29 September, 2000

Production Notes (Non-Normative)

Gavin Nicol, Inso EPS

The DOM specification serves as a good example of the power of using XML: all of the HTML documents, Java bindings, OMG IDL bindings, and ECMA Script bindings are generated from a single set of XML source files. This section outlines how this specification is written in XML, and how the various derived works are created.

A. The Document Type Definition

This specification was written entirely in XML, using a DTD based heavily on the DTD used by the XML Working Group for the XML specification. The major difference between the DTD used by the XML Working Group, and the DTD used for this specification is the addition of a DTD module for interface specifications.

The DTD module for interfaces specifications is a very loose translation of the Extended Backus-Naur Form (EBNF) specification of the OMG IDL syntax into XML DTD syntax. In addition to the translation, the ability to describe the interfaces was added, thereby creating a limited form of literate programming for interface definitions.

While the DTD module is sufficient for the purposes of the DOM WG, it is very loosely typed, meaning that there are very few constraints placed on the type specifications (the type information is effectively treated as an opaque string). In a DTD for object to object communication, some stricter enforcement of data types would probably be beneficial.

B. The production process

The DOM specification is written using XML. All documents are valid XML. In order to produce the HTML versions of the specification, the object indexes, the Java source code, and the OMG IDL and ECMA Script definitions, the XML specification is converted.

The tool currently used for conversion is COST by Joe English. COST takes the ESIS output of nsgmls, creates an internal representation, and then allows scripts, and event handlers to be run over the internal data structure. Event handlers allow document patterns and associated processing to be specified: when the pattern is matched during a pre-order traversal of a document subtree, the associated action is executed. This is the heart of the conversion process. Scripts are used to tie the various components together. For example, each of the major derived data sources (Java code etc.) is created by the execution of a script, which in turn executes one or more event handlers. The scripts and event handlers are specified using TCL.

The current version of COST has been somewhat modified from the publicly available version. In particular, it now runs correctly under 32-bit Windows, uses TCL 8.0, and correctly handles the case sensitivity of XML (though it probably could not correctly handle native language markup).

We could also have used Jade, by James Clark. Like COST, Jade allows patterns and actions to be specified, but Jade is based on DSSSL, an international standard, whereas COST is not. Jade is more powerful than COST in many ways, but prior experience of the editor with Cost made it easier to use this rather than Jade. A future version or Level of the DOM specification may be produced using Jade or an XSL processor.

The complete XML source files are available at: http://www.w3.org/TR/2000/WD-DOM-Level-1-20000929/xml-source.zip

Note: The DOM Level 1 Specification Second Edition has been produced using a DOM Level 2 implementation and an XPath implementation in Java.

C. Object Definitions

As stated earlier, all object definitions are specified in XML. The Java bindings, OMG IDL bindings, and ECMA Script bindings are all generated automatically from the XML source code.

This is possible because the information specified in XML is a superset of what these other syntax need. This is a general observation, and the same kind of technique can be applied to many other areas: given rich structure, rich processing and conversion are possible. For Java and OMG IDL, it is basically just a matter of renaming syntactic keywords; for ECMA Script, the process is somewhat more involved.

A typical object definition in XML looks something like this:

<interface name="foo">
  <descr><p>Description goes here...</p></descr>
  <method name="bar">
    <descr><p>Description goes here...</p></descr>
      <param name="baz" type="DOMString" attr="in">
        <descr><p>Description goes here...</p></descr>
    <returns type="void">
       <descr><p>Description goes here...</p></descr>
       <!-- Throws no exceptions -->  

As can easily be seen, this is quite verbose, but not unlike OMG IDL. In fact, when the specification was originally converted to use XML, the OMG IDL definitions were automatically converted into the corresponding XML source using common Unix text manipulation tools.