Data layer in dynamic applications and presentation formatting

Jan Pazdziora

Scheduled: Iterative Software room on Friday at 16:20-17:00 (paper # 30)

Abstract: To achieve multiple user designs and page layouts of Web applications, including output for paper media or handheld devices from one code base, the structure of the data must be well defined and described. It can then be transparently filled with output structures of mod_perl/Apache::Registry handlers, producing XML data output. Using AxKit, a series of transformations can be applied to this XML data, producing various outputs depending on many factors.

The approach presented is aimed at large systems with hundreds of applications where embedding the code into HTML markup leads to unmanageable setup and code lost inside of formatting details. The handlers are run in mod_perl environment and return only Perl hashes, programmers are not slowed down by generating HTML tags. XML data stream is created on the fly, available for logging, auditing or testing, and XSLT and other transformations produce client-friendly formatting on the server side.

As the markup is not produced by function calls in applications but rather rendered in single place, using fast XSLT transformers, the performance is comparable with templating approaches, but with data being strictly separated from presentation. The approach allows using the applications as Web enabled components using for example SOAP, and to combine them together, suitable for toolbars or sideboxes. Multilingual pages can be served without any language specific code in individual applications, thanks to dynamic processing offered by AxKit.


Single presentation point for hundreds of applications

Information system of my university consists of over 300 perl scripts run in mod_perl environment. We are happy with having the applications isolated into individual files that can be used immediately when copied into proper directory tree. But we feel we need more flexibility in creating and maintaining consistent presentation, with possibly more designs of the web site to choose from, more language versions, and support for more output media.

The requirements:

Even if Perl is very good for text processing and for HTML generation alike, application program should never emit HTML. Why? Once its code is tied to HTML, it's very hard (impossible?) to make it to also output WML, PDF or SOAP responses. Code would have to be duplicated and that is something that should be avoided.

Solution? Have application produce only data, in standard, compatible form. Any output presentation can be created from the data. Applications will get much simpler and presentation will be centralized into one, easily maintainable point. XML seems like a good choice these days; AxKit then adds instant framework that only needs to be extended a bit to support that kind of dynamic applications we need, and mod_perl provides the performance we expect.

Interpolation into XML

We could generate the XML using

	print "<name>$name</name>\n";
	print "<age>$computed_age</age>\n";
calls but that still leaves the burden of watching for and escaping characters that are special in XML on application programmer, not mentioning the basic requirement of producing well formed XML with matching tag names.

Since our scripts will be run in mod_perl environment and they will be thus invoked as subroutines inside of Apache::Registry or other dispatcher code, we can return native Perl structures from them:

	return {
		'name' => $name,
		'age' => $computed_age,
		'cars' => $dbh->selectall_arrayref(q!
			select year_from, year_until, make
			from car_owners
			where owner = ?
			!, {}, $id),
		'owner_id' => $id,
	};

These data structures can be serialized into XML automatically, by conversion of hashes and arrays into XML elements. Alternatively, the data can be interpolated into structural description of application's data which can be provided in another, external XML source. In our example, the structural description might be

  <application>
    <search_cars>
      <inputbox name="owner_id"/>
    </search_cars>
    <more_owners_found>
      <datalist name="more_owners">
        <id/><name/>
      </datalist>
    </more_owners_found>
    <owner_info>
      <data name="name"/><data name="age"/>
      <datalist name="cars">
        <year_from/><year_to/><car_make/>
      </datalist>
    </owner_info>
  </application>
and it both describes the logical structure of the application and its input parameter (which may be used for parameter validation), and provides placeholders for data returned from application.

After interpolation, the resulting XML is

  <application>
    <search_cars>
      <inputbox name="owner_id">72588</inputbox>
    </search_cars>
    <more_owners_found>
    </more_owners_found>
    <owner_info>
      <name>John Smith</name>
      <age>36</age>
      <cars>
      <row num="1"><year_from>1993</year_from>
	<year_to>1998</year_to>
	<car_make>Saab</car_make></row>
      <row num="2"><year_from>1997</year_from>
	<year_to></year_to>
	<car_make>Volvo</car_make></row>
      </cars>
    </owner_info>
  </application>

Application thus consists of the description of its structure and the actual code. While the first could be omitted, it provides good communication tool among application analyst, application programmer and presentation designers, and it can also serve for result verification.

Presentation in AxKit

In AxKit, we can write our own handlers that will setup presentation transformations based on user's requirements. Once the data is available in XML, we can use standard chain of transformers to provide desired design of pages, and also include output of some other standard components on the output page ("news should be displayed on top of all output pages, besides those intended for printing"). We can even provide localized stylesheets based on user's preferences (sent by browser in HTTP headers, for example). We have also been able to get PDF output, generated via stylesheet that creates TeX source and running pdfTeX on it. For non-authenticated pages, AxKit's builtin caching mechanism can speedup response and save CPU cycles.

The biggest problem with multiple format output seems to be design of good structure of presentation stylesheets. With hundreds of applications, hundreds of XML element names may appear in the system and it is not practical to write new transformation for all of them in each new design. Some sort of generalization is needed, which would make it possible to keep the size and number of stylesheets under control.

The bundle of modules that provide the interpolation of applications output into XML and run necessary transformation on that data layer is called RayApp and is available. For actual deployment, modification to local specifics may be needed. That includes setup of paths to stylesheets and handling of their localized versions; some coding rather than just configuring will probably be involved.

Future of this approach

Only a small part of existing system was converted to use the XML data layer. It provides great advantage in really separating presentation from content -- application programmer doesn't care about HTML markup, and applications do not even know if the actual request is for HTML output or for WML. All the application has to do is to run desired actions with data in database and produce output data structures.

With AxKit framework, all necessary pieces could be put together fast -- Apache::Registry(NG) for application execution, LibXSLT for XSLT transformation and Perl and its modules for the glue around it. Management of presentation stylesheets seems to be the hardest part, but it is not unique to this approach.

For environments where the number of individual applications is relatively small, systems with embedded code may be superior. If the number of application gets high, like with large systems, strictly separating presentation from code is the way to go.