Detecting newly introduced namespace attributes at individual element level?

[drott]

Hi, big thanks for maintaining this useful, ergonomic library.

For a specific StartElement event in the tree, I would like to find out if this element introduces any new namespaces.

<library xmlns:book="http://www.example.com/books">
    <book:title>The Great Novel</book:title>
    <book:author>John Doe</book:author>
    
    <review xmlns:feedback="http://www.example.com/feedback"> <feedback:rating>5 stars</feedback:rating>
        <feedback:comment>Excellent read.</feedback:comment>
     </review>
    
    <book:isbn>1234567890</book:isbn>
</library>

When I get the StartElement event for <review> how can I tell that xmlns:feedback is new here in the namespaces stack? Do I need to keep track of the previous BTreeMaps from the previous StartElement events and compare them to see which is new? Or is there a way I can tell which namespace was most recently introduced?

My understanding is that the namespace field of the StartElement event aggregates all namespaces encountered so far and reports those for every child element, independent of whether they are added on the current element or whether they were accumulated in parent elements.

Background: In a parser that converts the document to an internal structure, for example a DOM structure, without this sort of functionality I think I would otherwise be unable to reproduce the original document structure. Because I cannot tell directly from the API where a namespace was introduced, or can I?

[kornelski]

Yeah, some kind of explicit information about this could be useful.

The current design reflects the philosophy that XML wants to have that only the namespace URI matters, nothing else. The prefixes are meaningless. The xmlns attributes are just a serialization detail, a form of compression that reduces repetition, and you shouldn't care about them any more than you care about exact settings of gzip used to compress a document.

If you only keep namespace URIs, then xmlns and prefixes can be re-added in any way later, even on different elements with different prefix names, and the document will be semantically identical as long as individual elements map to the same URIs.

[drott]

If you only keep namespace URIs, then xmlns and prefixes can be re-added in any way later, even on different elements with different prefix names, and the document will be semantically identical as long as individual elements map to the same URIs.

I agree on semantic identity, but without information from parser events which prefixes / namespaces are introduced on which element - the resulting internal representation can not be used to serialize back to produce a document that is source-identical.

I'll experiment locally with diff'ing the BTreeMaps for now and I'll see if after that I have a proposal on what we can do here.