At 02:02 PM 3/5/2005, Bob Foster wrote:
>Andrew Jeavons wrote:
>>> From your description I'd guess that every verb/attribute combination
>>>and interaction has been hardcoded.  If so, you might get the best
>>>bang for your buck by "acquiring" (training/hiring) someone who can
>>>help you re-write your existing XSLT.
>>Exactly. Which is totally brain dead. That's what set me on my search for 
>>the BNF tool. Even with my minimal XSLT knowledge seeing the SAME code
>>over and over again made me very worried.....
>>The XSLT was done by external contractors, soon to be ex-contractors now 
>>if I can help it. I know that I can make a parser out of Bison and Flex 
>>with a fifth or less code
>>which will do far more than the XSLT does and be readable (well to me 
>>anyway....)
>
>That's what you think you know, but it isn't correct in general. BNF can 
>only do a surface syntax scan; you'll have to escape to a programming 
>language to do any sort of semantic check. XSLT is far more powerful. 
>That's not saying that the contractors' stylesheet actually uses any of 
>this power, but it's there.

Reasonable point. Yes, BNF provides a validity checking mechanism but I 
still have to write C/C++ or whatever to do anything concrete with the 
elements I parse. And C et al
are not great languages for these kinds of job it is true.

>Plus, you seem to be missing this major point: Before you can teach your 
>BNF to parse your XML application language, you first have to teach it to 
>parse XML. This is a much bigger PITA than you imagine.

Yes and no. When I look at the XML language we have it is very (very) 
simple, which is why I was aghast it taking 2666 lines of *ANYTHING* (XSLT, 
C et al) to parse it. And also
looking at other systems I've written/worked on that use BNF parsers I'm 
pretty certain it isn't a huge job. 20+ years of hacking code around...:-) 
So while I accept your point
about XSLT being badly used in this case, I do know for an absolute fact I 
can write a shorter parser in Bison, but it does seem clear now that the 
problem isn't so much using
XSLT for the parsing as the way it has been done...

>  The virtue of using existing schema languages is that they already know 
> how to deal with the ins and outs of XML syntax, so you write grammars 
> and rules at a much higher level than BNF.

Very true. What wasn't clear to me was that systems like XSLT could write 
higher order grammars. This code was the first XSLT I had seriously looked 
at and it didn't give a good
impression !

>As a parser-aware person, you should find RELAX NG compact syntax easy to 
>understand and able to express what you need. If there is something you 
>can't say, you can add Schematron assertions to the same schema. I'd 
>strongly recommend you check it out.

Thanks, I'll check it out, I appreciate the comments/information.

andrew

>Bob Foster
>http://xmlbuddy.com/
>
>

Andrew Jeavons

www.andrewjeavons.com

+1 513 871 8278