Re: [xml] Potential wrong usage of xmlIsID() in tree.c

[Rob Richards <rrichards ctindustries net>]

Daniel Veillard wrote:
> > Daniel, from a non-DOM perspective, what's your view on adding auto 
> > ID-ness detection to functions like addchild, addsibling, etc... In the 
> >     
>
>   I'm undecided. On one side the fact of adding an attribute to a document
> may potentially make the parent an ID for that document. On the other side
> you have the XML Infoset spec which state that IDness is a property of the
> attribute (set at parse time or creation time)
>   http://www.w3.org/TR/xml-infoset/#infoitem.attribute
>   [attribute type] == ID
> and well when you copy this attribute you copy its infoset and hence its
> properties.
>   
See comments below.
> > patch I'm working on I removed my original code for those and only fixed 
> > up some other issues, but do you want the ID stuff added there as well 
> > for consistency with the newprop, setprop functions or should detection 
> > just be left in the state it is in now? If you do think it should go in, 
> > the last question I have is creating a xml:id attribute without a parent 
> > element. Right now it marks it as an ID, but I think xmlNewPropInternal 
> > should be changed to check for a parent element first.
> >     
>
>   To me creating a new attribute should definitely make the lookup and
> set the attribute type, but if you copy an existing one I think the existing
> type should be maintained, it may differ if you reserialize and reparse
> but that's a different document.
>   
I see the creation of the infoset (non-parse time) as falling under the 
synthetic infoset category. To me this encompasses node creation, 
unlinking, appending, etc..
Based on a DTD or the xml:id spec, an attribute must have an element to 
be considered an ID. As far as the copying goes, if a lone attribute is 
copied, the type should be modified as it breaks the element/attribute 
relationship for IDs. Once added back to a an element then the IDness 
would be updated appropriately.

The closest API the Information Set spec even mentions for manually 
building an infoset is the DOM API (though it does't limit what API can 
be used), which we have seen has completely different rules for handling 
building IDness. From the non-DOM perspective if IDness is going to be 
determined from checking the DTD when manually creating an attribute, 
then it should also take care of it when inserting attributes as well. 
Which leads to my perspective on unlinking and copying.

Rob