[Mirrored from: http://www5conf.inria.fr/fich_html/slides/dday/sgml/all.htm]

An SGML-Based Web Server

Jon Bosak, SunSoft

What is SGML?

• Standard Generalized Markup Language: the international standard for structured document interchange, ISO 8879 (1986)
• Developed out of a search for a universal typesetting language begun in the late 1960s
• Descriptive, not procedural - the procedures are left to formatting/presentation systems
• Not a single markup language, but a metalanguage for the specification of an unlimited number of markup languages, each optimized for a particular category of documents
• The SGML description of a markup language is called a Document Type Definition (DTD)

Major industry DTDs (markup languages)

HTML is just one of many standardized special-purpose SGML markup languages.

HTML vs. most other SGML languages

Basic document model from DocBook:

<!ELEMENT Book - - ((Title, TitleAbbrev?)?, BookInfo?, ToC?, LoT*, Preface*,
                (((%chapter.gp;)+, Reference*) | Part+ | Reference+ |
                Article+), (%appendix.gp;)*, Glossary?, Bibliography?,
                (%index.gp;)*, LoT*, ToC? ) +(%ubiq.gp;) >
[...]
<!ELEMENT Chapter - - (DocInfo?, Title, TitleAbbrev?, (%sect1.gp;), (Index |
                Glossary | Bibliography)*) +(%ubiq.gp;) >
[...previously defined:]
<!ENTITY % sect1.gp "((%component.gp;)+, (Sect1* | RefEntry*)) | Sect1+ |
                RefEntry+" >
[...]
<!ELEMENT Sect1 - - (Title, TitleAbbrev?, (%nav.gp;)*, (((%component.gp;)+,
                (RefEntry* | Sect2*)) | RefEntry+ | Sect2+), (%nav.gp;)*)
                +(%ubiq.gp;) >

Basic document model from HTML 2.0:

<!ENTITY % html.content "HEAD, BODY">
<!ELEMENT HTML O O  (%html.content)>

<!ENTITY % body.content "(%heading | %text | %block | HR | ADDRESS)*">
<!ELEMENT BODY O O  %body.content>

HTML documents differ from documents marked up in most other standard SGML languages in that they lack a controlled hierarchical structure.

Implications of the HTML content model

• Difficult or impossible to validate document data structures (so that documents can be safely dropped into a database, for example) or to impose editorial control in projects with multiple authors
• Difficult or impossible to automatically generate navigational aids (dynamic tables of contents, etc.) directly from the document itself
• Navigation must generally be implemented by adding handcrafted hypertext links
• HTML browsers have no concept of entity management (no modular reuse)
• Context searching becomes difficult or impossible

HTML is too limited to serve as an adequate data format for large-scale commercial publishing.

HTML tools vs. generic SGML tools

• HTML tools are hardwired to a particular tag set, often containing proprietary extensions. Generic SGML tools allow designers to add tags at will; there is no such thing as a proprietary extension in generic SGML.
• Current HTML tools are hardwired to a particular typographic format. Generic SGML tools allow formatting and other presentational characteristics to be controlled by one or more output specifications.
• Future HTML tools will support stylesheets, but not other aspects of output control typically supported by generic SGML tools.

  (For example, in technical publishing, it is often necessary to generate different versions of an SGML document by selectively showing different portions of the source. The same capability is needed in magazine publishing to generate market-specific editions. It is also often necessary to show elements in an order different from their order in the source.)

HTML as a server format

Advantages of HTML on the server

• Cheap
• Easy (at first)

. . . but it does not scale

Example:

• Book with a five-level hierarchy
• Five subdivisions at each level

3125 manually created hypertext links are required to make a table of contents for this one book

And what about revisions?

HTML as a client format

An HTML-based Web server is limited to a flat, unorganized (or tediously handcrafted) document space. A generic SGML Web server, on the other hand, delivers the power of a hierarchical object-oriented document database.

Advantages of generic SGML on the server

• Based on international standards immune to current and future Internet politics, competing vendor strategies, and ad hoc HTML extensions
• Fully extensible markup language is completely under publisher control and precisely suited to the documents
• Documents can be formally validated and editorial guidelines enforced by structured authoring tools
• Links and navigational aids can be generated directly from the structure of documents
• Context searching increases the speed of user access 10-20 times over flat document databases
• Documents can easily be reused for different purposes
• All versions of a document (printed and online) can be generated from the same source
• User-selectable stylesheets allow dynamically configurable views of a documents (not just different typographical treatments)
• System administration of large document repositories is vastly simplified
• Lays the foundation for future deployment of object-oriented authoring/publishing systems

Some SGML-based Web servers

http://occam.sjf.novell.com:8080/docs/toc.pubs_server.html

http://www.sgi.com/Technology/TechPubs/lib/display.cgi?4097

http://cobweb.sybase.com:8000/

And see:

http://www.w3.org/pub/Conferences/WWW4/Papers/112

Case study: Novell

Note: The speaker left Novell to work for SunSoft in January, 1996. All descriptions of Novell's document server are valid as of that date but should not be taken as necessarily descriptive of Novell's current direction. However, everything said in this presentation may be taken in a general way as applying to SunSoft's current direction.

Novell's problem (1991-1994)

They started with this...

And needed to get to all of these...

Wrong answer

This is an m x n solution.

Right answer

This is an m + n solution.

The Novell Publications Server (January 1996)

The Novell Publications Server (future)

The next step: generic SGML on the Web

Example:

http://www.ncsa.uiuc.edu/SDG/Software/WinMosaic/Viewers/panorama.htm

then

http://www.sq.com

The case for generic SGML on the Web

• SGML-to-HTML servers are complex and CPU-intensive; only large corporations can afford them. By shifting more of the processing load to the client, generic SGML browsers can deliver many of the advantages of structured documents at a much lower cost.
• Structured SGML provides the basis for presentational controls far beyond what can be accomplished with any current form of HTML.
• Generic SGML allows for the transmission of much richer data to client-side applications (especially Java applets); "SGML gives Java something to do."

  Generic SGML is rich enough to support distributed document processing (not just distributed document rendering); HTML is not.

• Generic SGML is required whenever structured data from a database system must be processed at the client before transmission to some other database system.

  The interchange format must be capable of capturing all of the information in the source and conveying it to the target. HTML cannot do this.

Examples of distributed document processing requiring generic SGML

• TOCs (tables of contents) and library catalogs downloaded to the browser for increased performance
  • Key requirement: hierarchical data structure
• Patient medical histories from a hospital pasted into the entrance forms for a home care agency
  • Key requirements: data validation, controlled authoring, industry-specific standardized markup
• Semiconductor data from multiple manufacturers used to drive a circuit-modeling application
  • Key requirements: industry-specific standardized markup, rich semantic tagging
• Distributed interactive airline scheduling with client-side itinerary optimization
  • Key requirements: controlled authoring, rich semantic tagging
• Corporate travel authorization/expense program implemented as an "intranet" application
  • Key requirements: controlled authoring, easily extensible markup, tools that automatically adapt to changed markup