The title page of the document lists the following normative documents:

Normative Machine Consumable Files:
http://www.omg.org/spec/XMI/20131001/XMI.xsd
http://www.omg.org/spec/XMI/20131001/XMI-model.xmi

However, attempting to retrieve (for example) http://www.omg.org/spec/XMI/20131001/XMI.xsd returns HTTP redirect to https://www.omg.org/spec/XMI/20131001/XMI.xsd.
Attempt to retrieve https://www.omg.org/spec/XMI/20131001/XMI.xsd returns a 404.

Section B.2

• bullet 2 - smi:XMI should be xmi:XMI
• bullet 6 - smi:type should be xmi:type and smi:idref should be xmi:idref
  Section B.5.3
• second bullet - xmi:idreg should be xmi:idref

As suggested by Pete while reviewing the XMI files generated for SysML 1.5, the official XMI.xsd should have the processContent attribute set to "lax" instead of "strict". It would allow using it for validating official XMI files for specification related to UML (e.g. SysML) since UML has no XSD.

Table 9.4.1 states:

Choice of:
1. XMIObjectElement
2. Nested XMIReferenceElement
3. Nested XMIReferenceAttribute

Normally, serialized properties with isComposite = true are serialized as nested XMIObjectElements.
In the case where the model is split across more than one file then a nested XMIReferenceElement would be used. Exceptionally, even within one file, it may be the case that a containing object has more than one serialized class-typed property with isComposite = true that, contain the same object or include it among their collection of objects. In such an exceptional case, because of MOF contstraints, only one of those properties can have an opposite with a non-empty slot. Objects of the property with the non-empty opposite slot are serialized as nested XMIObjectElements, and the other references to the same object are serialized either as XMIReferenceAttributes or nested XMIReferenceElements.

This rule is very bad for several reasons:

1) The criteria for determining which of the 3 possible options is very difficult to understand

2) Words such as "Exceptionally" convey a sense that there is a distinction that should be seldom relevant in practice. This is definitely not the case. This rule affects all serializations of UML profiles and UML models that contain Structured Activities.

3) Whether a model is serialized in one or multiple files has nothing to do with the intent of this rule. Mentioning this adds an unnecessary layer of complexity to the rule.

4) The criteria for choosing between nested vs. reference/attribute serialization refers to unexplained considerations about MOF:

• no explanation in the XMI spec
• no cross-references to existing explanations in MOF specifications (it is even unclear which MOF specifications, if any, have such explanations)

5) The combination of no illustrative example and of exceptional opacity has contributed to legitimate doubts as to whether there exists any truly conforming implementation of XMI 2.x (incl. 2.5.1) and whether claims of conformance to the XMI 2.x specification can be independently verified by non XMI experts.

Suggest:

a) Reword this rule to clearly explain all of what is necessary to unambiguously and deterministically determine whether a value of a composite property typed by a class must be serialized as a nested element, a nested reference or an attribute reference.

b) Provide examples for all three options

Table 9.4.1 states:

When the value of a Property is null, it is serialized as XMIValueElement with attribute nil=’true’.

In UML, null means "lack of a value".
MOF (which is based on a subset of UML) inherits this – that is, for MOF, null also means "lack of a value".

Currently, many UML tools do not serialize null-valued properties according to the rule above, instead, the property is not serialized at all.

In principle, any metamodel can have a profile extending it.
The XMI spec only addresses producing an XMI schema for a metamodel. The XMI spec should also address the production of an XMI schema for a profile that extends one or model metamodels.

In principle, any model that is an instance of a metamodel extended by one or more profile can be augmented with the application of such profiles. The XMI spec only addresses the production of an XMI document for the model itself. The XMI document production rules should also cover the augmentation of such a model corresponding to the profile-defined stereotypes applied and to the profile-defined datatypes & classes instantiated.

Table 9.4.1 states:

Choice of:
1. XMIObjectElement
2. XMIValueAttribute
3. Nested XMIValueElement
By default instances of structured datatypes are serialized as if they were classes, as described in 7.8.7. This can be overridden by the tag org.omg.xmi.flattenStructuredDatatypes in which case the values of the Properties are serialized as a single string separated (by default) by commas. The default separator can be overridden by the XMI org.omg.xmi.valueSeparator tag.

The flag org.omg.xmi.flattenStructuredDatatypes adds a layer of complexity for XMI interchange because tools need to handle effectively two different serialization strategies: flatten vs. structured.

The benefit of flattened structured datatype value serialization is questionable because its relevance has not been established in current practice and current implementations.

Suggest eliminating org.omg.xmi.flattenStructuredDatatypes.

Canonical XMI, ptc/2013-08-28, states that xmi:uuid must be always present (B2, #5) except for values that are datatypes (B2, #7).
What does "values that are datatypes" in B2 #7 mean?

Are auxiliary XML elements like mofext:Tag and stereotype instances considered "values that are datatypes" ?

If not, then according to B2 #5, these auxiliary XML elements must have identification attributes (xmi:id and xmi:uuid).

Historically, the OMG used rather arbitrary conventions for producing the xmi:id of mofext:Tag — not a big deal because nobody really refers to them. However, these arbitrary conventions break the repeatability and reproducibility of Canonical XMI!

The Canonical XMI document production rules in B4 are incomplete: they should be augmented to specify the serialization, identification and ordering of auxiliary XML elements.

Consider augmenting the Canonical XMI specification B6 with the following rules:

Generate the xmi:id and xmi:uuid for mofext:Tag by appending "_mofext.Tag" to the xmi:id and xmi:uuid respectively of the referenced element.

Generate the xmi:id and xmi:uuid for stereotype instances by concatenating the xmi:id and xmi:uuid respectively of the stereotype with that of the element on which the stereotype is applied.

The subsection refers to documents [MOF2] and [UML2] which are not further specified in the document.

MOF/XMI defines the schemaType tag as follows: "The name of a datatype defined in the XML Schema Datatype specification."

First,the examples in the MOF/XMI specification and current use of this tag in OMG's XMI artifacts use datatype URIs, not datatype names.

Second, the definition of this tag is vague about which datatypes are allowed.

• The builtin datatypes explicitly defined in the XML Schema Datatype spec? http://www.w3.org/TR/2012/REC-xmlschema11-2-20120405/#built-in-datatypes
• Any datatype defined in the "xs" namespace?

Consider for example "precision decimal", a datatype that is explicitly mentioned but not explicitly defined in the XML Schema Datatypes spec: http://www.w3.org/TR/2012/REC-xmlschema11-2-20120405/#primitive-vs-derived
This datatype is explicitly defined in a different specification but in the "xs" namespace: http://www.w3.org/TR/xsd-precisionDecimal/#precisionDecimal

This example illustrates the ambiguity of the scope as currently specified, that is, is "xs:precisionDecimal" allowed as a value for schemaType?

Third, even if schemaType were to be clarified to be some subset of datatypes defined in the "xs" namespace, there are legitimate business reasons to expect greater flexibility.

- Several OMG specifications define XSDs (e.g., BPMN, DTV, …) If these include XML Schema datatype definitions, why are we precluded from referring to them via a PrimitiveType that has a schemaType tag pointing to their URI?

• For working with RDF, it would make sense to define a PrimitiveType whose schemaType tag maps it to rdf:PlainLiteral — I.e., http://www.w3.org/TR/rdf-plain-literal/#Definition_of_the_rdf:PlainLiteral_Datatype
• Similarly, for OWL, it would make sense to define PrimitiveType that would be mapped to owl:Rational and owl:Real: http://www.w3.org/TR/2012/REC-owl2-syntax-20121211/#Datatype_Maps

This problem is preventing me from implementing a XMI 2.0 Schema generator for a MOF 1.4 metamodeling environment. The specification is often unclear about which type of name (short name, fully-qualified name, or namespace-qualified name) should be used. >From section 1.8.1: “The XML element name for each metamodel class, package, and association in a document is its short name.” Then, in section 2.2.1, rule 4 we have “ClassTypeDef ::= "<xsd:complexType name=’" //Name of Class//…”. Now, the w3 xml schema specification (http://www.w3.org/TR/2001/REC-xmlschema-1-20010502/ states that “no complex type definition can have the same name as another simple or complex type definition.” If the spec means that we should use the short name of the Class in rule 4 above, then it isn’t in line with the xml schema specification, because it’s obvious that several classes in different namespaces of the same metamodel can have the same short name. Furthermore, the w3 schema specification defines the name attribute of a complexType declaration to be of type NCName, or ‘non-colonized name’. This means that we can’t use the namespace-qualified name for the Class either. The only solution is to use the fully-qualified name for metamodel elements** when defining their types. **The fully-qualified name of a metamodel element lists all of the encapsulating MOF Namespaces of the element. So, if you have Class A in nested Package B in outermost Package C, the qualified name of A would be ‘C.B.A’.

At the moment every metaclass has a XML element generated, allowing it to be the root of any interchange.

While this flexibility might be useful in some cases, it clogs up the XSD with elements that will never get used in practice.

Proposed resolution:

Define two boolean XMI Tags:

1. To define whether root elements should be restricted: org.omg.xmi.restrictRoots (defaults to true)

2. To mark classifiers (classes or associations) as being potential roots: org.omg.xmi.rootElement (defaults to false)

Given that the XMI spec contains a metamodel for XML Schema it would be possible to express the Schema production rules using QVT as a transformation from the MOF metamodel to the XSD metamodel. This would create a more rigorous specification that could be automated

These examples in the XMI spec s have nothing to do with modeling but are related to Departments and stoplights.

Furthermore they do not show xmi:ids and xmi:types.

They should also illustrate some of the more sophisticated cases such as that introduced by Issue 16330.

Finally, the Figures in Section 9.4 are all labeled Figure 1

Metamodels do not necessarily use names compliant with XML syntax.

XMI should include a name transformation algorithm to convert names e.g. replace spaces by ‘_’

While for older versions of XML the concrete syntax for persisting UML models was given in DTD, for newer versions of XMI a related concrete syntax specification e.g. as XML Schema is missing (even if the syntax is described in ptc/2010-12-06). Pure syntax compliance tests for given XMI files cannot be performed.

MOF/UML does not require names of elements in a metamodel to be valid XML names.

Therefore the XMI spec should document a transformation to be applied to cope with spaces, punctuation etc to be used for XML element and attribute names.

3:Extension is only referenced from 1a. However it certainly should be valid within an XMIObjectElement.

(3:Extension)* should be added to 2a