RE: [xml-dev] RE: RELAX NG and type derivation [Was:] XQuery and DTD/Schema?

["Derek Denny-Brown" <derekdb@microsoft.com>]

" Let us look at type assignment more closely. In traditional programming languages, type assignment is non-existent - here, type assignment for objects is done by the user. The user provides a set of type definitions, and whenever he creates an object, he explicitly declares the type to which the object belongs. However, it is different for XML Schemas."

I disagree with this.  C++ static struct instance initialization provides a limited form of type assignment. XML makes some things easier, since every sub-match is named, but then makes it all very complicated by with content-models.  I would argue that most people do not differentiate
	int foo;
From
	<int>foo</int>
(where there is the appropriate schema definitions)


My general opinion of subsumption based typing frameworks is that while the danger of allowing subsumptive-based subtyping, where there was not such intent is not worth the gain.  This is partly due to the fact that tag names tend to be based on language, and language allows multiple meanings of a given term, where the correct meaning is expected to be derived by context.  I strongly believe that subtyping must be explicit, specifically to remove the opportunity for unintended subtyping of trees.  Using a subsumption framework to validate explicit subtyping declarations is fine, but I am very wary of anything which attempts to derive intent that was not made explicit, when the penalty for incorrect derivation is high, as it is in this case.

Then again, I have little doubt that you can use a subsumptive framework to define a subtyping solution which is simpler and clearer than XSD. 


-derek

> -----Original Message-----
> From: Murali Mani [mailto:mani@CS.UCLA.EDU]
> Sent: Monday, July 08, 2002 10:24 AM
> To: Derek Denny-Brown
> Cc: Michael Fitzgerald; xml-dev@lists.xml.org
> Subject: Re: [xml-dev] RE: RELAX NG and type derivation [Was:] XQuery and
> DTD/Schema?
> 
> 
> Hi, I checked this message. I think my write-up is kind of preliminary,
> though itdiscusses exactly what Michael said. I think is is important to
> understand two different ways of doing subtyping --explicit subtyping and
> implicit subtyping. The issues in implicit subtyping are not fully solved.
> I will try to upgrade my report at my earliest, but I will encourage you
> to read the preliminary version, and please give any comments you may
> have.
> 
> you may please find the report at
> http://www.cs.ucla.edu/~mani/xmlresearch/datamodels/subtyping.doc
> (or)
> http://www.cs.ucla.edu/~mani/xmlresearch/datamodels/subtyping.pdf
> 
> I conjecture this will be an important topic to be discussed and probably
> solved in the next few months.
> 
> thanks and regards - murali.
> 
> On Mon, 8 Jul 2002, Derek Denny-Brown wrote:
> 
> > -----Original Message-----
> > >From: Michael Fitzgerald [mailto:mike@wyeast.net]
> > >
> > > RELAX NG can do type derivation by restriction, union, and list. In
> some
> > > ways, RELAX NG's list type derivation is more powerful that XML
> Schema's
> > > derivation.[1] All the facets that are available in XML Schema are
> available
> > > in implementations of RELAX NG. In what specific ways does RELAX NG
> fall
> > > short in type derivation?
> >
> > > As for named typing/subtyping in XML Schema, I think Murali Mani's
> analysis
> > > is worthy of consideration.[2] There he discusses implicit and
> explicit
> > > assignment, subsumption, and even (briefly) an instanceof method a la
> > > XQuery. RELAX NG does not offer named typing yet, but a proposal is on
> the
> > > table to provide it in 2.0.[3] As such it will most probably be
> presented in
> > > a module rather than added to the core specification, in keeping with
> RELAX
> > > NG's layered approach. I predict it will be a well-reasoned spec that
> won't
> > > later invoke pangs of regret and won't be unduly difficult to
> implement or
> > > use.
> >
> > I was not talking about simple-type (as WXS uses the term) derivation.Â
> I was talking about complex type derivation.  The ability to define a
> type 'book'
> > <xs:complexType name="book">
> > Â  <xs:sequence>
> > Â Â Â  <xs:element name="title" type="xs:string"/>
> > Â Â Â  <xs:element name="author" type="properName"/>
> > Â Â Â  <xs:any minOccurs="0" maxOccurs="unbounded" namespace="##any"/>
> > Â Â Â  <xs:element ref="chapter" maxOccurs="unbounded"/>
> > Â Â Â  <xs:any minOccurs="0" maxOccurs="unbounded" namespace="##any"/>
> > Â  </xs:sequence>
> > </xs:complexType>
> > Â
> > And then differentiate the sub-types via subtyping.
> > <xs:element name="referenceBook">
> > Â  <xs:complexType>
> > Â  Â  <xs:complexContent mixed="0">
> > Â Â Â Â Â  <xs:restriction base="book">
> > Â Â Â Â Â  Â  <xs:sequence>
> > Â  Â Â Â Â Â Â Â  <xs:element name="title" type="xs:string"/>
> > Â Â Â Â Â Â Â Â Â  <xs:element name="author" type="properName"/>
> > Â Â Â Â Â Â Â Â Â  <xs:element ref="chapter" maxOccurs="unbounded"/>
> > Â Â Â Â Â Â Â Â Â  <xs:element name="references" type="sourceReference"
> minOccurs="0" maxOccurs="unbounded"/>
> > Â Â Â Â Â Â Â Â </xs:sequence>
> > Â Â Â Â Â  </xs:restriction>
> > Â  Â  </xs:complexContent>
> > Â  </xs:complexType>
> > </xs:complexType>
> > Â
> > An application can now unambiguously query whether a given element is of
> type 'book'.  With RELAX-NG, the closest you can come is to ask if an
> instance matches 'book', but that in now way unambiguously determines
> whether it was intended to be considered a type of book or not.
> > Â
> > This ability to extend the type model of one schema is one of the core
> principles behind OOP, and as such is something that people want in a
> schema language.Â
> > Â
> > Also note, that I am not even remotely defending WXS as a complete,
> well-designed, nor elegant spec.  It just is, and it achieves certain
> things which people wanted.  I am simply arguing that it is not an utter
> failure, and that there exists nothing which could fully replace it
> currently.  There is no doubt in my mind that RELAX-NG is a more elegant
> creation, nor that it could be extended to fill the same role as WXS.Â
> The problem is that WXS is already out there, and a large number of
> vendors are already committed to it.  If that is not an issue for you,
> use RELAX-NG.  I would.
> > Â
> > -derek
> > Â
> >