XML is not extensible

[Bill de hÓra <bill.dehora@propylon.com>]

Rick Jelliffe wrote:

    >>>
And similarly, don't confuse XML with its infoset,
let alone the PSVI. It is the infoset or PSVI or
canonical XML that are (paradoxically) closest to
S-exprs (if we accept properties as part of S-exprs).
    >>>

Let's not stop there. These things cannot be XML because XML is not 
extensible.

XML isn't extensible is because it desn't have any rules of 
expression evaluation, the way Lisp has for general purpose 
programming or WXS has for type inference. XML can only be lexed. 
For many people, that XML cannot be evaulated is seen to be a 
valuable characteristic; it makes XML a general purpose syntactic form.

Since XML is in used in many domains, we can expect attempts to have 
it optimized for those domains. Optimization always results in 
specialization. The problem with tuning XML to enable type 
derivation, query expression, namespace punctuation and so forth is 
that it takes the most general purpose syntactic form in widespread 
use and attempts to specialize it for a particular problem domain by 
baking in a semantics (rules of evaluation). Take types, which seems 
to be the optimization of most concern. Over time, the worry is that 
XML ceases to be a purely syntactic form, but one that also has 
evaulation rules for type inference. Type inference mechanisms will 
be baked into XML processing, resulting in XML systems with low 
cohesion and high coupling, the sky will fall in, and so on.

    >>>
I think Simeon and Wadler's basic introductory spin is still wrong:

     * The property of self-describing as they
seem to use it (which I think is good), seems
depends on there being enough lexical forms
for each datatype. But by the time you add
dates and derived types, you would need to
extend basic S-expr syntax.
    >>>

The title of Simeon and Wadler's paper is as Dare said elsewhere, 
misleading. You have to wonder why they fail to make a distinction 
between typing the incriptions and typing the language inscribed.

Let's be clear about what XML extensibility is. When we say XML is 
extensible, we are saying encoding a domain specific language in XML 
allows that language to be extensible - inscribing a domain language 
with markup is more flexible than inscribing it with varchars for 
example. Thus we can say MathML /is/ extensible. But MathML is not 
an extension /of/ XML, it's a /use/ of XML, to encode another 
language. The structural flexibility afforded by XML encoding 
results in simplifying both syntactic and semantic extensions to MathML.

In the same way we can claim WXS is extensible (we can add new 
types), and that using XML as an encoding for WXS affords 
extensibility. But WXS is not an extension /of/ XML. Nor is it 
properly applied /to/ XML - in fact it can't be, which is why we 
apply WXS types over Infosets. Saying then that WXS is a type system 
for XML is nonsensical. WXS (or RNG) is used as a type system for 
domain language that happens to be encoded in XML.

The key characteristic of extensibility is that the evaluator
program for a language doesn't change with respect to domain
specific extensions of that language (little languages if you will). 
  Evaluators that can evaluate their own rules are called 
meta-circular. The most elegant realization of that idea continues 
to be Lisp. The most elegant realization of the idea that we shall 
not have any kind of evaluation whatsoever is XML. The ideas 
themselves are polar opposites.

Bill de hÓra