Help-Site Computer Manuals
  Algorithms & Data Structures   Programming Languages   Revision Control
  Cameras   Computers   Displays   Keyboards & Mice   Motherboards   Networking   Printers & Scanners   Storage
  Windows   Linux & Unix   Mac

HTML::Tree subclass for HTML templating via tree rewriting

HTML::Seamstress - HTML::Tree subclass for HTML templating via tree rewriting


HTML::Seamstress - HTML::Tree subclass for HTML templating via tree rewriting


HTML::Seamstress provides ``fourth generation'' dynamic HTML generation (templating).

In the beginning we had...

First generation dynamic HTML production

First generation dynamic HTML production used server-side includes:

 <p>Today's date is   <!--#echo var="DATE_LOCAL" --> </p>

Second generation dynamic HTML production

The next phase of HTML generation saw embedded HTML snippets in Perl code. For example:

 sub header {

   my $title = shift;

   print <<"EOHEADER";






Third generation dynamic HTML production

The 3rd generation solutions embed programming language constructs with HTML. The language constructs are either a real language (as is with the HTML::Mason manpage) or a pseudo/mini-language (as is with PeTaL, Template or the HTML::Template manpage). Let's see some Template code:

 <p>Hi there [% name %], are you enjoying your stay?</p>

Talkin' bout them generations...

Up to now, all approaches to this issue tamper with the HTML in some form or fashion:

Enter fourth generation dynamic HTML production

The fourth generation of HTML production is distinguished by no need for tampering with the HTML. There are a wealth of XML-based modules which provide this approach (the XML::Twig manpage, the XML::LibXML manpage, the XML::TreeBuilder manpage, the XML::DOM manpage). HTML::Seamstress is the one CPAN module based around HTML and the HTML::Tree manpage for this approach.


When looking at HTML::Seamstress, we are looking at a uniquely positioned 4th-generation HTML generator. Seamstress offers two sets of advantages: those common to all 4th generation htmlgens and those common to a subclass of the HTML::Tree manpage.

Reap 4th generation dynamic HTML generation benefits

What advantages does this fourth way of HTML manipulation offer? Let's take a look:

Guarantee yourself well-formed HTML

Because lower-generation dynamic HTML generators treat HTML as a string, there is no insurance against poorly formed HTML.

Take a look at these two Mason components, from :

  • Example 5-3. /autohandler
      % $m->call_next;
      <%method .body_tag>
        $bgcolor => 'white'
        $textcolor => 'black'
       <body onLoad="prepare_images( )" bgcolor="<% $bgcolor %>" text="<% $textcolor %>">
  • Example 5-4. /important_advice.mas
      <head><title>A Blue Page With Red Text</title></head>
      <& SELF:.body_tag, bgcolor=>'blue', textcolor=>'red' &>
       Never put anything bigger than your elbow into your ear.

There is nothing guaranteeing that open tags will match close tags or that close tags will even exist. To make the correspondence between open and close tags even more troublesome, they are in different files. And it is not easy for an HTML designer and/or design tool to manipulate things once they have been shredded apart like this.

With the tree-based approach of Seamstress, the end tag will exist and it will match the open tag. Well-formedness is job 1 in tree-based HTML rewriting!

HTML will be properly escaped

Separate HTML development and its programmatic modification

Software engineers refer to this as orthogonality. The contents of the document remain legal HTML/XML that can be be developed using standard interactive design tools. The flow of control of the code remains separate from the page. Technologies that mix content and data in a single file result in code that is often difficult to understand and has trouble taking full advantage of the object oriented programming paradigm.

Work at meta-level instead of object-level

The book ``Godel, Escher, Bach: An Eternal Golden Braid'' by Douglas R Hofstadter makes it clear what it means to operate at object-level as opposed to meta-level. When you buy into earlier-generation HTML generation systems you are working at object-level: you can only speak and act as the HTML with no ability to speak about the HTML.

Compare a bird's eye view of a city with standing on a city block and you have the difference between the 4th generation of HTML development versus all prior generations.

Reduced learning curve

If you have a strong hold on object-oriented Perl and a solid understand of the tree-based nature of HTML, then all you need to do is read the manual pages showing how Seamstress and related modules offer tree manipulation routines and you are done.

Extension just requires writing new Perl methods - a snap for any object oriented Perler.

Static validation and formatting

Mixing Perl and HTML (by any of the generation 1-3 approaches) makes it impossible to use standard validation and formatting tools for either Perl or HTML.

Two full-strength programming languages: HTML and Perl

Perl and HTML are solid technologies with years of effort behind making them robust and flexible enough to meet real-world technological demands.

Object-oriented reuse and extension of HTML

Class-based object-oriented programming makes use of inheritance and other techniques to achieve maximum code reuse. This typically happens by a certain base/superclass method containing common actions and a derived/subclass/mixin method containing extra actions.

A genuine tree-based approach (such as HTML::Seamstress) to HTML generation is supportive of all methods of object-oriented reuse: because manipulator and manipulated are separate and manipulators are written in oo Perl, we can compose manipulators as we please.

This is in contrast to inline simple object systems (as in Mason) and also in contrast to the if-then approach of tt-esque systems.

Per-page stereotyped substitution

[ FYI: you can run the two Seamstress approaches. They are in $DISTRO/samples/perpage ]

In the HTML::Mason book by O'Reilly:

we see a technique for doing simple text insertion which varies per page:


  <head><title>Welcome to Wally World!</title></head>

  <body bgcolor="#CCFFCC">

  <center><h1><% $m->base_comp->attr('head') %></h1></center>

  % $m->call_next;

  <center><a href="/">Home</a></center>


 # homepage.html


   head => "Wally World Home"


  Here at Wally World you'll find all the finest accoutrements.

 # productpage.html


   head => "Wally World Products"



 <table> ... </table>

So, how would we do this using Seamstress' pure Perl approach to HTML refinement?



  <head><title>Welcome to Wally World!</title></head>

  <body bgcolor="#CCFFCC">

  <center><h1 id=head>DUMMY_HEAD</h1></center>

  <span id=body>DUMMY_BODY</span>

  <center><a href="/">Home</a></center>



 package html::homepage;

 use base qw( HTML::Seamstress ) ;

 sub new {

  my ($class, $c) = @_;

  my $html_file = 'html/base.html';

  my $tree = __PACKAGE__->new_from_file($html_file);



 sub process {

  my ($tree, $c, $stash) = @_;

  $tree->content_handler(head => 'Wally World Home');

  $tree->content_handler(body => 

   'Here at Wally World you'll find all the finest accoutrements.');



 package html::productpage;

 use base qw( HTML::Seamstress ) ;

 sub new {

  my ($class, $c) = @_;

  my $html_file = 'html/base.html';

  my $tree = __PACKAGE__->new_from_file($html_file);



 sub process {

  my ($tree, $c, $stash) = @_;

  $tree->content_handler(head => 'Wally World Products);

  $tree->content_handler(body => html::productpage::body->new->guts)


We have solved our problem. However, we can create even more re-use because the both of these classes are very similar. They only vary in 2 things: the particular head and body they provide. You can abstract this with whatever methodmaker you like. I tend to prefer prototype-based oop over class-based, so with Class::Prototyped, here's how we might do it:

 package html::abstract::common;

 use base qw(HTML::Seamstress Class::Prototyped);

 sub head { 'ABSTRACT BASE METHOD' }

 sub body { 'ABSTRACT BASE METHOD' }


  html_file => 'html/base.html',


 sub new {

  my $self = shift;

  my $tree = $self->new_from_file($self->html_file);


 sub process {   

  my ($tree, $c, $stash) = @_;

  $tree->content_handler(head => $tree->head);

  $tree->content_handler(body => $tree->body);



and then have both of the above classes instantiate and specialize this common class accordingly.

[ Again: you can run the two Seamstress approaches. They are in 

F<$DISTRO/samples/perpage> ]

Parallel generation of a single page natural

A tree of HTML usually contains subtrees with no inter-dependance. They therefore can be manipulated in parallel. If a page contains 5 areas each of which takes N time, then one could realize an N-fold speedup.

Reap the benefits of using HTML::Tree

Pragmatic HTML instead of strict X(HT)ML

The real world is unfortunately more about getting HTML to work with IE and maybe 1 or 2 other browsers. Strict XHTML may not be acceptable under time and corporate pressures to get things to work with quirky browsers.

Rich API and User Contributions

the HTML::Tree manpage has a nice large set of accessor/modifier functions. If that is not enough, then take a gander at Matthew Sisk's contributions: as well as the HTML::Element::Library manpage.

Seamstress contains no voodoo elements whatsoever

If you know object-oriented Perl and know how to rewrite trees, then everything that Seamstress offers will make sense: it's just various boilerplates and scripts that allow your mainline code to be very succinct: think of it as Class::DBI for HTML::Tree.

Seamstress is here to help you use HTML::Tree, that's all.

Unify HTML and the processing of the HTML via a Perl class

Let's see why this is a good idea. In Mason, your Perl and HTML are right there together in the same file. Same with Template. Now, since Seamstress operates on the HTML without touching the HTML, the operations and the HTML are not in the same file. So we create a Perl module to glue the HTML file to the operations we plan to perform on it.

This module (auto-created by and perhaps has a constructor new(), which grabs the HTML file and constructs an HTML::Element tree from it and returns it to you.

It also contains a process() subroutine which processes the HTML in some way: text substitutions, unrolling list elements, building tables, and whatnot.

Finally, it contains a fixup() subroutine. This subroutine is designed to support the meat-skeleton paradigm, discussed above. The process() subroutine generated the $meat. After <$meat> has been placed in $skeleton, there may be some page-specific processing to the whole HTML page that you want to: pop in some javascript, remove a copyright notice, whatever. That's what this routine is for.

Now that I've said all that, please understand that you are perfectly free to call new() and do what you want with the HTML tree. You don't have to use process() and fixup(). But they are there and are used by the Catalyst::View::Seamstress manpage to make meat-skeleton dynamic HTML development quick-and-easy (and non-greasy).

A Perl class created by

Here is our venerable little HTML file:

 metaperl@pool-71-109-151-76:/ernest/dev/catalyst-simpleapp/MyApp/root/html$ cat hello_world.html 

 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">



    <title>Hello World</title>



  <h1>Hello World</h1>

    <p>Hello, my name is <span id="name">dummy_name</span>.

    <p>Today's date is <span id="date">dummy_date</span>.



Now let's abstract this as a Perl class:

 metaperl@pool-71-109-151-76:/ernest/dev/catalyst-simpleapp/MyApp/root/html$ --base_pkg=MyApp::View::Seamstress --base_pkg_root=`pwd`/../../lib hello_world.html

 comp_root........ /ernest/dev/catalyst-simpleapp/MyApp/root/

 html_file_path... /ernest/dev/catalyst-simpleapp/MyApp/root/html/

 html_file........ hello_world.html

 html_file sans... hello_world

 hello_world.html compiled to package html::hello_world


Now lets see what html::hello_world looks like. Everything other than process() was auto-generated:

 package html::hello_world;

 use strict;

 use warnings;

 use HTML::TreeBuilder;

 use base qw(MyApp::View::Seamstress);

 our $tree;

 sub new {

  my $file = __PACKAGE__->comp_root() . 'html/hello_world.html' ;

  -e $file or die "$file does not exist. Therefore cannot load";

  $tree =HTML::TreeBuilder->new;




  bless $tree, __PACKAGE__;


 sub process {

  my ($self, $c, $stash) = @_;

  $tree->look_down(id => $_)->replace_content($stash->{$_})

      for qw(name date);


 sub fixup { $tree }


The meat-skeleton paradigm

This section is written to help understanding of the Catalyst::View::Seamstress manpage for people who want to use Seamstress as the view for their Catalyst apps.

HTML pages typically have meat and a skeleton. The meat varies from page to page while the skeleton is fairly (though not completely) static. For example, the skeleton of a webpage is usually a header, a footer, and a navbar. The meat is what shows up when you click on a link on the page somewhere. While the meat will change with each click, the skeleton is rather static.

The perfect example of

Mason accomodates the meat-skeleton paradigm via an autohandler and $m->call_next(). Template accomodates it via its WRAPPER directive.

And Seamstress? Well, here's what you _can_ do:

  1. generate the meat, $meat
  2. This is typically what you see in the body part of an HTML page

  3. generate the skeleton, $skeleton
  4. This is typically the html, head, and maybe some body

  5. put the meat in the skeleton

So, nothing about this is forced. This is just how I typically do things and that is why Catalyst::View::Seamstress has support for this.

In all honesty, the meat-skeleton paradigm should be supported here and called from Catalyst::View::Seamstress. But the problem is, I don't want to create an abstract API here unless I have used the meat-skeleton paradigm from one other framework besides Catalyst. Then I will have a good idea of how to refactor it so any framework can make good use of the paradigm.


The best example of usage is the Quickstart directory in this distribution. You can read the HTML::Seamstress::Quickstart manpage and actually run the code in that directory at the same time. After doing so, the following sections are additional instruction.

Understand that HTML is a tree

The best representation of this fact is this slide right here:

If you understand this (and maybe the rest of the slides), then you have a good grip on seeing HTML as a tree.

the HTML::Tree::AboutTrees manpage does also teach this, but it takes a while before he gets to what matters to us. It's a fun read nonetheless.

Now that we've got this concept under our belts let's try some full examples.

Install and Setup Seamstress

The first thing to remember is that Seamstress is really just convenience functions for HTML::Tree. You can do entirely without Seamstress. It's just that my daily real-world obligations have lead to a set of library functions (HTML::Element::Library) and a convenient way to locate ``templates'' ( that work well on top of HTML::Tree

  • move and onto your execution $PATH
  • and are used to simplify the process of parsing an HTML file into HTML::Treebuilder object. In other words instead of having to do this in your Perl programs:
     use HTML::TreeBuilder;
     my $tree = HTML::TreeBuilder->new_from_file('/usr/htdocs/hello.html');

    You can do this:

     use htdocs::hello;
     my $tree = htdocs::hello->new;

    The lines of code is not much different, but abstracting away absolute paths is important in production environments where the absolute path may come from who knows where via who knows how.

  • run
  • will ask you 2 very simple questions. Just answer them. When it is finished, it will have installed a package named HTML::Seamstress::Base on your @INC. This module contains one function, comp_root() which points to a place you wouldn't typically have on your @INC but which you must have because your HTML file and corresponding .pm abstracting it are going to be there.

  • run
  • In the default seutp, no options need be supplied to this script. They are useful in cases where you have more than one document root or want to inherit from more than one place.
     metaperl@pool-71-109-151-76:~/www$ moose.html
     comp_root........ /home/metaperl/
     html_file_path... /home/metaperl/www/
     html_file........ moose.html
     html_file sans... moose
     moose.html compiled to package www::moose

  • load your abstracted HTML and manipulate it
  • Now, from Perl, to get the TreeBuilder object representing this HTML file, we simply do this:
     use www::moose;
     my $tree = www::moose->new;
     # manipulate tree...

    In a mod_perl setup, you would want to pre-load your HTML and Class::Cache was designed for this very purpose. But that's a topic for another time.

    In a setup with HTML files in numerous places, I recommend setting up multiple HTML::Seamstress::Base::here, HTML::Seamstress::Base::there for each file root. To do this, you will need to use the --base_pkg and --base_pkg_root options to

  • That's it!
  • Now you are ready to abstract away as many files as you want with the same call. Just supply it with a different HTML file to create a different package. Then use them, new them and manipulate them and $tree->as_HTML them at will.

    Now it's time to rock and roll!

Text substitution == node mutation

In our first example, we want to perform simple text substitution on the HTML template document:



   <title>Hello World</title>



 <h1>Hello World</h1>

   <p>Hello, my name is <span id="name">dummy_name</span>.

   <p>Today's date is <span id="date">dummy_date</span>.



First save this somewhere on your document root. Then compile it with Now you simply use the ``compiled'' version of HTML with API calls to HTML::TreeBuilder, HTML::Element, and HTML::Element::Library.

 use html::hello_world; 


 my $tree = html::hello_world->new; 

 $tree->look_down(id => name)->replace_content('terrence brannon');

 $tree->look_down(id => date)->replace_content('5/11/1969');

 print $tree->as_HTML;

replace_content() is a convenience function in the HTML::Element::Library manpage.

If-then-else == node(s) deletion

 <span id="age_dialog">

    <span id="under10">

       Hello, does your mother know you're 

       using her AOL account?


    <span id="under18">

       Sorry, you're not old enough to enter 

       (and too dumb to lie about your age)


    <span id="welcome">




Again, compile and use the module:

 use html::age_dialog;

 my $tree = html::dialog->new;


    (age_dialog =>


      under10 => sub { $_[0] < 10} , 

      under18 => sub { $_[0] < 18} ,

      welcome => sub { 1 }




  print $tree->as_HTML;

  # will only output one of the 3 dialogues based on which closure 

  # fires first

And once again, the function we used is the highlander method, also a part of the HTML::Element::Library manpage.

The following libraries are always available for more complicated manipulations:

Looping == child/sibling proliferation

Table unrolling, pulldown creation, li unrolling, and dl unrolling are all examples of a tree operation in which you take a child of a node and clone it and then alter it in some way (replace the content, alter some of its attributes), and then stick it under its parent.

Functions for use with the common HTML elements --- <table>, <ol>, <ul>, <dl>, <select> are documented in the HTML::Element::Library manpage and are prefaced with the words ``Tree Building Methods''.

What Seamstress offers

Beyond the ``compilation'' support documented above, Seamstress offers nothing more than a simple structure-modifying method, expand_replace(). And to be honest, it probably shouldn't offer that. But once, when de-Mason-izing a site, it was easier to keep little itty-bitty components all over and so I wrote this method to facilitate the process.

Let's say you have this HTML:

     <div id="sidebar">

        <div class="sideBlock" id=mpi>mc::picBar::index</div>

        <div class="sideBlock" id=mnm>mc::navBox::makeLinks</div>

        <div class="sideBlock" id=mg>mc::gutenBox</div>


In this case, the content of each sideBlock is the name of a Perl Seamstress-style class. As you know, when the constructor for such a class is called an HTML::Element, $E, will be returned for it's parsed content.

In this case, we want the content of the div element to go from the being the class name to being the HTML::Element that the class constructs. So to inline all 3 tags you would do the following;

 $tree->look_down(id => $_)->expand_replace for qw(mpi mnm mg);

What Seamstress works with


Useful in mod_perl environments and anywhere you want control over the timing of object creation.

The family of HTML::Tree contributions



This does the same thing as the TreeBuilder new_from_file() method, but it blesses the object into the invocant class. This makes the invocant class derive from Seamstress which means it has HTML::TreeBuilder, HTML::Element , and HTML::Element::Library at its disposal.


This method takes __FILE__, and optionally a desired $extension (defaults to 'html' if not given) and changes the extension on __FILE__ from .pm to $extension. This works well for common situations.


HTML Templating as Tree Rewriting: Part I: ``If Statements''

HTATR II: HTML table generation via DWIM tree rewriting

Survey of Surveys on HTML Templating systems

A fierce head-to-head between PeTaL and Seamstress goes on for several days in this thread!

The disadvantages of mini-languages

The disadvantages of mini-languages is discussed here:

A striking example of the limitations of mini-languages is shown here:

But the most cogent argument for using full-strength languages as opposed to mixing them occurs in the the Text::Template manpage docs:

 When people make a template module like this one, they almost always

 start by inventing a special syntax for substitutions. For example,

 they build it so that a string like %%VAR%% is replaced with the

 value of $VAR. Then they realize the need extra formatting, so they

 put in some special syntax for formatting. Then they need a loop, so

 they invent a loop syntax. Pretty soon they have a new little

 template language.

 This approach has two problems: First, their little language is

 crippled. If you need to do something the author hasn't thought of,

 you lose. Second: Who wants to learn another language? You already

 know Perl, so why not use it?

And for the Mason users whose retort is ``we do use Perl!'' the obvious reply is: ``granted, but in an embedded fashion with ad hoc, inflexible object mechanisms, non-tree-based (hence syntactically suspect) HTML manipulation, and no ability to statically validate the Perl or HTML''

Problems with JSP (JSP is similar to HTML::Mason)

Los Angeles Perl Mongers Talk on HTML::Seamstress

XMLC, A similar framework for Java

Similar Frameworks for Perl

Two other frameworks come to mind. Both are stricter with regard to the correctness of the HTML and both use a different means for node lookup and rewrite.

  • Petal
  • Based on Zope's TAL, this is a very nice and complete framework that is the basis of MkDoc, a XML application server. It offers a mini-language for XML rewriting, Seamstress does not. The philosophy of the Seamstress is the orthogonal integration of Perl and HTML not a mini-language and HTML.

  • the XML::LibXML manpage
  • By the XML guru Matt Sergeant, who is also the author of AxKit, another XML application server. This offers XPath for finding nodes

  • the XML::DOM manpage
  • If I wanted to ape XMLC entirely, I would have used TJ Mather's the XML::DOM manpage. Because XMLC is based around DOM API calls. However, TreeBuilder is very handy and has a lot of nice libraries around it such the HTML::PrettyPrinter manpage. The biggest win of XML::DOM is it's easy integration with the XML::Generator manpage

    From the docs, it looks like the XML::GDOME manpage is the successor to this module.

SUPPORT and DEVELOPMENT is the jumping-off page for all things Seamstress on the web.

Mailing List


Terrence Brannon,


I would like to thank


Copyright 1999-2006 by Terrence Brannon.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.