Re: [xml] XML Schema QNames

[Adam Dickmeiss <adam indexdata dk>]

Daniel Veillard wrote:

> On Fri, Mar 05, 2004 at 12:20:37AM +0100, Adam Dickmeiss wrote:
>  
>
> > I've seen what appears to be a bug in the handling of type names in XML
> > schema handler of libxml2. These type names are referred to as QNames in 
> > the XML
> > schema standard.
> >
> > See:
> > http://www.w3.org/TR/xmlschema-1/#section-Constraints-on-XML-Representations-of-Schemas
> >
> > The issue can be illustrated with the schema:
> >
> > <schema xmlns='http://www.w3.org/2001/XMLSchema'
> >  targetNamespace='http://my.ns/'>
> > <element name='my' type='string'/>
> > </schema>
> >
> > Here, the omitted prefix in 'string' is treated as NS http://my.ns/, 
> > when it should really be http://www.w3.org/2001/XMLSchema
> >    
>
>  Oh, okay.... my head did hurt each time I have tried to really understand
> from the spec the expected behaviour about namespace propagation and
> composition in the context of XML Schemas.
>  
I understand why that hurt.

I saw problem with mixed content and QNames due to the use of DTD to XSD 
converters. Then looked at literature and tested with Xerces as well.

>  
>
> > Attached is a patch.ns + 2 new tests which modifies 
> > xmlschemas.c:xmlGetQNameProp so that it returns default namespace if 
> > prefix is omitted for Qname. The patch also _removes_ 'fallback' to 
> > targetNamespace, which IMHO is not to be used.
> >    
>
>  okay,
>
>  
>
> > Removing this fallback made test po1_0.xsd fail. It turned out that this 
> > schema had two bad type references with omitted prefixes (when that was 
> > not allowed). Patch for po1_0.xsd is part of patch.ns.
> >    
>
>  Hum, I think I copied po1_0.xsd directly from the XML Schemas Primer,
> maybe I made a mistake doing that copy or I missed something else...
>  
You are referring to the primer on w3.org I assume.

Looking at the primer today, in section
"4.1 A Schema in Multiple Documents"
at least the reference to SKU/Items looks OK, i.e. it reads:

<element name="items"      type="ipo:Items"/>
...
<attribute name="partNum" type="ipo:SKU" use="required"/>

with the ipo: in place. So either the cut-n-paste went wrong or the primer was fixed. We don't know, and it 
doesn't matter :)


>  I applied the patch, maybe this will fix the problem that the xmlbench
> author reported about QName when running the benchmark suite.
> In general I very much welcome fixes you and others can provide w.r.t. XML
> Schemas implementation and cleanup. There is a large test suite a W3C but
> when I looked at it nearly 2 years ago libxml2 Schemas was really too
> incomplete to even start trying running against it. Maybe this would make
> more sense now, though there is still a parts which are not implemented
> like identity constraints.
>  
I won't promise to do more work for now. libxml2 WRT schemas does what I 
need now. Have never looked at the test suite from w3c.. It's tempting 
to see how things are now, admitted.


>  Thanks a lot, this was just commited in CVS,
>  
Thanks. Still can't see my first patch yet in anoncvs. I'll bug you, if 
it doesn't show up next week.

-- Adam

> Daniel
>
>