Issues for Mailing list of the Requirements Interchange Format V1.2 (ReqIF) Revision Task Force

List of issues (green=resolved, yellow=pending Board vote, red=unresolved)

List options: All ; Open Issues only; or Closed Issues only

Issue 16013: XHTML integration does not work as expected
Issue 16014: Associations of RelationGroup are limited to XML document scope
Issue 16399: Specification inconsistencies and misspelllings
Issue 17553: AttributeDefinition should allow mutliplicities for attribute values
Issue 17554: Allowed source and target elements on SpecRelationType level
Issue 18147: Missing class description for SpecElementsWithAttributes
Issue 18148: AttributeValue should not be subclass of Identifiable
Issue 18784: Implementation Guide

Issue 16013: XHTML integration does not work as expected (reqif-rtf)

A key feature of the Requirements Interchange Format is the ability to exchange formatted content (e.g. bulleted lists, text with fonts applied etc.) between requirements authoring tools. 
Transporting formatted content is implemented in ReqIF by the use of W3C’s XHMTL, and the XHTML integration does not work as proposed in the beta2 version of ReqIF, rendering the format almost useless for non-trivial exchanges.


Background:
In earlier, “non-OMG” versions of the Requirements Interchange Format, XHTML had already been incorporated, but using a proprietary, customized XHTML schema. 
For better compliance with international standards, XHTML has been incorporated using a XML schema driver for the beta2 OMG version. 


As only a subset of XHTML should be included, the project group needed to incorporate so-called XHTML XML schema “modules”, which proved to be a complex task. 
Not until the implementation of a ReqIF based transformation tool, it turned out that some XHTML element definitions were not pulled in from the W3C modules, but only their XML schema types.


This currently makes it impossible to use these XHTML elements in XML documents conforming to the ReqIF XML schema.



Resolution:
Rewrite the XHTML schema driver (which incorporates XHTML in ReqIF), in a way that all needed XHTML elements can be used in ReqIF XML documents.
In other words: replace the file driver.xsd with a new version that works as expected and adapt chapter 11.4 of the specification, which shows the content of driver.xsd, accordingly.
This new version has already been created and tested by ProSTEP.

ReqIF uses two distinct ways to reference XML identifiers in its XML schema: 
1. LOCAL-REF (which is used for XML references within the same XML document) and 
2. GLOBAL-REF (which is used for arbitrary, meaning document scope or cross document, XML references).
By mistake, the XML references representing the associations of the RelationGroup type have been made LOCAL-REF references.


The consequence of the mistake is: the source specification and target specification associated to a RelationGroup instance must be contained in the same XML document as the RelationGroup instance, or otherwise the references would be broken and the XML document would not validate against the ReqIF XML schema.


Putting both the source specification, the target specification and the relation group in one XML document can significantly increase the size of XML documents.  
Even more important, it would affect almost all exchange processes between companies and their suppliers who use older (pre-OMG) versions of Requirements Interchange Format, as they have been able to exchange single specifications and relation groups separately with these older versions. 
This would be a step backwards for these companies and reduce their willingness to use tools that adopt the ReqIF standard.


Resolution:
Simply change two XML attributes in the ReqIF XML schema (lines: <xsd:element name="SPECIFICATION-REF") from LOCAL-REF to GLOBAL-REF as shown below :


Before resolution of the issue:
<xsd:complexType name="RELATION-GROUP">
    <xsd:all>
      …
      <xsd:element maxOccurs="1" minOccurs="1" name="SOURCE-SPECIFICATION">
        <xsd:complexType>
          <xsd:choice maxOccurs="1" minOccurs="1">
            <xsd:element name="SPECIFICATION-REF" type="REQIF:LOCAL-REF"/>
          </xsd:choice>
        </xsd:complexType>
      </xsd:element>
      …
      <xsd:element maxOccurs="1" minOccurs="1" name="TARGET-SPECIFICATION">
        <xsd:complexType>
          <xsd:choice maxOccurs="1" minOccurs="1">
            <xsd:element name="SPECIFICATION-REF" type="REQIF:LOCAL-REF"/>
          </xsd:choice>
        </xsd:complexType>
      </xsd:element>
      …
  </xsd:complexType>


After resolution of the issue:
  <xsd:complexType name="RELATION-GROUP">
    <xsd:all>
      …
      <xsd:element maxOccurs="1" minOccurs="1" name="SOURCE-SPECIFICATION">
        <xsd:complexType>
          <xsd:choice maxOccurs="1" minOccurs="1">
            <xsd:element name="SPECIFICATION-REF" type="REQIF:GLOBAL-REF"/>
          </xsd:choice>
        </xsd:complexType>
      </xsd:element>
      …
      <xsd:element maxOccurs="1" minOccurs="1" name="TARGET-SPECIFICATION">
        <xsd:complexType>
          <xsd:choice maxOccurs="1" minOccurs="1">
            <xsd:element name="SPECIFICATION-REF" type="REQIF:GLOBAL-REF"/>
          </xsd:choice>
        </xsd:complexType>
      </xsd:element>
      …  </xsd:complexType>


As a side effect, the XML namespace needs to based on a new date (page 84 of the beta2 specification).

§10.2 p29 Fig10.2: Identifiable-AlternativeID composition, ident back linkage is named (with multiplicity) but shown as not navigable, the §10.8.2 p39 is unclear
§10.6.2 p36 Fig10.11: EnumValue-EmbeddedValue composition, properties role name, but multiplicity is 1 -> either * multiplicity or property role name
§10.6.3 p37 Fig10.12: AttributeValueXHTML dual compositions, if treated as compositions, duplicate back linkage attributeValue role names -> better treated as attributes (similar strong composition)
§10.8.3 p40 : AttributeDefinition, on composition, specType back linkage multiplicity should be 1 (and shown as 1 in Fig10.3 p30 ) -> 1 multiplicity
§10.8.9 p46 : AttributeDefinitionSimple associations (including composition) are redundant (when implemented) with associations of concrete realizations (AttributeDefinitionInteger, Boolean, Date ...)) -> to be removed
§10.8.18 p53 : AttributeValueSimple associations : same as AttributeDefinitionSimple -> to be removed
§10.8.18 p54 : AttributeValueSimple : description and semantics are identical -> precise Semantics if required
§10.8.20 p55 : AttributeValueXHTML : already noticed -> theValue and theOriginal should better be considered as attributes
§10.8.28 p64 : DatatypeDefinitionString, maxLength attribute, spelled in lower cases, but spelled with a L upper case in Fig10.9 &10.10 -> maxLength for consistency 
§10.8.30 p66 : EmbeddedValue : description and semantics are identical -> precise Semantics if required
§10.8.31 p66 : EnumValue : misspelling in role name of back linkage to DatatypeDefinitionEnumeration, missing 'y' and 'T' shoulf be 't' for consistencies -> datatypeDefEnum 
§10.8.31 p66 : EnumValue : properties association role name while 1 multiplicity, already mentioned 
§10.8.31 p67 : EnumValue : description and semantics are identical -> precise Semantics if required
§10.8.32 p67 : Identifiable : oonstraint #2 are redundant with Identifier attribute description
§10.8.35 p70 : ReqIFContent associations : redundant composition association with RelationGroupType and with SpecType
§10.8.36 p71 : SpecHierarchy : description and semantics are identical -> precise Semantics if required
§10.8.37 p73 : Specification : description and semantics are identical -> precise Semantics if required
§10.8.40 p75 : SpecObjectType back linkage associations : redundant back linkage with SpecType, back to ReqIFContent
§10.8.41 p76 : SpecRelation : description and semantics are identical -> precise Semantics if required
§10.8.42 p77 : SpecRelationType back linkage associations : redundant back linkage with SpecType, back to ReqIFContent


=> the documentation should better be automatically generated from the UML model (document and diagrams would be consistent). Or at least, one should implement this metamodel from the spec in a tool to check the consistency.

Section 10.3 says: "What is actually shared among the requirements is the requirement attribute definitions (the number of attributes, the names of the attributes, the default values for the attributes, and the datatypes of the attributes.)"


However, there is no possibility to allow more than one value for an attribute of a SpecType. This is a very fundamental conecept and should be provided by ReqIF 1.1.


A potential solution might be the following:


- Add a metaattribute upperValueBoundary to AttributeDefinition with default set to 1 (meaning, the attribute is allowed to have exactly one entry).


- Add two more metaattributes to AttributeDefinition: isOrdered:Boolean = true and isUnique:Boolean = true to enable the user specifying the nature of a list of attribute values

- Add a constraint to AttributeValue that ensures that the upperValueBoundary of the corresponding AttributeDefinition must be respected by an implementation

An implementation of the above requirements would require a change to the ReqIF model. The impact of such a change on the existing ReqIF implementations is unclear and needs further analysis. The ReqIF1.1 RTF, including the issue’s author Marc-Florian Wendland, agreed that changes to the ReqIF model should currently be omitted for standard stability.
Parts of the requested functionality can already be achieved using the ReqIF1.0.1 standard: by using an AttributeDefinitionEnumeration element with multiValued set to true, multiple string literals can be referenced from each SpecObject instance.
This issue will be reconsidered in a future RTF.
Disposition: Deferred

Currently, any SpecObject can be associated with any other SpecObject by using a SpecRealtion. Even though this is very generic and flexible, it is prone to erros, since the user is not restricted by the type of a SpecObject that can be used as source or target element for SpecRelation. So, subsequent tasks have always to ensure that only allowed SpecObject can be linked with each other. Unfortunately, it is not possible to deposit the knowledge about what SpecObjects are allowed in the model itself.

This situation can easily mitigated by adding a sourceType and targetType association to SpecRelationType. Further more, a constraint on SpecRelation must be defined that ensures that the only SpecObjects are linked as source and/opr target that adhere to the SpecRelationType source and target.

An implementation of the above requirements would require a change to the ReqIF model. The impact of such a change on the existing ReqIF implementations is unclear and needs further analysis. The ReqIF1.1 RTF, including the issue’s author Marc-Florian Wendland, agreed that changes to the ReqIF model should currently be omitted for standard stability.
Parts of the requested functionality can already be achieved using the ReqIF1.0.1 standard: by using two Specification instances, each instance referencing only SpecObjects of one SpecObjectType, and then using a RelationGroup with a sourceSpecification attribute and targetSpecification pointing to the above Specification instances. The SpecRelation instances referenced by that RelationGroup instance are thereby constrained to use only SpecObject elements from the above Specification instances.
This issue will be reconsidered in a future RTF.
Disposition: Deferred

The description of the very fundamental metaclass SpecElementWithAttributes is missing in the ReqIF class descriptions sections. This is a pure editorial issue, but important to be resolved since SpecElementWithAttributes is one of the central metaclasses of the abstract syntax.

The class description of AttributeValue states that AttributeValue is a subclass of Identifiable. This is not correct according to the ReqIF model and the ReqIF XML schema, which state that AttributeValue has no super class.


Making AttributeValue a subclass of Identifiable in the specification was simply a copy and paste error during initial specification creation, as AttributeValue instances can uniquely be identified by their related SpecElementWithAttributes and AttributeDefinition (both have identifiers).

So the proposal is: remove the Generalization relationship between AttributeValue and Identifiable in the text of chapter 10.8.12 of the specification.

As discussed at the co-located Req-IF/OMG meeting in Berlin (June 2013) it would be more in the spirit of OMG standard openness to include the non-normative recommendations for implementing the standard, in the standard document itself, rather than as a link to a non-OMG web site where there is a charge of 90 Euros to download the 'implementation guide'.  This issue should be considered for resolution and could either result in; 1/ removal of section 12 Annex A from the specification, so that it stands alone, 2/ creating the 'implementation guide' as a freely available OMG document (still linked from the specification) or 3/ including the non-normative implementation guidelines as a section directly in the specification.