Re: GSXML

[Richard Frith-Macdonald]

Nicola Pero wrote:

> > Reports of all such inconsistencies welcome ... the classes were 
> > originally written
> > by someone unfamiliar with OpenStep, and while Nicola and I have put 
> > some effort
> > into making them more consistent, the job is not complete.
>
> Since you quote me, I need to correct you - as I already said in public
> posts, I think it's not just a problem of consistency or of details ...
> the GSXML classes and their API are basically flawed and should be totally
> rewritten.
I think that's a bit misleading taken out of the context of the earlier 
discussion.

The basic flaw in the GSXML classes is that they are based on libxml, 
and suffer
the memory management problems of libxml.  This basically manifests if 
you want to
use them to create XML, not for parsing.  The 'quick' fix is to change 
the GSXML API
to better reflect the limitations of the underlying libxml model, and 
not to imply
that elements may be freely created and moved around.  This is what I'd 
like to see
done (opr might even do myself) in the short term.

The 'proper' fix is to rewrite from scratch, so that elements can be 
freely created, added
to documents, removed from documents, and moved around in documents.

The disadvantage of doing a proper fix is that it's time consuming ... 
the libxml
implementation has two advantages when parsing ... speed, and validation 
against DTDs
(the latter is quite tricky, and not matched by any other implementation 
I've seen).