[lxml-dev] lxml 2.2 validation question

[Stefan Behnel]

Piet van Oostrum wrote:
>> jholg at gmx.de wrote:
>  >>> nsmap={None: 'http://example.com', 'foo': 'http://example.com'}
>  >>> rootElem = etree.Element('Foo', {}, nsmap)
>
>> Note that this does not put Foo into the http://example.com NS. It
>> creates an element Foo wit no namespace. The nsmap is rather a collection
>> of known namespace prefixes in the context of an element.
>
> But that means that the serialization:
>
> <Foo xmlns="http://example.com">
> Contents
> </Foo>
>
> that etree.tostring produces is wrong.

Correct. This is actually an error case that we could catch at the API
level: the new element has no namespace *and* the nsmap defines a default
namespace, i.e. this should fail:

    el = etree.Element('thetag', nsmap={None : 'uri:some-namespace'})

We'd then need to make sure that you can write

    el = etree.Element('{}thetag', nsmap={None : 'uri:some-namespace'})

which would result in something like

    <ns0:thetag xmlns="uri:some-namespace" xmlns:ns0="" />

Similarly, adding a namespace-less subelement to a tree that defines a
default namespace will not work 'as expected' (whatever a user may expect
in doing that). In this case, we'd have to cut the default namespace
definition by inserting a xmlns="" on the new element, so this case would
not be an error. The same applies in the general case where you create a
tree without namespaces and one that uses a default namespace, and then
insert the namespace-less tree into the other one at some place.

Another sick case:

    el1 = etree.fromstring(
       '<p:qualified xmlns:p="uri:myns"><nons/></p:qualified>')
    el2 = etree.Element("{uri:tns}thetag", nsmap={None: "uri:otherns"})
    el2.append(el1)

So it looks like we'd have to integrate something similar into the
namespace fixing mechanism...

Ugly, ugly...

Stefan