One of the biggest assumptions I had about software development was shattered when I started working on the XML team at Microsoft. This assumption was that standards bodies know what they are doing and produce specifications that are indisputable. However I’ve come to realize that the problems of design by committee affects illustrious names such as the W3C and IETF just like everyone else. These problems become even more pernicious when trying to combine technologies defined in multiple specifications to produce a coherent end to end application.
An example of the problem caused by contradictions in core specifications of the World Wide Web is summarized in Mark Pilgrim’s article, XML on the Web Has Failed. The issue raised in his article is that determining the encoding to use when processing an XML document retrieved off the Web via HTTP, such as an RSS feed, is defined in at least three specifications which contradict each other somewhat; XML 1.0, HTTP 1.0/1.1 and RFC 3023. The bottom line being that most XML processors including those produced by Microsoft ignore one or more of these specifications. In fact, if applications suddenly started following all these specifications to the letter a large number of the XML documents on the Web would be considered invalid. In Mark Pilgrim’s article, 40% of 5,000 RSS feeds chosen at random would be considered invalid even though they’d work in almost all RSS aggregators and be considered well-formed by most XML parsers including the System.Xml.XmlTextReader class in the .NET Framework and MSXML.
The newest example of XML specifications that should work together instead of becoming a case of putting square pegs in round holes is Daniel Cazzulino’s article, W3C XML Schema and XInclude: impossible to use together??? which points out
The problem stems from the fact that XInclude (as per the spec) adds the xml:base attribute to included elements to signal their origin, and the same can potentially happen with xml:lang. Now, the W3C XML Schema spec says:
3 For each attribute information item in the element information item’s [attributes] excepting those whose [namespace name] is identical to
http://www.w3.org/2001/XMLSchema-instanceand whose [local name] is one of
noNamespaceSchemaLocation, the appropriate case among the following must be true:
And then goes on to detailing that everything else needs to be declared explicitly in your schema, including xml:lang and xml:base, therefore :S:S:S.
So, either you modify all your schemas to that each and every element includes those attributes (either by inheriting from a base type or using an attribute group reference), or you validation is bound to fail if someone decides to include something. Note that even if you could modify all your schemas, sometimes it means you will also have to modify the semantics of it, as a simple-typed element which you may have (with the type inheriting from xs:string for example), now has to become a complex type with simple content model only to accomodate the attributes. Ouch!!! And what’s worse, if you’re generating your API from the schema using tools such as xsd.exe or the much better XsdCodeGen custom tool, the new API will look very different, and you may have to make substancial changes to your application code.
This is an important issue that should be solved in .NET v2, or XInclude will be condemned to poor adoption in .NET. I don’t know how other platforms will solve the W3C inconsistency, but I’ve logged this as a bug and I’m proposing that a property is added to the XmlReaderSettings class to specify that XML Core attributes should be ignored for validation, such as XmlReaderSettings.IgnoreXmlCoreAttributes = true. Note that there are a lot of Ignore* properties already so it would be quite natural.
I believe this is a significant bug in W3C XML Schema that it requires schema authors to declare up front in their schemas where xml:lang, xml:base or xml:base may occur in their documents. Since I used to be the program manager for XML Schema technologies in the .NET Framework this issue would have fallen on my plate. I spoke to Dave Remy who toke over my old responsibilities and he’s posted his opinion about the issue in his post XML Attributes and XML Schema . Based on the discussion in the comments to his post it seems the members of my old team are torn on whether to go with a flag or try to push an errata through the W3C. My opinion is that they should do both. Pushing an errata through the W3C is a time consuming process and in the meantime using XInclude in combination with XML Schema is signficantly crippled on the .NET Framework (or on any other platform that supports both technologies). Sometimes you have to do the right thing for customers instead of being ruled by the letter of standards organizations especially when it is clear they have made a mistake.