XML and WML. Opera is available for free for several systems.



Html2Wml -- Program that can convert HTML pages to WML pages


Html2Wml can be used as either a shell command:

  $ html2wml file.html

or as a CGI:


In both cases, the file can be either a local file or a URL.


Html2Wml converts HTML pages to WML decks, suitable for being viewed on a Wap device. The program can be launched from a shell to statically convert a set of pages, or as a CGI to convert a particular (potentially dynamic) HTML resource.

Althought the result is not guarantied to be valid WML, it should be the case for most pages. Good HTML pages will most probably produce valid WML decks. To check and correct your pages, you can use W3C's softwares: the HTML Validator, available online at http://validator.w3.org and HTML Tidy, written by Dave Raggett.

Html2Wml provides the following features:

  • translation of the links

  • limitation of the cards size by splitting the result into several cards

  • inclusion of files (similar to the SSI)

  • compilation of the result (using the WML Tools, see LINKS)

  • a debug mode to check the result using validation functions


Please note that most of these options are also available when calling Html2Wml as a CGI. In this case, boolean options are given the value ``1'' or ``0'', and other options simply receive the value they expect. For example, --ascii becomes ?ascii=1 or ?a=1. See the file t/form.html for an example on how to call Html2Wml as a CGI.

Conversion Options

-a, --ascii

When this option is on, named HTML entities are converted to US-ASCII characters using the same 7 bit approximations as Lynx. For example, © is translated to ``(c)'', and ß is translated to ``ss''. This option is off by default.

--collapse, --nocollapse

This option tells Html2Wml to collapse redundant whitespaces, tabulations, carriage returns, lines feeds and empty paragraphs. The aim is to reduce the size of the WML document as much as possible. Collapsing empty paragraphs is necessary for two reasons. First, this avoids empty screens (and on a device with only 4 lines of display, an empty screen can be quite ennoying). Second, Html2wml creates many empty paragraphs when converting, because of the way the syntax reconstructor is programmed. Deleting these empty paragraphs is necessary like cleaning the kitchen :-)

If this really bother you, you can desactivate this behaviour with the --nocollapse option.

-c, --compile

Setting this option tells Html2Wml to use the compiler from WML Tools to compile the WML deck. If you want to create a real Wap site, you should seriously use this option in order to reduce the size of the WML decks. Remember that Wap devices have very little amount of memory. If this is not enought, use the splitting options.


This option tells Html2Wml to completly ignore all image links.

--img-alt-text, --noimg-alt-text

This option tells Html2Wml to replace the image tags with their corresponding alternative text (as with a text mode web browser). This option is on by default.

--linearize, --nolinearize

This option is on by default. This makes Html2Wml flattens the HTML tables (they are linearized), as Lynx does. I think this is better than trying to use the native WML tables. First, they have extremely limited features and possibilities compared to HTML tables. In particular, they can't be nested. In fact this is normal because Wap devices are not supposed to have a big CPU running at some zillions-hertz, and the calculations needed to render the tables are the most complicated and CPU-hogger part of HTML.

Second, as they can't be nested, and as typical HTML pages heavily use imbricated tables to create their layout, it's impossible to decide which one could be kept. So the best thing is to keep none of them.

[Note] Although you can desactivate this behaviour, and although there is internal support for tables, the unlinearized mode has not been heavily tested with nested tables, and it may produce unexpected results.

-n, --numeric-non-ascii

This option tells Html2wml to convert all non-ASCII characters to numeric entities, i.e., © becomes ©, and ß becomes ß. By default, this option is off.

-p, --nopre

This options tells Html2Wml not to use the <pre> tag. This option was added because the compiler from WML Tools 0.0.4 doesn't support this tag.

-o, --output

Use this option (in shell mode) to specify an output file. By default, Html2Wml prints the result to standard output.

Link Reconstruction Options


This options sets the template that will be used to reconstruct the href-type links. See LINKS RECONSTRUCTION for more information.


This option sets the template that will be used to reconstruct the src-type links. See LINKS RECONSTRUCTION for more information.

Splitting Options

-s, --max-card-size=SIZE

This option allows you to limit the size (in bytes) of the generated cards. Default is 1,500 bytes, which should be small enought to be loaded on most Wap devices. See DECK SPLITTING for more information.

-t, --card-split-threshold=SIZE

This option sets the threshold of the split event, which can occur when the size of the current card is between max-card-size - card-split-threshold and max-card-size. Default value is 50. See DECK SPLITTING for more information.


This options sets the label of the link that points to the next card. Default is ``[>>]'', which whill be rendered as ``[>>]``.


This options sets the label of the link that points to the previous card. Default is ``[&lt;&lt;]'', which whill be rendered as ``[<<]``.

Debugging Options

-d, --debug[=LEVEL]

This option activates the debug mode. This prints the output result with line numbering and with the result of the XML check. If the WML compiler was called, the result is also printed in hexadecimal an ascii forms. When called as a CGI, all of this is printed as HTML, so that can use any web browser for that purpose.

When this option is on, it send the WML output to XML::Parser to check its well-formedness.


The deck slicing is a feature that Html2Wml provides in order to match the low memory capabilities of most Wap devices. Many can't handle cards larger than 2,000 bytes, therefore the cards must be sufficiently small to be viewed by all Wap devices. To achieve this, you should compile your WML deck, which reduce the size of the deck by 50%, but even then your cards may be too big. This is where Html2Wml comes with the deck slicing feature. This allows you to limit the size of the cards, currently only before the compilation stage.

Slice by cards or by decks

On some Wap phones, slicing the deck is not sufficient: the WLM browser still tries to download the whole deck instead of just picking one card at a time. A solution is to slice the WML document by decks. See the figure below.

     _____________          _____________ 
    |    deck     |        |   deck #1   |
    |  _________  |        |  _________  |
    | | card #1 | |        | |  card   | |
    | |_________| |        | |_________| |
    |  _________  |        |_____________|
    | | card #2 | |        
    | |_________| |             . . .
    |  _________  |        
    | |   ...   | |         _____________
    | |_________| |        |   deck #n   |
    |  _________  |        |  _________  |
    | | card #n | |        | |  card   | |
    | |_________| |        | |_________| |
    |_____________|        |_____________|
      WML document           WML document
    sliced by cards        sliced by decks

What this means is that Html2Wml generates several WML documents. In CGI mode, only the appropriate deck is sent, selected by the id given in parameter. If no id was given, the first deck is sent.

Note on size calculation

Currently, Html2Wml estimates the size of the card on the fly, by summing the length of the strings that compose the WML output, texts and tags. I say ``estimates'' and not ``calculates'' because computing the exact size would require many more calculations than the way it is done now. One may objects that there are only additions, which is correct, but knowing the exact size is not necessary. Indeed, if you compile the WML, most of the strings of the tags will be removed, but not all.

For example, take an image tag: &lt;img src="images/dog.jpg" alt="Photo of a dog"&gt;. When compiled, the string "img" will be replaced by a one byte value. Same thing for the strings "src" and "alt", and the spaces, double quotes and equal signs will be stripped. Only the text between double quote will be preserved... but not in every cases. Indeed, in order to go a step further, the compiler can also encode parts of the arguments as binary. For example, the string "http://www." can be encoded as a single byte (8F in this case). Or, if the attribute is href, the string href="http:// can become the byte 4B.

As you see, it doesn't matter to know exactly the size of the textual form of the WML, as it will always be far superior to the size of the compiled form. That's why I don't count all the characters that may be actually written.

Also, it's because I'm quite lazy ;-)

Why compiling the WML deck?

If you intent to create real WML pages, you should really consider to always compile them. If you're not convinced, here is an illustration.

Take the following WML code snipet:

    <a href='http://www.yahoo.com/'>Yahoo!</a>

It's the basic and classical way to code an hyperlink. It takes 42 bytes to code this, because it is presented in a human-readable form.

The WAP Forum has defined a compact binary representation of WML in its specification, which is called ``compiled WML''. It's a binary format, therefore you, a mere human, can't read that, but your computer can. And it's much faster for it to read a binary format than to read a textual format.

The previous example would be, once compiled (and printed here as hexadecimal):

    1C 4A 8F 03 y a h o o 00 85 03 Y a h o o ! 00 01

This only takes 20 bytes. Half the size of the human-readable form. For a Wap device, this means both less to download, and easier things to read. Therefore the processing of the document can be achieved in a short time compared to the tectual version of the same document.

There is a last argument, and not the less important: many Wap devices only read binary WML.


Actions are a feature similar to (but with far less functionalities!) the SSI (Server Side Includes) available on good servers like Apache. In order not to interfere with the real SSI, but to keep the syntax easy to learn, it differs in very few points.


Basically, the syntax to execute an action is:

    <!-- [action param1="value" param2='value'] -->

Note that the angle brackets are part of the syntax. Except for that point, Actions syntax is very similar to SSI syntax.

Available actions

Currently, only two actions are available, but more can be implemented on request.



Includes a file in the document at the current point. Please note that Html2Wml doesn't check nor parse the file, and if the file cannot be found, will silently die (this is the same behavior as SSI).


virtual=url -- The file is get by http.

file=path -- The file is read from the local disk.


If you use the file parameter, an absolute path is recommend.



Returns the size of a file at the current point of the document.


virtual=url -- The file is get by http.

file=path -- The file is read from the local disk.


If you use the file parameter, an absolute path is recommend.


If you want to share a navigation bar between several WML pages, you can include it this way:

    <!-- [include virtual="nav.wml"] -->

Of course, you have to write this navigation bar first :-)


The links reconstruction engine is IMHO the most important part of Html2Wml, because it's this engine that allows you to reconstruct the links of the HTML document being converted. It has two modes, depending upon whether Html2Wml was launched from the shell or as a CGI.

When used as a CGI, this engine will reconstructs the links of the HTML document so that all the urls will be passed to Html2Wml in order to convert the pointed files (pages or images). This is completly automatic and can't be customized for now (but I don't think it would be really useful).

When used from the shell, this engine reconstructs the links with the given templates. Note that absolute URLs will be left untouched. The templates can be customized using the following syntax.


HREF Template

This template controls the reconstruction of the href attribute of the A tag. Its value can be changed using the --hreftmpl option. Default value is "{FILEPATH}{FILENAME}{$FILETYPE =~ s/s?html?/wml/o; $FILETYPE}".

Image Source Template

This template controls the reconstruction of the src attribute of the IMG tag. Its value can be changed using the --srctmpl option. Default value is "{FILEPATH}{FILENAME}{$FILETYPE =~ s/gif|png|jpe?g/wbmp/o; $FILETYPE}"


The template is a string that contains the new URL. More precisely, it's a Text::Template template. Parameters can be interpolated as a constant or as a variable. The template is embraced between curcly bracets, and can contain any valid Perl code.

The simplest form of a template is {<em>PARAM</em>} which just returns the value of PARAM. If you want to do something more complex, you can use the corresponding variable; for example {"foo $<em>PARAM</em> bar"}, or {join "_", split " ", <em>PARAM</em>}.

You may read Text::Template for more information on what is possible within a template.

If the original URL contained a query part or a fragment part, then they will be appended to the result of the template.

Available parameters


This parameter contains the original URL from the href or src attribute.


This parameter contains the base name of the file.


This parameter contains the leading path of the file.


This parameter contains the suffix of the file.

This can be resumed this way:

  URL = http://www.server.net/path/to/my/page.html
                             ------------^^^^ ----
                                 |        |     \
                                 |        |      \
                              FILEPATH FILENAME FILETYPE

Note that FILETYPE contains all the extensions of the file, so if its name is index.html.fr for example, FILETYPE contains ``.html.fr''.


To add a path option:


Using Apache, you can then add a Rewrite directive so that URL ending with $wap will be redirected to Html2Wml:

    RewriteRule  ^(/.*)\$wap$  /cgi-bin/html2wml.cgi?url=$1

To change the extension of an image:



Currently, only the well-formedness of the resulting WML can be tested, not its validity.

Inverted tags (like ``<b>bold <i>italic</b></i>'') may produce unexpected results. But only bad softwares do bad stuff like this.




This is the web site of the author, where you can find the archives of all the releases of Html2Wml.

[ http://www.maddingue.org/techie/ ]

Html2Wml on SourceForge

This is the web site of the Html2Wml project, hosted by SourceForge.net. All the stable releases can be downloaded from this site.

[ http://htmlwml.sourceforge.net/ ]


The WAP Forum

This is the official site of the WAP Forum. You can find some technical information, as the specifications of all the technologies associated with the WAP.

[ http://www.wapforum.org/ ]


This site has some useful information and links. In particular, it has a quite well done FAQ.

[ http://www.wap.com/ ]

The World Wide Web Consortium

Altough not directly related to the Wap stuff, you may find useful to read the specifications of the XML (WML is an XML application), and the specifications of the different stylesheet languages (CSS and XSL), which include support for low-resolution devices.

[ http://www.w3.org/ ]


This web site is dedicated to Mobile UniX systems. It leads you to a lot of useful hands-on information about installing and running Linux and BSD on laptops, PDAs and other mobile computer devices.

[ http://www.mobilix.org/ ]

Programmers utilities


This is a very handful utility which corrects your HTML files so that they conform to W3C standards.

[ http://www.w3.org/People/Raggett/tidy ]


Kannel is an open source Wap and SMS gateway. A WML compiler is included in the distribution.

[ http://www.kannel.org/ ]

WML Tools

This is a collection of utilities for WML programmers. This include a compiler, a decompiler, a viewer and a WBMP converter.

[ http://pwot.co.uk/wml/ ]

WML browsers and Wap emulators

Opera 5.0

Opera is originaly a Web browser, but the version 5 has a good support for

[ http://www.opera.com/ ]


wApua is an open source WML browser written in Perl/Tk. It's easy to intall and to use. Its support for WML is incomplete, but sufficient for testing purpose.

[ http://fsinfo.cs.uni-sb.de/~abe/wApua/ ]


Tofoa is an open source Wap emulator written in Python. Its installation is quite difficult, and its incomplete WML support makes it produce strange results, even with valid WML documents.

[ http://tofoa.free-system.com/ ]


EzWAP, from EZOS, is a commercial WML browser freely available for Windows 9x, NT, 2000 and CE. Compared to others Windows WML browsers, it requires very few resources, and is quite stable. Its support for the WML specs seems quite complete. A very good software.

[ http://tofoa.free-system.com/ ]


Deck-It is a commercial Wap phone emulator, available for Windows and Linux/Intel only. It's a very good piece of software which really show how WML pages are rendered on a Wap phone, but one of its major default is that it cannot read local files.

[ http://www.pyweb.com/php/test_adapt.php3 ]


WinWAP is a commercial Wap browser, freely available for Windows.

[ http://www.winwap.org/ ]


QWmlBrowser (formerly known as WML BRowser) is an open source WML browser, written using the Qt toolkit.

[ http://www.wmlbrowser.org/ ]


WAPreview is a Wap emulator written in Java. As it uses an HTML based UI and needs a local web proxy, it runs quite slowly.

[ http://wapreview.sourceforge.net ]


This is a Wap emulator written in Java. It uses the Java native GUI, and claims to support binary WML, but it doesn't seem to work at all at this time.

[ http://j2wap.sourceforge.net ]


Werner Heuser, for his numerous ideas, advices and his help for the debugging

Igor Khristophorov, for his numerous suggestions and patches

And all the people that send me bug reports: Daniele Frijia, Axel Jerabek


Sébastien Aperghis-Tramoni <maddingue@free.fr>