[SCL] DTD for XML/SCL

[John F. Sowa]

Chris and Bill,

XML is a very flexible markup language, which can be used in many
different ways for may different purposes.  However, it should also
be recognized that it can be easily abused in ways that degrade
performance and add useless verbosity for no real benefit.

SOAP is an example of a reaction against the complexity of Corba.
It uses XML to encode character strings in a much simpler way than
Corba, but in a way that is satisfactory for many applications for
which Corba is overkill.

However, people are now beginning to recognize that for many
applications, the verbosity of the XML tags is a major obstacle
to performance.  They are now looking at ways of replacing SOAP
with solutions that transmit character strings between applications
without the overhead of XML.

Bill's suggestion is a typical example of XML overkill:

 > <implies>
 >   <forall>
 >     <varlist>
 >       <var>x</var>
 >     </varlist>
 >     <atomic>
 >       <term>P</term>
 >       <fnterm>
 >         <term>succ</term>
 >         <term>x</term>
 >       </fnterm>
 >     </atomic>
 >   </forall>
 >   <atomic>
 >     <term>P</term>
 >     <fnterm>
 >       <term>succ</term>
 >       <term>0</term>
 >     </fnterm>
 >   </atomic>
 > </implies>

That takes a very simple one-line formula in KIF and expands it into
unreadable garbage whose only purpose is to exploit a one-size-fits
all XML parser, which will probably translate it back into KIF.

I have been a supporter of the GML-SGML-HTML-XML family of markup
languages since the early 1970s, but only for those applications
that make sense.  Examples like the above might be useful for some
applications, but KIF or CGIF notations are vastly simpler and
vastly preferable for a great many other applications.

Summary:  XML is an extremely flexible notation that can be used
and abused in many ways.  For different purposes, different kinds
of XML markups may be used.  For some purposes, the best way of
using XML is to provide an escape tag with an attribute that says
"language=KIF", "language=CGIF", or "language=Z".  I would prefer
to develop an SCL standard that suggests a couple of encodings in
a non-normative annex and leaves the way open for other standards
to choose appropriate encodings for their applications.

John Sowa