Issue 1446:  CorbaUnion should redefine the has_feature relationship (boca-rtf)
Source:  (, )
Nature:  Revision
Severity: 
Summary: Summary: Section 2.2.4 (BOCA Meta-Model in CDL)
 
 identifiable CorbaUnion: CorbaStructuredType{
         //attribute CorbaType discriminator_type;
 relationship discriminator_type Uses CorbaType;
 };
 ...
 identifiable CorbaUnionMember:StructuralFeature{
         attribute any label;
 };
 
 The corresponding UML diagram (section 5.3) shows an association wherein CorbaUnion is a composition 
 of CorbaUnionMember(s).  Yet, this is not reflected in the CDL.  It is not reflected in the CDL because of the 
 fact that this relationship is actually a specialization of the relationship between FeatureOwner and Feature.  
 It is understandable that this relationship was not declared in the CDL because of the following problem:
 

Resolution: 
Revised Text:  CorbaStructuredType{
Actions taken:
June 3, 1998: received issue
Discussion: 

End of Annotations:=====
*ISSUE 18*

Section 2.2.4 (BOCA Meta-Model in CDL)

identifiable CorbaUnion: CorbaStructuredType{
        //attribute CorbaType discriminator_type;
relationship discriminator_type Uses CorbaType;
};
...
identifiable CorbaUnionMember:StructuralFeature{
        attribute any label;
};

The corresponding UML diagram (section 5.3) shows an association
wherein CorbaUnion is a composition 
of CorbaUnionMember(s).  Yet, this is not reflected in the CDL.  It is
not reflected in the CDL because of the 
fact that this relationship is actually a specialization of the
relationship between FeatureOwner and Feature.  
It is understandable that this relationship was not declared in the
CDL because of the following problem:

The mapping of a BOCA relationship declaration to server IDL (rather
than to meta-object IDL) yields a set 
of operations.  A redefinition of the same relationship in subtypes
would be, essentially, a covariant 
specialization, redefining the set operations of operations in the
subtype.  If the redefinition of the 
relationship narrows the target type then the mapped operations in the
server IDL are also redefined such 
that their parameter types are narrower.  Since IDL does not allow
redefinition of an operation in a subtype 
(as does, for example, Eiffel), BOCA specifically prohibits this form
of covariant redefinition in static subtypes 
and, as a corollary, cannot allow this form of redefinition of a
relationship in a static subtype. (Note: This is 
allowed in dynamic subtypes, i.e. those with a RoleOf relationship to
another type.)

There has been some debate about how to handle this unfortunate
limitation imposed by the necessity to 
map to the legacy IDL meta-model.

Recommendation:  CorbaUnion should redefine the has_feature
relationship it inherits from FeatureOwner 
by adding a constraint saying that the relationship can be populated
only with CorbaUnionMember 
instances.  CorbaUnionMember should redefine the defines relationship
it inherits from Feature by adding a 
constraint saying that the relationship can be populated only with a
CorbaUnion instance.

Explanation: BOCA specifically permits adding constraints to features
in static subtypes, because 
constraints don't affect the server IDL operations and attributes that
are generated.  In general, this should 
be the approach to  working around the prohibition on narrowing the
parameter types of features when they 
are redefined in subtypes, including the redefinition of attributes,
operations, and relationshipsrowing the types referenced by the
features directly, constraints that in effect narrow the types 
should be declared.

Note!: It is not clear whether BOCA expressions support testing the
type of an element.  If they do not, an 
appropriate upgrade to expression capabilities will be needed (see
issue 63)