FW: [RESEND] Derivation by restriction

One of my colleagues has challenged my interpretation of the spec.  
He claims to have found evidence that *any* valid member of a substitution
group can validly be the content of a type derived by restriction containing
the SG head.  

He appears to have a point.  
The issue is probably "when" the substitution group head is replaced by its
equivalent choice group, and when and how many of the members of the
equivalent choice group are considered.  
Given that members may be added to the substitution group in external
schemas, then this could have quite an impact on a validating service.  

My head is now spinning on this.  
Can any of you guys tease it out?

-----Original Message-----
From: Clemens Portele [mailto:portele@interactive-instruments.de]
Sent: Wednesday, 3 April 2002 3:14 PM
To: Simon.Cox@csiro.au
Subject: RE: [RESEND] Derivation by restriction


Simon wrote:

> My understanding is that this states that when deriving a type by
> restriction, all the particles in its content model must be 
> the same or
> derived by restriction.  

As I read the specs (and John H's remark) this is true only for local
elements but not for global ones; in this case a valid substitution
group is sufficient to create valid instances of the base. This means:
The best practice example on Roger Costello's page is valid, also if
BookType extends PublicationType - which will almost certainly be the
case. Do we agree on this? Probably not, since you write (or am I
missing something again):

> In GML we use a pattern where in the content model of a 
> derived type, we
> restrict a substitution group (i.e. an implicit choice group) 
> to a specified
> member.  This is fine when the type of the member is a 
> restriction of the
> type of the head.  But fails if the type of the member in the 
> content model
> of the derived type is an extension of the type of the head 
> element in the
> content model of the base type.  
> 
> Note that it is OK for a substitution group member to have a 
> type derived by
> extension from the type of the head element.  
> But the constraints are on how you can use it within content models.  

Let me repeat my analysis from some days ago. If you are correct, where
is my mistake?

<xsd:element name="E1" type="X1"/>
<xsd:complexType name="B">
	<xsd:sequence>
		<xsd:element ref="E1"/>
	</xsd:sequence>
</xsd:complexType>

<xsd:element name="E2" type="X2" substitutionGroup="E1"/>
<xsd:complexType name="R">
	<xsd:complexContent>
	<xsd:restriction base="B">
	<xsd:sequence>
		<xsd:element ref="E2"/>
	</xsd:sequence>
	</xsd:restriction>
	</xsd:complexContent>
</xsd:complexType>

Let's assume that the element definitions of E1 and E2 are valid and X2
is derived from X1 by extension.

Then the question is: Is "R" valid?

XML Schema Part 1, 3.4.6, Rule "Schema Component Constraint: Derivation
Valid (Restriction, Complex)", subitem 5.3 says:

"If the {content type} of the {base type definition} is mixed or the
{content type} of the complex type definition itself is element-only,
then the particle of the complex type definition itself must be a ·valid
restriction· of the particle of the {content type} of the {base type
definition} as defined in Particle Valid (Restriction) (§3.9.6)."

I read this as 

  <xsd:element ref="E2"/>

must be a valid "restriction" of 

  <xsd:element ref="E1"/>.

In 3.9.6, Rule "Schema Component Constraint: Particle Valid
(Restriction)" the relevant subitem seems to be 2.1: 

"Any top-level element declaration particle (in R or B) which is the
{substitution group affiliation} of one or more other element
declarations is treated as if it were a choice group whose {min occurs}
and {max occurs} are those of the particle, and whose {particles}
consists of one particle with {min occurs} and {max occurs} of 1 for the
top-level element declaration and for each of the declarations in its
·substitution group·." 

That is:

<xsd:element ref="E1"/>

is treated as

<xsd:choice>
	<xsd:element ref="E1"/>
	<xsd:element ref="E2"/>
</xsd:choice>

and we get:

<xsd:element name="E1" type="X1"/>
<xsd:complexType name="B">
	<xsd:sequence>
		<xsd:choice>
			<xsd:element ref="E1"/>
			<xsd:element ref="E2"/>
		</xsd:choice>
	</xsd:sequence>
</xsd:complexType>

<xsd:element name="E2" type="X2"/>
<xsd:complexType name="R">
	<xsd:complexContent>
	<xsd:restriction base="B">
	<xsd:sequence>
		<xsd:element ref="E2"/>
	</xsd:sequence>
	</xsd:restriction>
	</xsd:complexContent>
</xsd:complexType>

Then the rule to be applied seems to be "Schema Component Constraint:
Particle Derivation OK (Elt:All/Choice/Sequence -- RecurseAsIfGroup)":

"For an element declaration particle to be a ·valid restriction· of a
group particle (all, choice or sequence) a group particle of the variety
corresponding to B's, with {min occurs} and {max occurs} of 1 and with
{particles} consisting of a single particle the same as the element
declaration must be a ·valid restriction· of the group as defined by
Particle Derivation OK (All:All,Sequence:Sequence -- Recurse) (§3.9.6),
Particle Derivation OK (Choice:Choice -- RecurseLax) (§3.9.6) or
Particle Derivation OK (All:All,Sequence:Sequence -- Recurse) (§3.9.6),
depending on whether the group is all, choice or sequence." 

I.e. this transforms our schema to:

<xsd:element name="E1" type="X1"/>
<xsd:complexType name="B">
	<xsd:sequence>
		<xsd:choice>
			<xsd:element ref="E1"/>
			<xsd:element ref="E2"/>
		</xsd:choice>
	</xsd:sequence>
</xsd:complexType>

<xsd:element name="E2" type="X2"/>
<xsd:complexType name="R">
	<xsd:complexContent>
	<xsd:restriction base="B">
	<xsd:sequence>
		<xsd:choice>
			<xsd:element ref="E2"/>
		</xsd:choice>
	</xsd:sequence>
	</xsd:restriction>
	</xsd:complexContent>
</xsd:complexType>

Then the rule to be applied according to the last rule is "Schema
Component Constraint: Particle Derivation OK (Choice:Choice --
RecurseLax)":

"For a choice group particle to be a ·valid restriction· of another
choice group particle all of the following must be true: 
1 R's occurrence range is a valid restriction of B's occurrence range as
defined by Occurrence Range OK (§3.9.6); 
2 There is a complete ·order-preserving· functional mapping from the
particles in the {particles} of R to the particles in the {particles} of
B such that each particle in the {particles} of R is a ·valid
restriction· of the particle in the {particles} of B it maps to as
defined by Particle Valid (Restriction) (§3.9.6). 
NOTE: Although the ·validation· semantics of a choice group does not
depend on the order of its particles, derived choice groups are required
to match the order of their base in order to simplify checking that the
derivation is OK."

Here, the definition of "order-preserving" is important:

"[Definition:]  A complete functional mapping is order-preserving if
each particle r in the domain R maps to a particle b in the range B
which follows (not necessarily immediately) the particle in the range B
mapped to by the predecessor of r, if any, where "predecessor" and
"follows" are defined with respect to the order of the lists which
constitute R and B." 

And since <xsd:element ref="E2"/> is a valid restriction of itself and
the order-preservation rule is fulfilled, the type definition of "R" is
valid (which was your point).

Received on Wednesday, 3 April 2002 02:38:11 UTC