Norm Walsh writes about a very nasty problem waiting to happen when anybody naive enough is trying to validate an XML document after XInclude processing against XML Schema. Ooops, XInclude processor appends xml:base attributes at top-level included elements, which you gotta allow in the schema, most likely on every element! And XML Schema of course doesn't support declaring of such globally allowed attributes, so you have to add its definition to every single element declaration. And that's not all! Norm forgot about xml:lang attribute, which XInclude processor can and should append too. I mentioned that issue in the "Combining XML Documents with XInclude" article.
One interesting point Norm raised is how come this glitch happened:
I think what pains me most about this situation is that XInclude was in development for just over five years. It went through eleven drafts[1] including three Candidate Recommendations.Basically I disagree. Of course this incompatibility wasn't unnoticed. Moreover you can find the following in the "DocBook Technical Committee Meeting Minutes: 19 Nov 2002":
Why didn't we notice this until several months after XInclude was a Recommendation?
615587 Support xml:baseYou get the idea of what is the common solution - just declare xml:base and xml:lang in your schema.
Any object to adding xml:base to the common attributes? None.
Accepted.
And what about better solutions? Norm talks about fixing either XML Schema or XInclude. I don't see why XInclude needs to be fixed here. It's clear that if a fragment comes from a different place its base URI must be preserved somehow, otherwise say good bye to relative URIs in included fragment. And xml:base is currently the only feasible facility of base URI manipulation. But when it comes to XML Schema I see plenty of room for improvements - it could either allow to define globally allowed attributes or to allow xml: attributes to appear anywhere, just like xsi:type at al.
What's wrong with the latter solution? xml:base, xml:lang, xml:space, xml:id - isn't it ridiculous to be forced to declare them on every element in XML Schema? They are orthogonal to validation and so XML Schema validation should be orthogonal to them.
Another interesting tidbit - this issue was reported to the MSDN Product Feedback Center by kzu and his suggestion to introduce a flag to the XmlSchemaValidationFlags enum saying that xml: attributes should be ignored during validation seems to be favorable by Microsoft. There is a chance that in .NET 2.0 XML Schema processor will optionally allow xml:base and friends even if not declared in schema. If you like it - go vote for this suggestion. I did.
It's time to update and say that finally the flag was added to .NET 2.0 and now XInclude can be used with XSD validation in .NET!!! :D
basically this seems to me to be related to xml schemas not taking into account a common extensibility methodology of xml, that being the ignoring of unknown namespaces, currently to get this functionality you need to use any at each level, you can not use it in a reasonable way with sequences, and if you do any namespace = other your extensibility model is going to be screwed up if you try to reuse this schema in a schema with another target namespace. this is one of the main and most irritating ways in which xml schema is broken, irritating in that this way of working with disparate namespaces was a very widespread strategy before xml schema was finalized.