Re: XML tools for Octave

[Andy Adler]

On Thu, 29 Jun 2006, Bill Denney wrote:

> On Thu, 29 Jun 2006, Andy Adler wrote:
>
> > On Thu, 29 Jun 2006, Bill Denney wrote:
> >
> > > Andy Adler wrote:
> > > > I'm trying to write this. My idea is that XML like this
> > > > <a b="c" d="e"> text <f g="h"/> more text <f>data</f> </a>
> > >
> > > Shouldn't it parse as something more like
> > >
> > > v.a.ATTS.b = "c"
> > > v.a.ATTS.d = "e"
> > > v.a.CHILD{1} = " text "
> > > v.a.CHILD{2}.f.ATTS.g => "h"
> > > v.a.CHILD{3} = " more text  "
> > > v.a.CHILD{4}.f.CHILD{1}   => "data"
> >
> > My concern is that this output makes writing software to parse the xml 
> > output really frustrating - you need to loop through the CHILD vectors to 
> > find what you're looking for. This would result in people taking shortcuts 
> > that make the code fragile.
>
> But not doing it this way would make an incorrect representation for nested 
> structures: what about just parsing xhtml like
> <p>abcd <i>efg</i> jklm</p>
>
> would turn into
>
> v.p.TEXT = 'abcd  jklm';
> v.p.i.TEXT = 'efg';
>
> which would not reverse correctly because all of these could turn into the 
> above:
> <p>abcd<i>efg</i>  jklm</p>
> <p><i>efg</i>abcd  jklm</p>
> <p>abcd  jklm<i>efg</i></p>
> ...
>
> To me, it should be a reversible transformation.  Also, without keeping the 
> order, you would lose the ability to do full XSLT interpretation (what do you 
> do about sibling commands).
>
> > This is actually a big debate in the XML semantics community - the fact 
> > that XML does not map easily to data structures.
>
> I realize that it doesn't map easily, but it should map reversibly.

I honestly don't know how to address this. Your suggestion would 
basically mean that we expose the full DOM API. Matlab did this with
a thin wrapper over Java's DOM.

However, this is a really bad idea. It means that all the effort of
XML parsing falls on the user - who will either make mistakes or
shortcuts, with the result of really fragile code.

For example a user will parse
     <data><item> 1 </item></data>

using
   v.data.CHILD{1}.item.TEXT{1}

But this will break when you have
     <data> <item> 1 </item></data>

or
     <data><metadata/><item> 1 </item></data>


So I don't think that pushing all the complexity to the user is right.
Somehow it should be easy to do easy things, but possible to do 
correct things.

How about:
 <a b="c" d="e"> text <f g="h"/> more text <f>data</f> </a>

 v.a.ATTS.b      : "c"
 v.a.ATTS.d      : "e"
 v.a.TEXT{1}     : " text "
 v.a.TEXT{2}     : " more text "
 v.a.TEXT{3}     : " "
 v.a.f{1}.g      : "h"
 v.a.f{2}.TEXT{1}: "data"

and some extra information in:

 v.a.NAMESPACE
 v.a.ORDEREDELEMS
 v.a.UTFNAMES

This is starting to look like a big project ;-<


--
Andy Adler <address@hidden> 1(613)562-5800x6218