Lists Home |
Date Index |
John Cowan wrote:
> A WXS schema imposes three kinds of constraints:
> 1) Schema-instance constraints: if an attribute is declared to have
> a certain (simple) type in the schema, then all values of that
> attribute in the instance must conform to the type.
> 2) Instance-instance constraints: if an attribute is declared to be
> unique, then no two values of that attribute in the instance may be
> equal (according to the definition of "equality" for the attribute's
> 3) Schema-schema constraints: if a complex type is said to be
> derived by restriction from another complex type, then it must
> redeclare the components of that type either unchanged or reduced in
> what may be called "strength".
> RNG does not have constraints of types 2 or 3. Therefore, it cannot
> specify complex type derivation by restriction. In addition,
> restricting simple types by facet is not a first-class operation, so
> once a built-in simple type has been restricted by facets, it can't
> be restricted further. This is a place where WXS is genuinely more
> orthogonal than RNG.
> (IMHO the way in which complex type derivation by restriction is
> done is bogus, because it generates fragile subclasses, a kind of
> bassackwards version of the well-known "fragile superclass" problem
> in OO.)
Couldn't <concur> be used to provide (very powerful and loosely
coupled) derivation by restriction if it was reintroduced to RELAX NG?
Along the lines of:
... base definition ...
<ref name="base" />
... restrictions ...