Re: [xml] redicting parts of trees

[Kasimier Buchcik <kbuchcik 4commerce de>]

Hi,

On Mon, 2005-06-27 at 10:44 -0400, Rob Richards wrote:
> Kasimier Buchcik wrote:
>
> > The context has just an initial "void * _private" field, to be able to
> > extend it. Adding fields to it, at the end, won't break ABI
> > compatibility.
> >  
> How do you not break compatibility? For example, A new field is added, 
> which if not NULL would cause the execution of some function while 
> inside the adopt node function. If someone where to create a context 
> structure, without using some initialization function, then a crash most 
> likely would occur as the field is not initialized to NULL.
>
> Like the ctxt->aquireNsDecl(ctxt, cur->ns, &ns) function. When that 
> actually gets added, how do I make sure it doesnt execute?
> Will there be a flags field (some int)  that indicates the functions off 
> the context to execute or will it end up testing:
>
> if (ctxt->flags & DOM_CTXT_AQUIRE_NS) {
>     ctxt->aquireNsDecl(ctxt, cur->ns, &ns)
> }
>
> or
>
> if (ctxt->aquireNsDecl) {
>     ctxt->aquireNsDecl(ctxt, cur->ns, &ns)
> }

I was thinking of the latter version.

> To avoid breaking stuff, shouldn't an additional function be added which 
> returns an initialized context with all fields set to null and be the 
> required way to grab the context or run the risk of crashing in future 
> releases? This way when a context is needed, they would call the init 
> function and then populate their context? As it would be bad if someone 
> used the context structure directly not realizing new fields have been 
> added and need to be initialized.

Yes, a factory function, e.g. xmlDOMWrapNewCtxt(), which would memset
the structure to 0, should be added.

Regards,

Kasimier