HTML::Bare - Minimal HTML parser implemented via a C state engine




  use HTML::Bare;
  my $ob = new HTML::Bare( text => '<html><name>Bob</name></html>' );
  # Parse the html into a hash tree
  my $root = $ob->parse();
  # Print the content of the name node
  print $root->{html}->{name}->{value};
  # Load html from a file ( assume same contents as first example )
  my $ob2 = new HTML::Bare( file => 'test.html' );
  my $root2 = $ob2->parse();
  $root2->{html}->{name}->{value} = 'Tim';
  # Save the changes back to the file
  # Load html and verify against XBS ( HTML Bare Schema )
  my $html_text = '<html><item name=bob/></html>''
  my $schema_text = '<html><item* name=[a-z]+></item*></html>'
  my $ob = new HTML::Bare( text => $html_text, schema => { text => $schema_text } );
  $ob->parse(); # this will error out if schema is invalid


This module is a 'Bare' HTML parser. It is implemented in C. The parser itself is a simple state engine that is less than 500 lines of C. The parser builds a C struct tree from input text. That C struct tree is converted to a Perl hash by a Perl function that makes basic calls back to the C to go through the nodes sequentially.

The parser itself will only cease parsing if it encounters tags that are not closed properly. All other inputs will parse, even invalid inputs. To allowing checking for validity, a schema checker is included in the module as well.

The schema format is custom and is meant to be as simple as possible. It is based loosely around the way multiplicity is handled in Perl regular expressions.

Supported HTML

To demonstrate what sort of HTML is supported, consider the following examples. Each of the PERL statements evaluates to true.

  • Node containing just text

      HTML: <html>blah</html>
      PERL: $root->{html}->{value} eq "blah";
  • Subset nodes

      HTML: <html><name>Bob</name></html>
      PERL: $root->{html}->{name}->{value} eq "Bob";
  • Attributes unquoted

      HTML: <html><a href=index.htm>Link</a></html>
      PERL: $root->{html}->{a}->{href}->{value} eq "index.htm";
  • Attributes quoted

      HTML: <html><a href="index.htm">Link</a></html>
      PERL: $root->{html}->{a}->{href}->{value} eq "index.htm";
  • CDATA nodes

      HTML: <html><raw><![CDATA[some raw $~<!bad html<>]]></raw></html>
      PERL: $root->{html}->{raw}->{value} eq "some raw \$~<!bad html<>";
  • Multiple nodes; form array

      HTML: <html><item>1</item><item>2</item></html>
      PERL: $root->{html}->{item}->[0]->{value} eq "1";
  • Forcing array creation

      HTML: <html><multi_item/><item>1</item></html>
      PERL: $root->{html}->{item}->[0]->{value} eq "1";
  • One comment supported per node

      HTML: <html><!--test--></html>
      PERL: $root->{html}->{comment} eq 'test';

Schema Checking

Schema checking is done by providing the module with an XBS (HTML::Bare Schema) to check the HTML against. If the HTML checks as valid against the schema, parsing will continue as normal. If the HTML is invalid, the parse function will die, providing information about the failure.

The following information is provided in the error message:

  • The type of error

  • Where the error occured ( line and char )

  • A short snippet of the HTML at the point of failure

XBS ( HTML::Bare Schema ) Format

  • Required nodes

      HTML: <html></html>
      XBS: <html/>
  • Optional nodes - allow one

      HTML: <html></html>
      XBS: <html item?/>
      or XBS: <html><item?/></html>
  • Optional nodes - allow 0 or more

      HTML: <html><item/></html>
      XBS: <html item*/>
  • Required nodes - allow 1 or more

      HTML: <html><item/><item/></html>
      XBS: <html item+/>
  • Nodes - specified minimum and maximum number

      HTML: <html><item/><item/></html>
      XBS: <html item{1,2}/>
      or XBS: <html><item{1,2}/></html>
      or XBS: <html><item{1,2}></item{1,2}></html>
  • Multiple acceptable node formats

      HTML: <html><item type=box volume=20/><item type=line length=10/></html>
      XBS: <html><item type=box volume/><item type=line length/></html>
  • Regular expressions checking for values

      HTML: <html name=Bob dir=up num=10/>
      XBS: <html name=[A-Za-z]+ dir=up|down num=[0-9]+/>
  • Require multi_ tags

      HTML: <html><multi_item/></html>
      XBS: <html item@/>

Parsed Hash Structure

The hash structure returned from HTML parsing is created in a specific format. Besides as described above, the structure contains some additional nodes in order to preserve information that will allow that structure to be correctly converted back to HTML.

Nodes may contain the following 3 additional subnodes:

  • _i

    The character offset within the original parsed HTML of where the node begins. This is used to provide line information for errors when HTML fails a schema check.

  • _pos

    This is a number indicating the ordering of nodes. It is used to allow items in a perl hash to be sorted when writing back to html. Note that items are not sorted after parsing in order to save time if all you are doing is reading and you do not care about the order.

    In future versions of this module an option will be added to allow you to sort your nodes so that you can read them in order. ( note that multiple nodes of the same name are stored in order )

  • _att

    This is a boolean value that exists and is 1 iff the node is an attribute.

Parsing Limitations / Features

  • CDATA parsed correctly, but stripped if unneeded

    Currently the contents of a node that are CDATA are read and put into the value hash, but the hash structure does not have a value indicating the node contains CDATA.

    When converting back to HTML, the contents of the value hash are parsed to check for html incompatible data using a regular expression. If 'CDATA like' stuff is encountered, the node is output as CDATA.

  • Node position stored, but hash remains unsorted

    The ordering of nodes is noted using the '_pos' value, but the hash itself is not ordered after parsing. Currently items will be out of order when looking at them in the hash.

    Note that when converted back to HTML, the nodes are then sorted and output in the correct order to HTML. Note that nodes of the same name with the same parent will be grouped together; the position of the first item to appear will determine the output position of the group.

  • Comments are parsed but only one is stored per node.

    For each node, there can be a comment within it, and that comment will be saved and output back when dumping to HTML.

  • Comments override output of immediate value

    If a node contains only a comment node and a text value, only the comment node will be displayed. This is in line with treating a comment node as a node and only displaying immediate values when a node contains no subnodes.

  • PI sections are parsed, but discarded

  • Unknown <! sections are parsed, but discarded

  • Attributes may use no quotes, single quotes, quotes, or backtics

  • Quoted attributes cannot contain escaped quotes

    No escape character is recognized within quotes. As a result, regular quotes cannot be stored to HTML, or the written HTML will not be correct, due to all attributes always being written using quotes.

  • Attributes are always written back to HTML with quotes

  • Nodes cannot contain subnodes as well as an immediate value

    Actually nodes can in fact contain a value as well, but that value will be discarded if you write back to HTML. That value is equal to the first continuous string of text besides a subnode.

      ( the value of node is text )
      ( the value of node is text )
      ( the value of node is "\n  " )
  • Entities are not parsed

    No entity parsing is done. This is intentional. Future versions of the module may include a feature to automatically parse entities, but by default any such feature will be disabled in order to keep from slowing down the parser.

    Also, this is done so that round trip ( read and then write back out ) behavior is consistent.

  • Nodes named value

    Previously iterations of this module had problems with nodes named 'value', due to the fact that node contents are stored under the 'value' key already. The current version should parse such files without any problem, although it may be confusing to see a parsed tree with 'value' pointing to another hash containing 'value' as well.

    In a future version of the module it will be possible to alter the name that values are stored under.

    Note that node values are stored under the key 'content' when the "simple" parsing mode is used, so as to be consistent with HTML::Simple.

Module Functions

  • $ob = HTML::Bare->new( text => "[some html]" )

    Create a new HTML object, with the given text as the html source.

  • $object = HTML::Bare->new( file => "[filename]" )

    Create a new HTML object, with the given filename/path as the html source

  • $object = HTML::Bare->new( text => "[some html]", file => "[filename]" )

    Create a new HTML object, with the given text as the html input, and the given filename/path as the potential output ( used by save() )

  • $object = HTML::Bare->new( file => "data.html", scheme => { file => "scheme.xbs" } )

    Create a new HTML object and check to ensure it is valid html by way of the XBS scheme.

  • $tree = $object->parse()

    Parse the html of the object and return a tree reference

  • $tree = $object->simple()

    Alternate to the parse function which generates a tree similar to that generated by HTML::Simple. Note that the sets of nodes are turned into arrays always, regardless of whether they have a 'name' attribute, unlike HTML::Simple.

    Note that currently the generated tree cannot be used with any of the functions in this module that operate upon trees. The function is provided purely as a quick and dirty way to read simple HTML files.

  • $tree = htmlin( $htmlext, keeproot => 1 )

    The htmlin function is a shortcut to creating an HTML::Bare object and parsing it using the simple function. It behaves similarly to the HTML::Simple function by the same name. The keeproot option is optional and if left out the root node will be discarded, same as the function in HTML::Simple.

  • $text = $object->html( [root] )

    Take the hash tree in [root] and turn it into cleanly indented ( 2 spaces ) HTML text.

  • $text = $object->html( [root], [root node name] )

    Take the hash tree in [root] and turn it into nicely colorized and styled html. [root node name] is optional.

  • $object->save()

    The the current tree in the object, cleanly indent it, and save it to the file parameter specified when creating the object.

  • $value = xval $node, $default

    Returns the value of $node or $default if the node does not exist. If default is not passed to the function, then '' is returned as a default value when the node does not exist.

  • ( $name, $age ) = xget( $personnode, qw/name age/ )

    Shortcut function to grab a number of values from a node all at the same time. Note that this function assumes that all of the subnodes exist; it will fail if they do not.

  • $text = HTML::Bare::clean( text => "[some html]" )

    Shortcut to creating an html object and immediately turning it into clean html text.

  • $text = HTML::Bare::clean( file => "[filename]" )

    Similar to previous.

  • HTML::Bare::clean( file => "[filename]", save => 1 )

    Clean up the html in the file, saving the results back to the file

  • HTML::Bare::clean( text => "[some html]", save => "[filename]" )

    Clean up the html provided, and save it into the specified file.

  • HTML::Bare::clean( file => "[filename1]", save => "[filename2]" )

    Clean up the html in filename1 and save the results to filename2.

  • $html = HTML::Bare::tohtml( text => "[some html]", root => 'html' )

    Shortcut to creating an html object and immediately turning it into html. Root is optional, and specifies the name of the root node for the html ( which defaults to 'html' )

  • $object->add_node( [node], [nodeset name], name => value, name2 => value2, ... )

        $object->add_node( $root->{html}, 'item', name => 'Bob' );
  • $object->add_node_after( [node], [subnode within node to add after], [nodeset name], ... )

  • $object->del_node( [node], [nodeset name], name => value )

        Starting HTML:
          $html->del_node( $root->{html}, 'a', b=>'1' );
        Ending HTML:
  • $object->find_node( [node], [nodeset name], name => value )

        Starting HTML:
          $object->find_node( $root->{html}, 'ob', key => '1' )->{val}->{value} = 'test';
        Ending HTML:
  • $object->find_by_perl( [nodeset], "[perl code]" )

    find_by_perl evaluates some perl code for each node in a set of nodes, and returns the nodes where the perl code evaluates as true. In order to easily reference node values, node values can be directly referred to from within the perl code by the name of the node with a dash(-) in front of the name. See the example below.

    Note that this function returns an array reference as opposed to a single node unlike the find_node function.

        Starting HTML:
          $object->find_by_perl( $root->{html}->{ob}, "-key eq '1'" )->[0]->{val}->{value} = 'test';
        Ending HTML:
  • HTML::Bare::merge( [nodeset1], [nodeset2], [id node name] )

    Merges the nodes from nodeset2 into nodeset1, matching the contents of each node based up the content in the id node.


        my $ob1 = new HTML::Bare( text => "
          </html>" );
        my $ob2 = new HTML::Bare( text => "
          </html>" );
        my $root1 = $ob1->parse();
        my $root2 = $ob2->parse();
        merge( $root1->{'html'}->{'a'}, $root2->{'html'}->{'a'}, 'id' );
        print $ob1->html( $root1 );
  • HTML::Bare::del_by_perl( ... )

    Works exactly like find_by_perl, but deletes whatever matches.

  • HTML::Bare::forcearray( [noderef] )

    Turns the node reference into an array reference, whether that node is just a single node, or is already an array reference.

  • HTML::Bare::new_node( ... )

    Creates a new node...

  • HTML::Bare::newhash( ... )

    Creates a new hash with the specified value.

  • HTML::Bare::simplify( [noderef] )

    Take a node with children that have immediate values and creates a hashref to reference those values by the name of each child.

  • HTML::Bare::hash2html( [hashref] )

    Take a recursive hash tree ( perhaps generated by the simplify function ) and turn it into a raw HTML string. Note that this function does not indent nicely. You will need to feed this string back into the parser and output it again if you want it to look nice. ( or you could use the 'clean' function to do it in one go )

  • HTML::Bare->new( text => "[html]", unsafe => 1 )

    An extra speedy way to parse HTML. It is unsafe; may harm pets and children. Don't say you weren't warned. 30% speed boost compared to the normal parsing. You -must- use $ob->simple() in combination with this for it to work properly.

    The speed boost is gained by skipping checks for the end of the string when in the middle of properly formatted HTML. The only time the check is done is within "values" ( which includes the space after the final closing </html> )

    Also, in the unsafe mode, tags, complete with their attributes, must be on one line. Node contents of course, can still have carriage returns...

  • $object->read_more( text => "[html fragment]" )

    Add more HTML text to be handled. Note that this function must be called before calling the parse function.


        my $ob = HTML::Bare->new( text => "
            <node>a</node>" );
        $ob->read_more( text => "<node>b</node>" );
        $ob->read_more( text => "</html>" );
        my $root = $ob->parse();
        print $ob->html( $root );

    Warning! Reading in additional HTML fragments only works properly at proper "division points". Currently the parser will -not- work properly if you split in the middle of a node value, or in the middle of a node name. A future version of the module will be properly updated to handle these cases.

    Currently there is little to no benefit to parsing this way, rather than simple concatenating the two strings together and then reading all the HTML in at once.

Functions not yet documented

  • find_by_att() find_by_attr() find_by_id() find_by_idr() find_by_tagname() find_by_tagnamer() nav()

Functions Used Internally

  • check() checkone() readxbs() free_tree_c()

  • lineinfo() c_parse() c_parse_unsafely() c_parse_more() c_parsefile() free_tree() html2obj()

  • obj2htmlcol() get_root() obj2html() html2obj_simple()


Since the creation of this module there has been a fair amount of controvesy surrounding it. A number of authors of other HTML parsers have gone so far as to publicly attack this module and claim that it 'does not parse HTML', and 'it is not HTML compliant'. Some of the same people seem to be angered by the inclusion of a benchmark, claiming that it is an unfair comparison, and that if the proper options and setup are used, that other HTML parsers are better.

The module should parse any HTML document that conforms to the standardized HTML specifications, there is no need for alarm and fear that the module will corrupt your HTML documents on reading.

To be blunt about how the parser works, very little has been done to make the parser follow the specification known as 'HTML'. The parser is meant to be flexibile and somewhat resilient, and will parse HTML like garbage that would cause other parsers to error out.

As far as I am concerned, as the author of the module, the 'HTML' in 'HTML::Bare' should be thought of to mean 'eXtremely Mad Language', because the module was written from scratch without referring to the specification known as 'HTML'.

In regard to the complaints about the unfairness of the included benchmarks, please make your own intelligent decision as to what module you like by trying multiple modules and/or running the performance tests yourself. If you like some other module, use that module. If you like HTML::Bare and think it is the fastest thing on the planet, that is cool too.

If you hate HTML::Bare and want to go around on the internet trashing it and telling people to use something else, I think perhaps you may want to seek counseling.


In comparison to other available perl html parsers that create trees, HTML::Bare is extremely fast. In order to measure the performance of loading and parsing compared to the alternatives, a templated speed comparison mechanism has been created and included with HTML::Bare.

The include file runs when you make the module and creates perl files within the bench directory corresponding to the .tmpl contained there.

Currently there are three types of modules that can be tested against, executable parsers ( exe.tmpl ), tree parsers ( tree.tmpl ), and parsers that do not generated trees ( notree.tmpl ).

A full list of modules currently tested against is as follows:

  EzHTML (exe)
  Tiny HTML (exe)
  HTML::Descent (notree)
  HTMLIO (exe)
  HTML::LibHTML (notree)
  HTML::Parser (notree)
  HTML::Parser::Expat (notree)
  HTML::Simple using HTML::Parser
  HTML::Simple using HTML::SAX::PurePerl
  HTML::Simple using HTML::LibHTML::SAX::Parser
  HTML::Simple using HTML::Bare::SAX::Parser

To run the comparisons, run the appropriate perl file within the bench directory. (,, or )

The script measures the milliseconds of loading and parsing, and compares the time against the time of HTML::Bare. So a 7 means it takes 7 times as long as HTML::Bare.

Here is a combined table of the script run against each alternative using the included test.html:

  -Module-                   load     parse    total
  HTML::Bare                  1        1        1
  HTML::TreePP                2.3063   33.1776  6.1598
  HTML::Parser::EasyTree      4.9405   25.7278  7.4571
  HTML::Handler::Trees        7.2303   26.5688  9.6447
  HTML::Trivial               5.0636   12.4715  7.3046
  HTML::Smart                 6.8138   78.7939  15.8296
  HTML::Simple (HTML::Parser)  2.3346   50.4772  10.7455
  HTML::Simple (PurePerl)     2.361    261.4571 33.6524
  HTML::Simple (LibHTML)       2.3187   163.7501 23.1816
  HTML::Simple (HTML::Bare)    2.3252   59.1254  10.9163
  HTML::SAX::Simple           8.7792   170.7313 28.3634
  HTML::Twig                  27.8266  56.4476  31.3594
  HTML::Grove::Builder        7.1267   26.1672  9.4064
  HTML::XPath::HTMLParser      9.7783   35.5486  13.0002
  HTML::LibHTML (notree)       11.0038  4.5758   10.6881
  HTML::Parser (notree)       4.4698   17.6448  5.8609
  HTML::Parser::Expat(notree) 3.7681   50.0382  6.0069
  HTML::Descent (notree)      6.0525   37.0265  11.0322
  Tiny HTML (exe)                               1.0095
  EzHTML (exe)                                  1.1284
  HTMLIO (exe)                                  1.0165

Here is a combined table of the script run against each alternative using the included feed2.html:

  -Module-                   load     parse    total
  HTML::Bare                  1        1        1
  HTML::Bare (simple)         1        0.7238   ?
  HTML::Bare (unsafe simple)  1       ~0.5538   ?
  HTML::Fast                  1.516    0.9733   1.4783
  HTML::TreePP                0.6393   30.5951  2.6874
  HTML::MyHTML                 1.8266   14.2571  2.7113 
  HTML::Parser::EasyTree      1.5208   22.8283  2.9748 
  HTML::Trivial               2.007    25.742   3.615  
  HTML::Tiny                  0.1665   61.4918  4.3234  
  HTML::XPath::HTMLParser      2.5762   33.2567  4.6742  
  HTML::Smart                 1.702    59.4907  5.7566
  HTML::Simple (HTML::Parser)  0.5838   64.7243  5.0006  
  HTML::DOM::Lite             4.5207   17.4617  5.4033
  HTML::Simple (LibHTML)       0.5904   161.7544 11.5731
  HTML::Twig                  8.553    56.9034  11.8805 
  HTML::Grove::Builder        7.2021   30.7926  12.9334
  HTML::Handler::Trees        6.8545   33.1007  13.0575
  HTML::LibHTML::Simple        14.0204  11.8482  13.8707
  HTML::Simple (PurePerl)     0.6176   321.3422 23.0465 
  HTML::Simple                2.7168   90.7203  26.7525
  HTML::SAX::Simple           8.7386   94.8276  29.2166
  HTML::LibHTML (notree)       11.0023  5.022    10.5214
  HTML::Parser (notree)       4.3748   25.0213  5.9803
  HTML::Parser::Expat(notree) 3.6555   51.6426  7.4316
  HTML::Descent (notree)      5.9206   155.0289 18.7767
  Tiny HTML (exe)                               1.2212
  EzHTML (exe)                                  1.3618
  HTMLIO (exe)                                  1.0145

These results show that HTML::Bare is, at least on the test machine, running all tests within cygwin, faster at loading and parsing than everything being tested against.

The following things are shown as well: - HTML::Bare can parse HTML and create a hash tree in less time than it takes LibHTML just to parse. - HTML::Bare can parse HTML and create a tree in less time than all three binary parsers take just to parse. - HTML::Fast is theoretically faster at parsing than the default 'full' mode of HTML::Bare. Despite that, the 'simple' mode of HTML::Bare is even faster.

Note that the executable parsers are not perl modules and are timed using dummy programs that just uses the library to load and parse the example files. The executables are not included with this program. Any source modifications used to generate the shown test results can be found in the bench/src directory of the distribution


  Copyright (C) 2008 David Helkowski
  This program is free software; you can redistribute it and/or
  modify it under the terms of the GNU General Public License as
  published by the Free Software Foundation; either version 2 of the
  License, or (at your option) any later version.  You may also can
  redistribute it and/or modify it under the terms of the Perl
  Artistic License.
  This program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  GNU General Public License for more details.