RE: [xsl] Normalize / Simplify HTML-Tables with row-span / col-span

Subject: RE: [xsl] Normalize / Simplify HTML-Tables with row-span / col-span
From: David Tolpin <dvd@xxxxxxxxxxxxxx>
Date: Wed, 18 Feb 2004 22:33:57 +0400 (AMT)
> bugs aren't found for a long time. It's not quite true that optimization
> is the root of all evil, but it's certainly something you have to
> balance against reliability.

Yes. And my opinion is that eliminating node-set through hiding
different things inside the implementation core is a risk to lose
reliability. I have no doubts that it can be done right.

I am sure you can do this thing right in your implementation; as
well as many other things, since except for my (and some others')
perversive and overstressed use of XSLT, Saxon is the winner.

The problem is not whether a bright and talented engineer can do
it right, it is whether there are any warranties that it will be
done right. Or are we at the mercy of good will and common sense
of implementors. Either the specification must demand that certain
conditions hold (Haskell). Or it should refrain from introducing
things  which can become a hidden danger (Pascal).

Otherwise, the only portability that can be achieved is the portability
of the implementation, not of the language. This is the way of TeX
(the typesetting system) and Perl (the programming/scripting
language). I am sure that Saxon 7 will be an excellent implementation
of XSLT 2.0. Seriously. Honestly. I do not see a mechanism to ensure
that my XSLT program will exhibit consistent behaviour among different
implementations of XSLT 2.0. 

If the name of the language is SAXON, and Saxon 7 is the reference
implementation, I am gladly accepting it, because it is a good tool 
(the implementation) and a usable formalism (the language). But I 
fear that if the language is XSLT 2.0, and the expectation is to
have many implementation with the spec as the only thing they share,
it will not be possible to write a serious XSLT 2.0 program in
such a way that it works with consistent results and performance
with all (or most, or several) implementations without testing
it with all of them.

Java, in its core, is a smaller and simpler language than XSLT 2.0.
Yet there is only one conformant implementation of JVM for Java 1.2;
all others, both coming from opensource community and from giants
of the industry fail to run complex java programs -- I know it
from sad experience of being an author of a commercial Java program
of moderate size which we had to adapt to bugs of IBM JVM.

I am afraid that in order to ensure that there will be any interoperability
among implementation, many things must be either added to the specification
or removed from it. 

David Tolpin

 XSL-List info and archive:

Current Thread