[XML-DEV Mailing List Archive Home] [By Thread] [By Date] [Recent Entries] [Reply To This Message]

RE: Mixed content in data-binding (Was: Re: I nteres

• To: 'Bob Foster' <bob@o...>
• Subject: RE: Mixed content in data-binding (Was: Re: I nteresting pair of comments
• From: "Bullard, Claude L (Len)" <len.bullard@i...>
• Date: Fri, 15 Jul 2005 14:04:01 -0500
• Cc: 'Ed Day' <edday@o...>, Elliotte Harold <elharo@m...>, Michael Kay <mike@s...>, 'Peter Gerstbach' <peter.gerstbach@g...>, 'XML-dev' <xml-dev@l...>

Good catch, Bob, but I don't alter my position.  Build profiles 
and you require a lot of process rot to maintain them. 
Caveat emptor.

That mixed content annoys you doesn't mean it isn't a ubiquitously 
found means of notation.  That your model doesn't include 
them simply means it doesn't model the widest number of 
cases for the notation; it rewrites it into an alternative 
structure such as a typed node tree.  That's fine.  But 
as I've said before, markup IS a human interface.  Failing 
to notice that is also a reason for many permathreads.  My 
story just points out how old this permathread is.

History shows that the implementors will implement the wider 
notation and create a model that supports it if that is what 
the market they sell too exhibits in the data.   We have 
profiles in X3D.  The second largest one is the one that 
gets implemented most frequently because few want to 
buy self-limiting tools.  Where smaller subsets are implemented 
is typically because a tool only uses the language in 
a very limited context such as exporting.  Editors and 
browser implementors go for the largest practical set 
because they otherwise limit their market.

I also tried to get rid of mixed content in DTD designs. 
All it did was force the programmer to detect it in the code.
The problem boils down to where one detects it.
 
We've never been able to eliminate it from the human hand 
and in the case of automated systems, if we introduce them 
we also have to introduce conformance testing and signalling 
(eg, UDDI/COM/OLE) for the world's most unruly system.  As 
Liam Quin pointed out to me recently, failing to account for 
diversity was a major reason the SGML/Hytime/DSSSL systems 
failed to scale and the HTML systems did scale.

Microdiversity drives speciation.  Selection doesn't produce 
optimum populations or designs; it produces learning populations. 
So far we've learned that we can't arbitrarily eliminate mixed 
content.  Caveat vendor.

len


From: Bob Foster [mailto:bob@o...]

Bullard, Claude L (Len) wrote:
> Some years ago during the IADS work, our chief designer said that 
> any language (at that time, SGML) that enabled mixed content 
> should be eliminated from the universe.  Instead, the follow
> on, XML, thrived and won.  While I understand the parser-centric 
> data model position, history is not on the side of eliminating 
> mixed models.

There is no such thing as a mixed model. This is a notational issue.

<p>This is an <i>example</i> paragraph.</p>

actually represents a node structure like this:

{element p}
   {text}This is an {/text}
   {element i}
     {text}example{/text}
   {/element i}
   {text} paragraph.{/text}
{/element p}

That is the model; the former is a notational convenience. SGML is rife 
with notational shorthand designed to make documents more pleasing to 
human writers. XML eliminated some of it. The human interface oriented 
features it didn't eliminate - mixed notation, general entities, default 
attributes - remain to annoy us and periodically reignite permathreads.

XML would be improved if it had two dialects: the current cuddly 
language and a stripped down, possibly more verbose subset that did 
nothing but convey a structured node sequence as text.

That's the permathread this and most others here boil down to.

• Prev by Date: Re: Mixed content in data-binding (Was: Re: Interesting pair of comments
• Next by Date: namespace URI of xmlns prefix
• Previous by thread: Atom 1.0 is Baked
• Next by thread: namespace URI of xmlns prefix
• Index(es):
  • Date
  • Thread