re-reading the "least power" finding .... was RE: [xml-dev] 2007 Predict

["Michael Champion" <mc@xegesis.org>]

> -----Original Message-----
> From: Michael Champion [mailto:mc@xegesis.org]
> Sent: Saturday, January 20, 2007 7:18 AM
> To: 'noah_mendelsohn@us.ibm.com'
> Cc: 'XML Developers List'
> Subject: RE: [xml-dev] 2007 Predictions
> 


> This finding reminds me of  the emperor in "Amadeus" telling Mozart that
he
> used too many notes.  If the Web really did follow the W3C's lead, the
current
> state-of-the-art web applications would never have been invented. These
tend
> to download semi-opaque blobs of Javascript ...

After re-reading the final, edited-by-Noah version of the Rule of Least
Power, I don't have quite as negative a reaction as I did to my memory of
TimBL's "axioms" and the TAG list discussion a year ago.  If this is mainly
a warning that putting executable code on the Web is dangerous, it's hard to
disagree, but of course there are benefits that outweigh the potential
dangers for most people. After all, we all risk horrible death in
automobiles every day because their power and convenience outweighs their
dangers.  Also, it's clear from a re-reading that they're talking about
languages used to publish information, not the complexity of a service
implementation behind the Web.  BUT it seems worth noting that
first-generation XML standards widely deployed have similar problems.  Maybe
I would have had a less negative reaction to the finding had these problems
been called out along with those of AJAX.

First, XML is in some sense "too powerful" even though it is not Turing
complete. The recursive entity definition mechanism allows documents to be
created that require an exponential time order of magnitude to parse (the
"billion laughs" attack).   Also, XSLT *is* Turing complete, as the finding
notes.  In principle an XML parser or XSLT engine could do the analysis that
the finding suggests, but AFAIK actual implementations today just run the
program and see what happens, just as with Javascript. My very limited
discussions with smarter people than I am about adding logic to an XML
parser that would detect DoS attacks while maintaining full conformance with
the spec indicate that this would make it unacceptably slow.

The practical advice is not "use the language with the least power", since
sufficient power to anything interesting on today's Web is sufficient power
to do evil.   Hmm, I could go on but Len just  said it so much better than I
possibly could :-)