s*:\s*}{};
my $inject = '';
$inject .= "my ${invocant} = shift;" if $invocant;
$inject .= "my ($proto) = \#_;" if defined $proto and length $proto;
$inject .= "$inject_code" if $inject_code;
$inject .= '();'; # fix for empty method body
return $inject;
}
Moops and Kavorka provide a syntax almost compatible with MooseX::Declare and MooseX::Method::Signatures, and are designed to be very extensible (even from within!) via traits. I'll draw your attention to the following section of documentation for MooseX::Declare:
Warning: MooseX::Declare is based on Devel::Declare, a giant bag of crack originally implemented by mst with the goal of upsetting the perl core developers so much by its very existence that they implemented proper keyword handling in the core.
[...]
If you want to use declarative syntax in new code, please for the love of kittens get yourself a recent perl and look at Moops instead.
MooseX::Declare itself is not very easy to extend. I know. I've tried.
So bearing all that in mind, and also because I wrote Moops, I'll use that for the example. Here we define a role Kavorka::TraitFor::Sub::ProvidesMe which is will inject a little bit of code into a method. We then apply that role to a method using does ProvideMe.
package main;
use Moops;
role Kavorka::TraitFor::Sub::ProvideMe
{
around inject_prelude (#_)
{
my $prelude = $self->$next(#_);
$prelude .= 'my ($me) = $self->me;();';
return $prelude;
}
}
class MyClass
{
method me () { "tobyink" }
method example () does ProvideMe
{
# This gets injected: my ($me) = $self->me;
return $me;
}
}
my $obj = MyClass->new;
say $obj->example; ## says "tobyink"
Related
I'd like to be able to "inject" methods in a class on the fly, similarly to what happens with Mojolicious helpers. Something like this:
my $s = SomeThing->new;
$s->helper(do_this => sub {
my $self = shift;
$foo = shift;
});
$s->do_this('bar');
I've made it some distance, but I would like the subs that get injected to be operating in the namespace of the class they get injected into, not in the main one. In other words this currently works as follows:
$s->do_this('bar');
print 'in main: ', $foo;
this prints "bar" - and I'd like it not to, while I'd like this
print 'in SomeThing: ', $SomeThing::foo;
to print "bar" instead
while this works but seems clunky to me
$s->helper(do_this => sub {
my $self = shift;
${(ref $self) . '::foo'} = shift;
});
$s->do_this('foo');
print 'in SomeThing: ', $SomeThing::foo; # now this prints "foo"
The package where all this happens looks like this:
package SomeThing {
use Mojo::Base -base;
use Carp;
sub helper {
my $self = shift;
my $name = shift || croak "The helper name is required";
my $sub = shift || sub {};
my $namespace = __PACKAGE__;
no strict 'refs';
{
*{"$namespace\::$name"} = $sub
}
}
};
Is there a way to do this? I suspect I'd be messing up strictness real bad - but I kind of don't want to give up just yet (and it'd be a nice trick to learn).
You are asking to change the package associated with an already-compiled anon sub for the purpose of variable lookups. I don't know if that's possible.
Even if it was possible, it's not something you want to do because your code still wouldn't work. You'd have to add use vars qw( foo ); to the file in which the sub { } literal is found. And that's in addition to using our $foo; or use vars qw( $foo ); in Something.pm if you accessed it there.
That's pretty magical and messy. And it's easily avoided by using accessors. Simple replace
$s->helper(
do_this => sub {
my $self = shift;
$foo = shift;
},
);
with
$s->helper(
do_this => sub {
my $self = shift;
$self->foo(shift);
},
);
If you also need to add the accessor, you can use the following:
$s->helper(
foo => sub {
shift;
state $foo;
$foo = shift if #_;
$foo
},
do_this => sub {
my $self = shift;
$self->foo(shift);
},
);
As an aside, monkey_patch from Mojo::Util can be used as a replacement for helper. (Credit to #brian d foy for bringing it up.) It does the same thing, but it has the two added benefits:
You don't need to support it.
It sets the name of the anon sub so that stack traces use a meaningful name instead of __ANON__.
Switching to monkey_patch doesn't address your problem, but I do recommend using it (or similar) in addition to the change of approach I mentioned above.
use Mojo::Util qw( );
sub helper { shift; Mojo::Util::monkey_patch(__PACKAGE__, #_); }
Consider roles.
# role module
package SomeThing::Role::Foo;
use Role::Tiny;
sub foo { 42 }
1;
# user
use strict;
use warnings;
use SomeThing;
use With::Roles;
my $something_with_foo = SomeThing->with::roles('+Foo');
# new subclass of SomeThing, doesn't affect other usage of SomeThing
my $obj = $something_with_foo->new;
# can also dynamically apply to an existing object
my $obj = SomeThing->new->with::roles('+Foo');
print $obj->foo;
How can I hide a "tie" call from the user so calling an accessor will implicitly do it for them?
I want to do this, because I have a data structure that can be accessed by the user, but values stored in this structure can be modified without the user's knowledge.
If an attribute in the data structure changes, I want any variables referencing that attribute modified as well so the user will always be using fresh data. Since the user will always want fresh data, it's simpler and more intuitive if the user doesn't even need to know it's happening.
This is what I have so far... it doesn't seem to work though, the output is:
hello
hello
What I want is:
hello
goodbye
Code:
#!/usr/bin/perl
use warnings;
use strict;
use feature qw{ say };
{
package File;
use Moose;
has '_text' => (is => 'rw', isa => 'Str', required => 1);
sub text {
my ($self) = #_;
tie my $text, 'FileText', $self;
return $text;
}
}
{
package FileText;
use Tie::Scalar;
sub TIESCALAR {
my ($class, $obj) = #_;
return bless \$obj, $class;
}
sub FETCH {
my ($self) = #_;
return $$self->_text();
}
sub STORE {
die "READ ONLY";
}
}
my $file = 'File'->new('_text' => 'hello');
my $text = $file->text();
say $text;
$file->_text('goodbye');
say $text;
I would not recommend doing this. You're introducing "action at a distance" which leads to some very difficult to catch bugs. The user thinks they're getting a string. A lexical string can only be altered by changing it directly and obviously. It has to be altered in place or obviously passed into a function or a reference attached to something.
my $text = $file->text;
say $text; # let's say it's 'foo'
...do some stuff...
$file->text('bar');
...do some more stuff...
# I should be able to safely assume it will still be 'foo'
say $text;
That block of code is easy to understand because all the things which could affect $text are immediately visible. This is what lexical context is all about, isolating what can change a variable.
By returning a thing which can change at any time, you've quietly broken this assumption. There's no indication to the user that assumption has been broken. When they go to print $text and get bar it is non-obvious what changed $text. Anything in the whole program could change $text. That small block of code is now infinitely more complicated.
Another way to look at it is this: scalar variables in Perl have a defined interface. Part of that interface says how they can be changed. You are breaking this interface and lying to the user. This is how overloaded/tied variables are typically abused.
Whatever problem you're trying to solve, you're solving it by adding more problems, by making the code more complex and difficult to understand. I would step back and ask what problem you're trying to solve with tying.
What I would do instead is to just return a scalar reference. This alerts the user that it can be changed out from under them at any time. No magic to cover up a very important piece of information.
#!/usr/bin/perl
use warnings;
use strict;
use feature qw{ say };
{
package File;
use Moose;
has 'text_ref' => (
is => 'rw',
isa => 'Ref',
default => sub {
return \("");
}
);
sub BUILDARGS {
my $class = shift;
my %args = #_;
# "Cast" a scalar to a scalar ref.
if( defined $args{text} ) {
$args{text_ref} = \(delete $args{text});
}
return \%args;
}
sub text {
my $self = shift;
if( #_ ) {
# Change the existing text object.
${$self->text_ref} = shift;
return;
}
else {
return $self->text_ref;
}
}
}
my $file = 'File'->new('text' => 'hello');
my $text = $file->text();
say $$text;
$file->text('goodbye');
say $$text;
That said, here's how you do what you want.
I would recommend against using tie. It is very slow, considerably slower than a method call, buggy and quirky. One of its quirks is that the tied nature is attached to the variable itself, not the referenced data. That means you can't return a tied variable.
Instead, I would recommend using an overloaded object to store your changing text.
{
package ChangingText;
# Moose wants class types to be in a .pm file. We have to explciitly
# tell it this is a class type.
use Moose::Util::TypeConstraints qw(class_type);
class_type('ChangingText');
use overload
'""' => sub {
my $self = shift;
return $$self;
},
fallback => 1;
sub new {
my $class = shift;
my $text = shift;
return bless \$text, $class;
}
sub set_text {
my $self = shift;
my $new_text = shift;
$$self = $new_text;
return;
}
}
Overloaded objects have their own caveats, mostly due to code which expects strings writing things like if !ref $arg, but they are easier to deal with than the deep tie bugs.
To make this transparent, store the ChangingText object in the File object and then put a hand made text accessor around it to handle plain strings. The accessor makes sure to reuse the same ChangingText object.
To complete the illusion, BUILDARGS is used to change plain text initialization arguments into a ChangingText object.
{
package File;
use Moose;
has 'text_obj' => (
is => 'rw',
isa => 'ChangingText',
default => sub {
return ChangingText->new;
}
);
sub BUILDARGS {
my $class = shift;
my %args = #_;
# "Cast" plain text into a text object
if( defined $args{text} ) {
$args{text_obj} = ChangingText->new(delete $args{text});
}
return \%args;
}
sub text {
my $self = shift;
if( #_ ) {
# Change the existing text object.
$self->text_obj->set_text(shift);
return;
}
else {
return $self->text_obj;
}
}
}
Then it works transparently.
my $file = File->new('text' => 'hello');
my $text = $file->text();
say $text; # hello
$file->text('goodbye');
say $text; # goodbye
return $text just returns the value of the variable, not the variable itself. You can return a reference to it, though:
sub text {
my ($self) = #_;
tie my $text, 'FileText', $self;
return \$text;
}
You then have to use $$text to dereference it:
my $file = 'File'->new('_text' => 'hello');
my $text = $file->text();
say $$text;
$file->_text('goodbye');
say $$text;
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.
Using sub prototypes, we can define our own subs that look like map or grep. That is, the first coderef argument has shorter syntax than a normal anonymous sub. For example:
sub thunked (&) { $_[0] }
my $val = thunked { 2 * 4 };
Works great here, since the first argument is the coderef. For latter arguments however, it simple won't parse properly.
I made a with sub designed to make writing GTK2 code cleaner. It's meant to look like this (untested since it's hypothetical code):
use 5.012;
use warnings;
use Gtk2 '-init';
sub with ($&) {
local $_ = $_[0];
$_[1]->();
$_;
}
for (Gtk2::Window->new('toplevel')) {
$_->set_title('Test Application');
$_->add(with Gtk2::VBox->new {
my $box = $_;
$box->add(Gtk2::Button->new("Button $_")) for (1..4);
});
$_->show_all;
}
Gtk2->main;
It doesn't work because with needs to take the block as a first argument for the nice syntax to work. Is there any way to pull it off?
The module Devel::Declare contains tools for extending Perl's syntax in a relatively safe way.
Using Devel::Declare you would create a hook on the with token, which will stop the parser when it reaches that word. From there, you have control over the parser and you can read ahead until you reach a { symbol. At that point, you have what you need to work with, so you rewrite it into valid Perl, and pass it back to the parser.
in the file With.pm:
package With;
use warnings;
use strict;
use Devel::Declare;
sub import {
my $caller = caller;
Devel::Declare->setup_for (
$caller => {with => {const => \&parser}}
);
no strict 'refs';
*{$caller.'::with'} = sub ($&) {
$_[1]() for $_[0];
$_[0]
}
}
our $prefix = '';
sub get {substr Devel::Declare::get_linestr, length $prefix}
sub set { Devel::Declare::set_linestr $prefix . $_[0]}
sub parser {
local $prefix = substr get, 0, length($_[0]) + $_[1];
my $with = strip_with();
strip_space();
set "scalar($with), sub " . get;
}
sub strip_space {
my $skip = Devel::Declare::toke_skipspace length $prefix;
set substr get, $skip;
}
sub strip_with {
strip_space;
my $with;
until (get =~ /^\{/) {
(my $line = get) =~ s/^([^{]+)//;
$with .= $1;
set $line;
strip_space;
}
$with =~ s/\s+/ /g;
$with
}
and to use it:
use With;
sub Window::add {say "window add: ", $_[1]->str}
sub Window::new {bless [] => 'Window'}
sub Box::new {bless [] => 'Box'}
sub Box::add {push #{$_[0]}, #_[1..$#_]}
sub Box::str {"Box(#{$_[0]})"}
sub Button::new {"Button($_[1])"}
with Window->new {
$_->add(with Box->new {
for my $num (1 .. 4) {
$_->add(Button->new($num))
}
})
};
Which prints:
window add: Box(Button(1) Button(2) Button(3) Button(4))
A completely different approach would be to skip the with keyword altogether and write a routine to generate constructor subroutines:
BEGIN {
for my $name (qw(VBox)) { # and any others you want
no strict 'refs';
*$name = sub (&#) {
use strict;
my $code = shift;
my $with = "Gtk2::$name"->new(#_);
$code->() for $with;
$with
}
}
}
and then your code could look like
for (Gtk2::Window->new('toplevel')) {
$_->set_title('Test Application');
$_->add(VBox {
my $box = $_;
$box->add(Gtk2::Button->new("Button $_")) for (1..4);
});
$_->show_all;
}
One way that you could deal with it is to add a fairly useless keyword:
sub perform(&) { $_[0] }
with GTK2::VBox->new, perform { ... }
where perform is really just a sugarier alternative to sub.
Another way is to write a Devel::Declare filter or a Syntax::Keyword:: plugin to implement your with, as long as you have some way to tell when you're done parsing the with argument and ready to start parsing the block — balanced parentheses would do (so would an opening curly brace, but then hashes become a problem). Then you could support something like
with (GTK2::VBox->new) { ... }
and let the filter rewrite it to something like
do {
local $_ = GTK2::VBox->new;
do {
...;
};
$_;
}
which, if it works, has the advantage of not actually creating a sub, and thus not interfering with #_, return, and a few other things. The two layers of do-age I think are necessary for being able to install an EndOfScope hook in the proper place.
The obvious disadvantages of this are that it's tricky, it's hairy, and it's a source filter (even if it's a tame one) which means there are problems you have to solve if you want any code using it to be debuggable at all.
I am writing a module and I want a specific piece of code to be executed before each of the functions in it.
How do I do that?
Is there no other way than to just have a function-call at the beginning of every function?
You can do this in Moose with method modifiers:
package Example;
use Moose;
sub foo {
print "foo\n";
}
before 'foo' => sub { print "about to call foo\n"; };
Wrapping a method is also possible with method attributes, but this route is not well-used in Perl and is still evolving, so I wouldn't recommend it. For normal use-cases, I would simply put the common code in another method and call it at the top of each of your functions:
Package MyApp::Foo;
sub do_common_stuff { ... }
sub method_one
{
my ($this, #args) = #_;
$this->do_common_stuff();
# ...
}
sub method_two
{
my ($this, #args) = #_;
$this->do_common_stuff();
# ...
}
And, in case someone is wondering how to achieve the effect of Hook* modules or Moose's "before" explicitly (e.g. what actual Perl mechanism can be used to do it), here's an example:
use strict;
package foo;
sub call_before { print "BEFORE\n"; } # This will be called before any sub
my $call_after = sub { print "AFTER - $_[0]\n"; };
sub fooBar { print "fooBar body\n\n"; }
sub fooBaz { print "fooBaz body\n\n"; }
no strict; # Wonder if we can get away without 'no strict'? Hate doing that!
foreach my $glob (keys %foo::) { # Iterate over symbol table of the package
next if not defined *{$foo::{$glob}}{CODE}; # Only subroutines needed
next if $glob eq "call_before" || $glob eq "import" || $glob =~ /^___OLD_/;
*{"foo::___OLD_$glob"} = \&{"foo::$glob"}; # Save original sub reference
*{"foo::$glob"} = sub {
call_before(#_); &{"foo::___OLD_$glob"}(#_); &$call_after(#_);
};
}
use strict;
1;
package main;
foo::fooBar();
foo::fooBaz();
The explanation for what we're excluding via "next" line:
"call_before" is of course the name I gave to our "before" example sub - only need this if it is actually defined as a real sub in the same package and not anonymously or code ref from outside the package.
import() has a special meaning and purpose and should generally be excluded from "run this before every sub" scenario. YMMV.
___OLD_ is a prefix we will give to "renamed" old subs - you don't need to include it here unless you're worried about this loop being execute twice. Better safe than sorry.
UPDATE: Below section about generalization is no longer relevant - at the end of the answer I pasted a general "before_after" package doing just that!!!
The loop above can obviously be easily generalized to be a separately-packaged subroutine which accepts, as arguments:
an arbitrary package
a code ref to arbitrary "before" subroutine (or as you can see, after)
and a list of sub names to exclude (or sub ref that checks if a name is to be excluded) aside from standard ones like "import").
... and/or a list of sub names to include (or sub ref that checks if a name is to be included) aside from standard ones like "import"). Mine just takes ALL subs in a package.
NOTE: I don't know whether Moose's "before" does it just this way. What I do know is that I'd obviously recommend going with a standard CPAN module than my own just-written snippet, unless:
Moose or any of the Hook modules can't be installed and/or are too heavy weight for you
You're good enough with Perl that you can read the code above and analyze it for flaws.
You like this code very much, AND the risk of using it over CPAN stuff is low IYHO :)
I supplied it more for informational "this is how the underlying work is done" purposes rather than practical "use this in your codebase" purposes, though feel free to use it if you wish :)
UPDATE
Here's a more generic version as mentioned before:
#######################################################################
package before_after;
# Generic inserter of before/after wrapper code to all subs in any package.
# See below package "foo" for example of how to use.
my $default_prefix = "___OLD_";
my %used_prefixes = (); # To prevent multiple calls from stepping on each other
sub insert_before_after {
my ($package, $prefix, $before_code, $after_code
, $before_filter, $after_filter) = #_;
# filters are subs taking 2 args - subroutine name and package name.
# How the heck do I get the caller package without import() for a defalut?
$prefix ||= $default_prefix; # Also, default $before/after to sub {} ?
while ($used_prefixes{$prefix}) { $prefix = "_$prefix"; }; # Uniqueness
no strict;
foreach my $glob (keys %{$package . "::"}) {
next if not defined *{$package. "::$glob"}{CODE};
next if $glob =~ /import|__ANON__|BEGIN/; # Any otrher standard subs?
next if $glob =~ /^$prefix/; # Already done.
$before = (ref($before_filter) ne "CODE"
|| &$before_filter($glob, $package));
$after = (ref($after_filter) ne "CODE"
|| &$after_filter($glob, $package));
*{$package."::$prefix$glob"} = \&{$package . "::$glob"};
if ($before && $after) { # We do these ifs for performance gain only.
# Else, could wrap before/after calls in "if"
*{$package."::$glob"} = sub {
my $retval;
&$before_code(#_); # We don't save returns from before/after.
if (wantarray) {
$retval = [ &{$package . "::$prefix$glob"}(#_) ];
} else {
$retval = &{$package . "::$prefix$glob"}(#_);
}
&$after_code(#_);
return (wantarray && ref $retval eq 'ARRAY')
? #$retval : $retval;
};
} elsif ($before && !$after) {
*{$package . "::$glob"} = sub {
&$before_code(#_);
&{$package . "::$prefix$glob"}(#_);
};
} elsif (!$before && $after) {
*{$package . "::$glob"} = sub {
my $retval;
if (wantarray) {
$retval = [ &{$package . "::$prefix$glob"}(#_) ];
} else {
$retval = &{$package . "::$prefix$glob"}(#_);
}
&$after_code(#_);
return (wantarray && ref $retval eq 'ARRAY')
? #$retval : $retval;
};
}
}
use strict;
}
# May be add import() that calls insert_before_after()?
# The caller will just need "use before_after qq(args)".
1;
#######################################################################
package foo;
use strict;
sub call_before { print "BEFORE - $_[0]\n"; };
my $call_after = sub { print "AFTER - $_[0]\n"; };
sub fooBar { print "fooBar body - $_[0]\n\n"; };
sub fooBaz { print "fooBaz body - $_[0]\n\n"; };
sub fooBazNoB { print "fooBazNoB body - $_[0]\n\n"; };
sub fooBazNoA { print "fooBazNoA body - $_[0]\n\n"; };
sub fooBazNoBNoA { print "fooBazNoBNoA body - $_[0]\n\n"; };
before_after::insert_before_after(__PACKAGE__, undef
, \&call_before, $call_after
, sub { return 0 if $_[0] eq "call_before"; $_[0] !~ /NoB(NoA)?$/ }
, sub { return 0 if $_[0] eq "call_before"; $_[0] !~ /NoA$/ } );
1;
#######################################################################
package main;
use strict;
foo::fooBar("ARG1");
foo::fooBaz("ARG2");
foo::fooBazNoB("ARG3");
foo::fooBazNoA("ARG4");
foo::fooBazNoBNoA("ARG5");
#######################################################################
If you search CPAN for 'hook', and then branch out from there, you'll find several options, such as:
Hook::WrapSub
Hook::PrePostCall
Hook::LexWrap
Sub::Prepend
Here's an example using Hook::LexWrap. I don't have experience with this module except for debugging. It worked fine for that purpose.
# In Frob.pm
package Frob;
sub new { bless {}, shift }
sub foo { print "foo()\n" }
sub bar { print "bar()\n" }
sub pre { print "pre()\n" }
use Hook::LexWrap qw(wrap);
my #wrappable_methods = qw(foo bar);
sub wrap_em {
wrap($_, pre => \&pre) for #wrappable_methods;
}
# In script.pl
use Frob;
my $frob = Frob->new;
print "\nOrig:\n";
$frob->foo;
$frob->bar;
print "\nWrapped:\n";
Frob->wrap_em();
$frob->foo;
$frob->bar;
See the Aspect package on CPAN for aspect-oriented computing.
before { Class->method; } qr/^Package::\w+$/;