Issue for XMI for MOF 2

The specification of XMI for MOF 2 does not describe handling of MOF's
object type, which is a supertype of ordinary class types as well as data
types. Particularly, it does not describe what XML results from the
following small example. The whole model is this:

In a package M there is one class X having one owned attribute named Y of
type TypeLibrary::Objects::Object with multiplicity 1..1.

Consider the following 4 instances of X:

1. An instance of X having Y set to the second instance of X.
2. An instance of X having Y set to the integer 3.
3. An instance of X having Y set to the string "this UNICODE string".
4. An instance of X having Y set to that same fourth instance of X.

What would an XMI-based serialization of these four instances of X look
like?

What would an XML schema for the package M look like?

The XMI for MOF 2 Specification seems to have been written as if referring to a MOF 1 Specification. For example, it defines mappings and mapping parameters for AssociationEnd, which is not in MOF 2.

Recommended Resolution: Restate the XMI rules and mapping parameters in terms of MOF 2 and UML 2.

The XMI for MOF 2 specification has not been updated to reflect one of the key goals in the alignment of MOF 2 with UML 2, which is that XMI becomes applicable to UML class models in general rather than to MOF-based metamodels only.

Recommended Resolution: Replace "metamodel" with "model" in most places within the XMI for MOF 2 specification. Also, in describing the mapping from classes, properties, etc. to XML, do not use wording that implies a restriction to a MOF context except where MOF is , such as for elements of the MOF Type Library.

The XMI for MOF 2 specification defines several tagged values for adding
data to a UML model that guide generation of an XML Schema and that guide
document production. But these tags appear to be defined as if for MOF 1.
To be appropriate for MOF 2 these must be defined using either an extension
to the MOF or UML core models (as a package that adds to MOF and/or UML
infrastructure) or they must be defined as a UML 2 profile. In either case,
an official XML rendering of the model or profile is required. And given
the big change from MOF 1, an example of such tags as they would appear in
an XMI-based XML document of an example model would be very helpful.

Remarks to chapter "3.4.3 Derived Types and References" in XMI Spec. , v 2.0, S. 3-15
There it says:

"... An instance of Company can use xsi:type to
indicate that its HQAddress is actually of type USAddress and includes a zipcode:
<Company xmi:id="Company_1" name="Acme">
<HQAddress xmi:type="USAddress" xmi:id="Address_1"
Street="Side Street" City="Hometown" zipcode="90210"
... "
The 'HQAdress'-Element has a 'xmi:type'-Attribute. Parsing this with Xerces 2.4.0 gives problems, since a 'xsi:type' Attribute is required. Derived types must be specified with the 'xsi:type'-Attibute. Maybe the line
"... An instance of Company CAN use xsi:type ..."
should be changed in
" ... An instance of Company MUST use xsi:type ..."

maybe this is just because of my ignorance, but I'm running into
a problem trying to validate documents against an XMI based schema
here, and it looks like the validator has a point.

As usual, my base document is XMI 2.0, formal/03-05-02.

Section 1.10.2 "Linking" describes the usage of XLink simple links
and XPointers to link across documents.

However, looking at the schema production rules (page 2-2 to 2-5,
rules 4g:ClassAttListItems and 1g:XMIFixedAttribs, as well as the
LinkAttribs attributeGroup in the Fixed Schema Declarations on
page 2-10, the xlink:href is never declared as a legal attribute
for complexTypes.

Hence I see why the parser (Xerces-C, when requested to validate)
complains about attempts to use XLinks in a document.

I think it is paramount that documents validate against the schema.
(Am I the first to attempt validation? Can't be.)

I realize that there doesn't seem to be a schema for XLinks yet,
so the short-term solution might be to allow the "normal" href
attribute to contain an XPointer.

As a minor nitpicking, page 1-23 gives the XLink namespace as
"http://www.w3.org/1999/XLink", whereas the W3C recommendation
for XLink uses "http://www.w3.org/1999/xlink" (note case).

Doc references are based on formal/03-05-02.

In a Schema that is produced by the XML Schema Production
rules, enforcement of multiplicities is incompatible with
the use of proxies.

According to 1.10.1, page 1-21, "Elements act as a union,
where they are either a definition or a proxy. Proxies use
the LinkAttribs attribute group to define the link, and
contain no nested elements."

However, if the org.omg.xmi.enforceMinimumMultiplicity is
true, then the generated schema does not allow the complex-
Type's content to be empty, and so proxies cannot exist in
an XML document.

As a simple fix, the 4b:ClassContents could be wrapped into
an additional choice element, as in

<xsd:choice minOccurs="0">
<xsd:sequence>
4b:ClassContents
</xsd:sequence>
</xsd:choice>

This would allow the element to be empty. However, in this
form, an XML document that contained both link attributes
and contents would still validate.

A stronger solution would be to make a choice between a
link element and contents, by removing the LinkAttribs from
the ObjectAttribs attribute group and by, in rule set 4,
defining something along the lines of

<xsd:choice>
<xsd:sequence>
4b:ClassContents
</xsd:sequence>
<xsd:element name="href" type="xsd:string">
<xsd:element name="idref" type="IDREF">
</xsd:choice>

This way, an element in a validating document could only be
either a proxy or not.

As a side note to this issue, there are obviously unpleasant
side effects when org.omg.xmi.enforceMinimumMultiplicity is
true and org.omg.xmi.element is false. org.omg.xmi.enforce-
MinimumMultiplicity=true should have the same effect as
if org.omg.xmi.element=true and org.omg.xmi.attribute=false.

Section 1.10.2.2 (page 1-23) gives the URL for the xmi namespace
as http://schema.omg.org/spec/XMI/2.0

Section 2.2.1 (page 2-2) uses http://www.omg.org/XMI instead

Document refs are against formal/03-05-02.

According to rule 4d of the Schema production rules on page 2-5,
"The type is 'xsd:string' for simple attributes".

I assume here that "simple attributes" refers to MOF primitive
types (Boolean, Integer, Long, Float, Double, String).

I wonder why xsd:string was used for all these primitive types,
when perfectly suitable data types are predefined for XML Schemas:
there's xsd:boolean, xsd:int, xsd:long, xsd:float and xsd:double
that exactly match the data types used by MOF.

I propose to make use of these existing types to strengthen the
generated schema.

Document refs are against formal/03-05-02.

Maybe it's because I'm still struggling with the details
of XML Schemas and MOF, but I'm confused with rule 4d of
the XML Schema Production rules (EBNF on page 2-4, expla-
nation on page 2-5).

• According to the EBNF, a type='xmi:Any' is an output
  option, but according to the explanation, this option
  is never used.

• The explanation goes, "The type is 'xsd:string' for simple
  attributes ..." It is not clear to me what constitutes a
  "simple" vs. "non-simple" attributes. I assume this refers
  to MOFs PrimitiveTypes, but in any case a clarification
  would be useful.

• The explanation goes on, "The type is [...] part of the
  of the value of the org.omg.xmi.schemaType tag". Why does
  it say "part of"? What should be removed from the value?

• There is also a question of precedence. I guess the
  intention is that the schemaType tag should take
  precedence over the "simple attribute" decision?

• There is no "else:" What if the attribute is not simple,
  not an enumeration, and the schemaType tag is not set?
  (Or maybe that is not possible in MOF?)

The Schema production rule for 'ClassReferences' (4.e) does not conform with the Document Production Rule 'Reference Element' (4). In the Class References the linking attribues are missing