Robert Barta > WWW-Agent > WWW::Agent::Plugins

Download:
WWW-Agent-0.03.tar.gz

Dependencies

Annotate this POD

View/Report Bugs
Module Version: 0.01   Source  

NAME ^

WWW::Agent::Plugins - Abstract web browser, Plugin Documentation

SYNOPSIS ^

   # this is not a package to invoke

DESCRIPTION ^

Introduction

Writing plugins for the WWW::Agent simply involves some the following things:

intercepting specific agent events

Here you hooked your plugin into the agent which will invoke you at certain times in the request/response cycle (see WWW::Agent for this list).

maintaining information

You may want to hold some information about your state. This can be initialize by intercepting the init event which all plugins receive at the very beginning.

To avoid a mixup every plugin should use a different namespace in a global data structure (the heap as POE calls it). If the agent realizes that two plugins are trying to register the same namespace, it will raise an exception.

triggering additional, private events

No one can stop you to set up additional events (for the POE system). To avoid confusion, you should name these event by prefixing it properly.

A Trivial Example

Let us write a plugin which simply makes the agent wait a certain number of seconds after any request is made. Also, the number of requests should be counted; if that reaches a certain number, then the agent should terminate, whether other plugins still have business to do.

As a convenience, we want to control the waiting time and the URL limit with parameters from the application:

  use WWW::Agent;
  use WWW::Agent::Plugins::GoldCoasting;
  my $l = new WWW::Agent (plugins => [
                                      new WWW::Agent::Plugins::GoldCoasting (wait  => 1,
                                                                             limit => 3),
                                      ]);

Before we start the agent we send an event to fetch a page:

  use POE;
  POE::Kernel->post ('agent', 'cycle_start', 'newtab', new HTTP::Request ('GET', 'http://www.rumsti.org/'));

This will be sent as a first (almost) event to the agent, making it run through the request/response cycle. Now we can start the agent:

  $l->run;

The plugin is simply a package with a new constructor:

   package WWW::Agent::Plugins::GoldCoasting;

   use strict;
   use Data::Dumper;
   use POE;

   sub new {
       my $class   = shift;
       my %options = @_;
       return bless { .... }, $class;
   }

   1;

whereby the constructor must return an object which has the following components:

namespace

As data from all plugins will be stored into one data structure, plugins have to register for a namespace which they will use consistently throughout. Saying this, it is possible to access data of other plugins. Not sure, whether this is good or bad.

hooks

This hash reference contains your handlers, i.e. the events for which your plugin registers interest to be called back together with anonymouse subroutines which actually do something in that case.

In our example this could look like the following:

    return bless {
                  namespace => 'laziness',
                  hooks => {
                      'init' => sub {
                                my ($kernel, $heap)  = (shift, shift);
                                $heap->{laziness}->{wait}       = $options{wait}  || 10;
                                $heap->{laziness}->{limit}      = $options{limit} ||  3;
                                return 1;
                                },
                       # ... maybe more handlers
                  }, $class;

The init event is actually done before anything else and it allows plugins to initialize their private data structure first. In our case we might not be overly interested in a reference to the POE kernel, but the $heap is the place to put the private data. Here we decide to copy relevant configuration information so that we can access it later. Note that we have to return 1 to indicate that everything is fine.

When the agent has completed a request successfully, it will receive an event cycle_pos_response. Here we can add the required wait time and here we can also increment our counter.

            # yet another event
            'cycle_pos_response' => sub {
                my ($kernel, $heap) = (shift, shift);
                my ($tab, $response) = (shift, shift);
                my $url  = $response->request->uri;

                warn "# before $url: working very hard for some secs";
                sleep $heap->{laziness}->{wait}; # you should not use blocking...
                $heap->{laziness}->{counter}++; # we do not care which tab it is

Again we get the POE kernel and the heap as parameters. For this event, though, we also get the tab within which the request was done and the response as an HTTP::Response.

After retrieving the original URL for this response, we wait the necessary seconds. This is actually NOT appropriate within a POE environment as it lets the process block here without being able to process other events.

If the counter is below our limit we simply let the agent request the page again (and again and again) using the same tab and the request which still sits in the response object.

                if ($heap->{laziness}->{counter} < $heap->{laziness}->{limit}) {
                    $kernel->yield ('cycle_start', $tab, $response->request);
                } else {
                    $kernel->yield ('laziness_end', $tab);
                }
                return $response;
                },

Finally the protocol with the agent requires that we return the response object (we could have modified it).

If we have reached the limit, the we post another event to the agent, one which is NOT defined natively by the agent. Instead it is an event which we cover ourselves:

            # yet another event
            'laziness_end' => sub {
                my ($heap) = $_[HEAP];
                warn "# we call it a life-style to stop after ".$heap->{laziness}->{limit}." requests";
            },

This is a fully-fledged POE event. When the plugin is configured, the agent had realized that it does not know about this particular event and has registered it for itself.

Event Classes

The agent distinguishes between the following kinds of events:

native, boolean events

These events are native to the agent and require that the plugin returns a boolean (1 or 0) value to indicate whether a particular condition is met or not.

init

parameters: kernel, heap

cycle_consider

parameters: kernel, heap, tab, request

cycle_complete

parameters: kernel, heap, tab

native filter events

These events expect that a request (or response) object is analyzed, possibly modified and returned. Every event handler this behaves as a filter.

cycle_prepare_request

parameters: kernel, heap, tab, request

cycle_pos_response, cycle_neg_response

parameters: kernel, heap, tab, response

private event handlers

These are POE states (POE is infamous for using misleading names).

Caveats and Tips

Blocking

By choosing POE as an external event loop and by programming the agent to be a reactive, rather than a active one, the agent (and the plugins) can coexist with other reactive POE components.

If this is no concern for you, then there is nothing wrong of letting your plugin block, either doing blocking IO or by sleeping some seconds.

Returning Values

In some cases a plugin might want to return values to the application when the agent has terminated. The simplest way, probably is to memorize

@@@@

SEE ALSO ^

WWW::Agent

AUTHOR ^

Robert Barta, <rho@bigpond.net.au>

COPYRIGHT AND LICENSE ^

Copyright (C) 2005 by Robert Barta

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.8.4 or, at your option, any later version of Perl 5 you may have available.

syntax highlighting: