Re: [Libxmlplusplus-general] Status

[murrayc t-online de (Murray Cumming)]

On Thu, 2003-02-06 at 15:50, Stefan Seefeld wrote:
> Murray Cumming wrote:
> 
> >>My point was that, while lacking a lot of features, libxml++ before my
> >>initial patch was consistent in the way it worked. My patch changed the 
> >>perspective of what an xmlpp::Node actually is (for example),
> > 
> > 
> > No patch should introduce regressions. 
> > 
> > Could you explain more? I thought you just changed the implementation
> > without changing the API?
> 
> yes and no. The API didn't change, but semantics did. With my patch
> the document (a libxml2 antity) is the owner of all xmlpp::Nodes (incl.
> Attributes), so we must not allow anybody to create any of these objects
> directly.

We had already stated (not clearly) that no-one should create Nodes or
Attributes directly. I would not expect the API to work if someone did.

> The Node constructor/destructors are called by libxml2 when the 
> corresponding xmlNode, xmlAttr, etc. objects are created/destroyed,
> meaning the implementation (xmlNode) owns its own wrapper (xmlpp::Node).

Is that an additional point? It seems like implementation to me?

> >>all we need to do is branch from the version just before the patch, and 
> >>declare that branch to be the 'stable 1.0' branch. No need to reapply
> >>anything.
> > 
> > 
> > I liked your patch.
> 
> Ok. Then please consider the SaxParser API change,

OK. I will look at it.

>  and possibly others
> that make public construction of xmlpp::Nodes (, Attributes, ...)
> impossible.

Yes, that would be good.

-- 
Murray Cumming
murray usa net
www.murrayc.com