付爸爸的神奇日記本

This article provides an update on the popular and mature CGI::Application framework for web applications. It assumes a basic understanding of the system, so reviewing the previous Perl.com article about CGI::Application may be helpful background reading.

CGI::Application and Catalyst Compared

You may recall the Perl.com article on another MVC web framework, Catalyst. First, I want to clear up possible confusion by explaining how the CGI::Application and Catalyst relate.

With the many plugins available for CGI::Application and Catalyst, both frameworks offer many of the same features.

Both provide convenient methods to access many of the same underlying modules including Data::FormValidator, HTML::FillInForm and templating systems such as Template Toolkit and HTML::Template.

Both frameworks work in CGI and mod_perl environments, although CGI::Application loads faster in CGI. Each one provides unique features to help with development and debugging. Catalyst includes a built-in web server for easy offline testing and development. CGI::Application provides a persistent development pop-up window that provides convenient reports on HTML validation, application performance, and more.

While CGI::Application and Catalyst share many of the same strengths, they also face the same challenge of attracting users and developers.

As the expectations for web site features and quality increase, the toolkit that a web developer depends on must increase and expand as well. The more plugins that are compatible with our framework, the easier our job is. We each have a selfish incentive to attract others to use the same framework. Users become contributors, and contributors write plugins to make our lives easier.

CGI::Application and Catalyst already share users and developers of many of the Perl modules they depend on. With PHP and Ruby on Rails both on the rise as web development solutions, those who prefer Perl have an incentive to promote the best the language has to offer.

What A Difference Half A Decade Makes

CGI::Application development took off around the 4.0 release for two reasons. To start with, it formalized a plugin system, which led to the release of some initial plugins. Next, the 4.0 release added a callback system, allowing the plugin authors to automatically add actions that take place at particular points in the request cycle.

For example, the AutoRunmode plugin registers itself at the "prerun" phase, allowing it to adjust which run mode is selected. Another plugin might register to add a cleanup action in the "teardown" phase.

The combined result was a boom in plugin development. While the core of CGI::Application has remained small and stable, there are over three dozen plugins now on the CPAN.

Here's a tour of some of the enhancements that have come about as a result of the new plugin and callback system.

Simplified Runmode Syntax

The built-in way to register a run mode typically involves calling run_modes() within setup():

sub setup {
   my $self = shift;
   $self->run_modes([qw/
        my_run_mode
   /]);
}

# later...
sub my_run_mode {
 ...
}

With the AutoRunmode plugin, it's now very easy to declare that a method is a "run mode" handling a CGI request rather than an internal function. The syntax is simply:

    sub my_run_mode : Runmode {
        my $self = shift;
        # ...    
    }

You can still use setup(), but it's no longer necessary.

by Mark Stosberg

New Dispatcher Provides Clean URLS

A large project built with CGI::Application typically has many small "instance scripts" that drive modules full of run modes. All of these instance scripts look basically the same:

use Project::Widget::View;
my $app = Project::Widget::View->new();
$app->run();

A corresponding URL might look like:

/cgi-bin/project/widget/view.cgi?widget_id=23

CGI::Application::Dispatch 2.0 allows you to replace all of these instance scripts with a single dispatch script to produce much cleaner URLs. Such a dispatch script might look like:

#!/usr/bin/perl
use CGI::Application::Dispatch;
CGI::Application::Dispatch->dispatch(
        prefix              => '',
        table               => [
            ':app/:rm/:id'  => {},
        ],

);

Now add a dash of mod_rewrite magic provided in the documentation, and the URL will transform from:

/cgi-bin/project/widget/view.cgi/detailed_view?widget_id=23

to:

/widget/detailed_view/23

Clean and simple.

The widget_id is easily accessible from within the run mode:

my $widget_id = $self->param('id');

The dispatcher takes care of that detail for you, saving some manual munging of PATH_INFO.

First-Class Templating Support

While CGI::Application integrates by default with HTML::Template, it seems an equal number of the users prefer the Template Toolkit templating system. Both camps now have access to several new templating-related features.

Default template names

To keep the code cleaner and consistent, it's now possible to generate template names automatically. Typically, you want to load one template to correspond with each run mode. Simply loading a template might look like:

    sub my_run_mode: Runmode {
        my $self = shift;

        # XXX The old way, with redundant file name 
        # my $t = $self->load_tmpl('my_run_mode.html');

        # Look Ma! No explicit file name needed!
        my $t = $self->load_tmpl;
        return $t->output;
    }

Easy access to the application object from the template

The TT plugin introduced easy access to the CGI::Application object from the template, allowing easy constructions by using the c parameter to access the application object.

Hello [% c.session.param('username') || 'Anonymous User' %]
<a href="[% c.query.self_url %]">Reload this page</a>

Authors of open source web applications will surely appreciate the AnyTemplate plugin, which allows you to use a single templating syntax in your code, and lets users choose the templating system that best integrates with their existing project. There was no ready-made way to do this in the past.

Conveniently, HTML::Template and TT users can use a familiar syntax to drive AnyTemplate.

TT style:

$self->template->process('file_name', \%params);

HTML::Template style:

# Yes, it really can be identical to the standard load_tmpl() syntax!
my $template = $self->load_tmpl('file_name');
$template->param('foo' => 'bar');
$template->output;

A great example of this template abstraction is CGI::Application::Search, a reusable application that integrates with the Swish-E search engine. Whether you prefer HTML::Template or Template Toolkit, it's easy to add this as a search solution for a larger project--even if the rest of your website does not use CGI::Application.

CGI::Application also offers improved support for other output formats. The Stream plugin makes it a snap to stream a document to the user, such as a PDF or Excel file that is built on the fly. This saves the busy work of remembering the related details for unbuffered output, binmode, file chunking, and MIME types. That now takes basically one line of syntax:

$self->stream_file( $file );

The XSV plugin simplifies building CSV files. This tedium is now a single function call for simple cases:

  return $self->xsv_report_web({
    fields     => \@headers,
    # Get values from the database
    values     => $sth->fetchall_arrayref( {} );
    csv_opts   => { sep_char => "\t" },
    filename   => 'members.csv',
  });

by Mark Stosberg

Lazier Than Ever

One frequent feature you'll find in CGI::Application plugins is lazy loading. This means that loading and configuring the plugin often has little resource penalty. Take the DBH plugin. It's convenient to configure the database handle once for a whole website project and then use the handle whenever you want.

Before this plugin arrived, it would be tempting to stuff the database handle into the param method to achieve a similar effect:

sub cgiapp_init {
    my $self = shift;

    my $dbh = DBI->connect($data_source, $username, $auth, \%attr);

    # save for later!
    $self->param('dbh',$dbh);
}

That works OK, but it misses a valuable feature: lazy loading.

Lazy loading creates the database connection only if any code needs to use it. This avoids needlessly creating a database connection for scripts that don't need it, while still being very convenient. Here's an example:

# define the database connection parameters once in a super class, for a whole
# suite of child applications:

sub cgiapp_init  {
   my $self = shift;

   # use the same args as DBI->connect(); 
   $self->dbh_config($data_source, $username, $auth, \%attr);

}

Then, whenever you need a database handle:

sub my_run_mode : Runmode {

    my $self = shift;

    my $result = $self->dbh->selectrow("...");

    # ...

}

Easy. dbh_config() will get called on every request, but it simply stores the configuration details. The database handle gets created only during calls to the dbh() method.

Another notable lazy-loading plugin is the Session plugin, which provides easy access to a CGI::Session object. It further takes advantage of the CGI::Application framework by automatically setting the session cookie for you, so you don't have to deal with cookies unless you want to.

Ready for High-Performance Environments

I mostly use CGI::Application with plain CGI, because it performs well enough and works well in a shared hosting environment since no resources persist between requests.

However, CGI::Application is ready for high-performance applications.

Often, code written for CGI::Application will run without changes under mod_perl's Apache::Registry mode, as 1-800-Save-A-Pet.com does.

To squeeze a little more juice out of mod_perl, there is an Apache plugin, which uses Apache::Request instead of CGI.pm.

A current popular alternative for increasing performance is FastCGI. Use CGI::Application::FastCGI, and add, usually, just one line of code to make your application work in this environment.

by Mark Stosberg

Easy Form Handling

A lot of tedium can be involved in processing web forms. The first plugin, ValidateRM, helped with that.

my $results =  $self->check_rm('display_form', '_form_profile' ) 
   || return $self->dfv_error_page;

This simple syntax calls a Data::FormValidator profile into action. If validation fails, the page with the original form is redisplayed with the previous values intact, and error messages appear next to each field that is missing or invalid.

Fans of Data::FormValidator will appreciate an upcoming related module from the CGI::Application community: JavaScript::DataFormValidator.

This module makes it easy to use the same Perl data structure to add an additional level of validation in JavaScript. I expect Catalyst and CGI::App users alike will be putting this to use.

Finally, there's a new plugin to simplify filling in a web form from a database record. This is the FillInForm plugin. The syntax is simple:

   # fill in the HTML form from the query object
    $self->fill_form($html);

In part, this plugin solves bug #13913 in HTML::FillInForm, which means the interface detects what kind of input you are giving it, rather than requiring you to explicitly declare that you have hashref or scalarref and so forth. As you can see from the example, if you are using the query object as input, you don't need to pass it in at all.

DevPopUp: A Unique Developer Tool

CGI::Application offers a unique developer tool in the form of the DevPopUp plugin. You can see DevPopUp in action on Rhesa's demo site. (Make sure your pop-up blocker doesn't trap it!).

The tool creates a persistent pop-up window that gives you feedback about each run mode as soon as it completes. "What were the HTTP Headers? How long did it take? Was the resulting HTML valid?"

The real kicker is that DevPopUp is itself pluggable, allowing other developers to add their own reports. I look forward to Sam Tregar releasing his graphical DBI profiling tool, which would make a nice addition here.

Easier Error Messages with DebugScreen

CGI::Application users in Japan recently brought us the DebugScreen plugin. This is a welcome change from referencing the web server log to find the most recent line that needs debugging.

Hello, Web 2.0: AJAX Integration

In another story of cross-pollination with Catalyst, CGI::Application integrates easily with the JavaScript Prototype library. Prototype provides easy access to plenty of interesting AJAX effects, such as auto-completing based on a lookup to the server. This uses a thin plugin-wrapping HTML::Prototype, which was written with Catalyst in mind.

CGI::Application and Catalyst

The appearance of Catalyst has been a great benefit to the CGI::Application community. Both projects support the use and development of focused, reusable components. When a new patch arrives for a module both projects use, both projects benefit. As the maintainer of Data::FormValidator, I'm well aware that the two camps are collaborating through this project.

Catalyst releases its code under a license that allows CGI::Application to reuse and integrate their work (and vice-versa). Often a plugin written for Catalyst takes only a little effort to port to work with CGI::Application. For example, Catalyst recently added PAR support, which allows the distribution and execution of a complex web application as a single binary. This helps a Perl project with complex module dependencies compete with the installation ease of typical PHP software.

This will be a great reference as CGI::Application users evaluate options for easier web application deployment.

Finally, Catalyst demonstrates an alternate approach as a web framework. This helps the CGI::Application community better evaluate their own options.

Conclusion

CGI::Application has always been about providing a clean structure for web applications. With the advent of myriad plugins, it is also about simplifying access to the many great tools that Perl offers web developers through the CPAN. With more plugins being developed on a regular basis, the life of the web developer is getting easier by the day.

Perl.com Compilation Copyright © 1998-2006 O'Reilly Media, Inc.