File Coverage

blib/lib/Aspect/Pointcut.pm

Criterion	Covered	Total	%
statement	78	88	88.6
branch	22	26	84.6
condition	17	30	56.6
subroutine	19	24	79.1
pod	8	10	80.0
total	144	178	80.9

package Aspect::Pointcut;

=pod

=head1 NAME

Aspect::Pointcut - API for determining which events should be hooked

=head1 DESCRIPTION

Aspect-Oriented Programming implementations draw much of their power from the
flexibility that can be applied to when a function call should or should not
be hooked.

B provides a robust and powerful API for defining the rules
for when a function call should be hooked, and then applying the rules as
optimally as possible. This optimisation is particularly important for any
pure-Perl implementation, which cannot hook deeply into the underlying
virtual machine as you might with a Java or Perl XS-based implementation.

A running program can be seen as a collection of events. Events like a
sub returning from a call, or a package being used. These are called join
points. A pointcut defines a set of join points, taken from all the join
points in the program. Different pointcut classes allow you to define the
set in different ways, so you can target the exact join points you need.

Pointcuts are constructed as trees; logical operations on pointcuts with
one or two arguments (not, and, or) are themselves pointcut operators.
You can construct them explicitly using object syntax, or you can use the
convenience functions exported by Aspect and the overloaded operators
C, C<&> and C<|>.

=head1 METHODS

=cut

use strict;
use Aspect::Pointcut::Or  ();
use Aspect::Pointcut::And ();
use Aspect::Pointcut::Not ();

our $VERSION = '1.04';

use overload (
	# Keep traditional Perl boolification and stringification
	'bool' => sub () { 1 },
	'""'   => sub { ref $_[0] },

	# Overload bitwise boolean operators to perform logical transformations.
	'|'    => sub { Aspect::Pointcut::Or->new(  $_[0], $_[1] ) },
	'&'    => sub { Aspect::Pointcut::And->new( $_[0], $_[1] ) },
	'!'    => sub { Aspect::Pointcut::Not->new( $_[0]        ) },

	# Everything else should fail to match and throw an exception
);





######################################################################
# Constructor

=pod

=head2 new

The C constructor creates new pointcut objects.

All pointcut classes define their own rules around the parameters that are
provided, but once created these pointcuts can then all be mixed together in
an arbitrary fashion.

Note: Unlike most Perl objects the default and recommended underlying datatype
for pointcut objects is an C reference rather than C references.
This is done because pointcut code can directly impact the speed of function
calls, and so is extremely performance sensitive.

=cut

sub new {
	my $class = shift;
	bless [ @_ ], $class;
}





######################################################################
# Weaving Methods

my %PRUNE;
my %IGNORE;
BEGIN {
	# Classes we should not recurse down into
	%PRUNE = map { $_ => 1 } qw{
		main
		B
		CORE
		DB
		Aspect
	};

	# Classes we should not hook functions in
	%IGNORE = map { $_ => 1 } qw{
		Aspect
		B
		Carp
		Carp::Heavy
		Config
		CORE
		DB
		DynaLoader
		Exporter
		Exporter::Heavy
		IO
		IO::Handle
		Regexp
		Sub::Uplevel
		UNIVERSAL
		attributes
		base
		feature
		fields
		lib
		strict
		warnings
		warnings::register
	};
}

=pod

=head2 match_all

  my @fully_resolved_function_names = $pointcut->match_all;

The C method is the primary compile-time function called on the
pointcut model by the core Aspect library.

It will examine the list of all loaded functions and identify those which
could potentially match, and will need to have hooks installed to intercept
calls to those functions.

These functions will not necesarily all result in Aspect code being run.
Some functions may be called in all cases, but often further run-time
analyis needs to be done before we can be sure the particular function call
respresents a match.

Returns a list of fully-resolved function names
(e.g. "Module::Name::function")

=cut

sub match_all {
	my $self    = shift;
	my @matches = ();

	# Curry the pointcut and compile the weave-time function
	my $curried  = $self->curry_weave;
	my $compiled = $curried ? $self->compiled_weave : sub () { 1 };
	unless ( $compiled ) {
		die "Failed to generate weave filter";
	}

	# Quick initial root package scan to remove the need
	# for special-casing of main:: in the recursive scan.
	no strict 'refs';
	my @search = ();
	my ($key,$value);
	while ( ($key,$value) = each %{*{"::"}} ) {
		next unless defined $value;
		local (*ENTRY) = $value;
		next unless defined *ENTRY{HASH};
		next unless $key =~ /^([^\W\d]\w*)::\z/;

		# Suppress aggressively ignored things
		if ( $IGNORE{$1} and $PRUNE{$1} ) {
			next;
		}

		push @search, $1;
	}

	# Search using a simple package list-recursion
	while ( my $package = shift @search ) {
		no strict 'refs';
		my ($key,$value);
		while ( ($key,$value) = each %{*{"$package\::"}} ) {
			next if $key =~ /[^\w:]/;
			next unless defined $value;
			$_ = "$package\::$key";
			local(*ENTRY) = $value;

			# Is this a matched function?
			if (
				defined *ENTRY{CODE}
				and
				not $IGNORE{$package}
				and
				not $Aspect::EXPORTED{$_}
				and
				$compiled->()
			) {
				push @matches, $_;
			}

			# Is this a package we should recurse into?
			if (
				not $PRUNE{$package}
				and
				s/::\z//
				and
				defined *ENTRY{HASH}
			) {
				push @search, $_;
			}
		}
	}

	return @matches;
}

=pod

=head2 match_define

  my $should_hook = $pointcut->match_define;

At compile time, the only common factor in predicting the future state of
a function call is the name of the function itself.

The C method is called on the pointcut for each
theoretically-matchable function in the entire Perl namespace that part of
an ignored namespace, passing a single parameter of the fully-resolved
function name.

The method will determine if the function B match, and needs to be
hooked for further checking at run-time, potentially calling C
on child objects as well.

Returns true if the function might match the pointcut, or false if the
function can never possibly match the pointcut and should never be checked
at run-time.

=cut

sub match_define {
	my $class = ref $_[0] || $_[0];
	die("Method 'match_define' not implemented in class '$class'");
}

=pod

=head2 compile_weave

The C method generates a custom function that is used to test
if a particular named function should be hooked as a potential join point.

=cut

# Most pointcut conditions always match at weave time, so default to that
sub compile_weave {
	return 1;
}

sub compiled_weave {
	my $self = shift;
	my $code = $self->compile_weave;
	return $code if ref $code;
	return eval "sub () { $code }";
}

=pod

=head2 compile_runtime

The C method generates a custom function that is used to test
if a particular named function should be hooked as a potential join point.

=cut

sub compile_runtime {
	my $class = ref $_[0] || $_[0];
	die "Missing compile_runtime method for $class";
}

sub compiled_runtime {
	my $self = shift;
	my $code = $self->compile_runtime;
	return $code if ref $code;
	return eval "sub () { $code }";
}

=pod

=head2 match_contains

  my $calls = $pointcut->match_contains('Aspect::Pointcut::Call');

The C method provides a convenience for the validation and
optimisation systems. It is used to check for the existance of a particular
condition type anywhere within the pointcut object tree.

Returns the number of instances of a particular pointcut type within the tree.

=cut

sub match_contains {
	my $self = shift;
	return 1 if $self->isa($_[0]);
	return 0;
}

=pod

=head2 match_always

  my $always = $pointcut->match_contains('Aspect::Pointcut::Throwing');

The C method provides a convenience for the validation and
optimisation systems. It is used to check that a particular condition type will
be tested at least once for a matching join point, regardless of which path
the match takes through branching pointcut logic.

Returns true if an expression type is encounter at least once in all branches,
or false if there is any branch path that can be taken in which the condition
won't be encountered.

=cut

sub match_always {
	die "CODE NOT IMPLEMENTED";
}

=pod

=head2 curry_runtime

  my $optimized_pointcut = $raw_pointcut->curry_runtime;

In a production system, pointcut declarations can result in large and
complex B object trees.

Because this tree can contain a large amount of structure that is no longer
relevant at run-time, it can end up making a long series of prohibitively
expensive cascading method or function calls before every single regular
function call.

To reduce this cost down to something more reasonable, pointcuts are run
through a currying process (see L).

A variety of optimisations are used to simplify boolean nesting, to remove
tests that are irrelevant once the compile-time hooks have all been set up,
and to remove other tests that the currying process can determine will
never need to be tested.

The currying process will generate and return a new pointcut tree that is
independent from the original, and that can perform a match test at the
structurally minimum computational cost.

Returns a new optimised B object if any further testing
needs to be done at run-time for the pointcut. Returns null (C in
scalar context or C<()> in list context) if the pointcut can be curried
away to nothing, and no further testing needs to be done at run-time.

=cut

sub curry_runtime {
	my $class = ref $_[0] || $_[0];
	die("Method 'curry_runtime' not implemented in class '$class'");
}

=pod

=head2 curry_weave

The C method is similar to the C method, except
that instead of reducing the pointcut to only elements that are relevant at
run-time, it reduces the pointcut to only elements that are relevant at weave
time.

By remove purely run-time elements, the compile weave test code is made both
faster and more accurate (some complicated situations can occur when there is
a L in the tree).

=cut

sub curry_weave {
	my $class = ref $_[0] || $_[0];
	die("Method 'curry_weave' not implemented in class '$class'");
}

sub match_runtime {
	return 1;
}





######################################################################
# Optional XS Acceleration

BEGIN {
	local $@;
	eval <<'END_PERL';
use Class::XSAccessor::Array 1.08 {
	replace => 1,
	true    => [ 'compile_weave', 'match_runtime' ],
};
END_PERL
}

1;

__END__