UML22 — UML2: Need a better mechanism for integrating UML2 Profiles

UML2 Superstructure specifies how to define a Profile, how Profiles can reference other Profiles through PackageImport and ElementImport, and how one stereotype could extend another through generalization/specialization. However, this is insufficient for profile integration as it results in too much coupling between profiles. What is needed is a more flexible mechanism for integrating UML2 profiles.

For example, both UPDM and SysML are UML2 profiles. UPDM would like to reuse certain stereotypes from SysML in order to provide effective integration in cases where consumers want to use both. However, UPDM would also like to be able to stand alone in cases where SysML isn't needed. The problem is how to model the overlapping stereotypes and classes without creating coupling that would require all applications of the UPDM profile to also require an application of SysML.

Consider a concrete example of overlap between the profiles, the stereotype ViewPoint. Both UPDM and SysML have a similar concept of ViewPoint, for similar purposes. However, each has its own specializations of ViewPoint, and possibly associations between ViewPoint and other stereotypes. There are a number of approaches for handling this overlap, but none are adequate or practical.

1. Profile refactoring: Each profile could factor its stereotypes into packages, and arrange the navigability of its associations to decouple its stereotypes in order to support anticipated reuse. This is what UML2 did, quite unsuccessfully, with the Abstractions packages. This isn't practical because 1) no existing profiles do it, 2) it is impossible to anticipate all the possible reuse opportunities and to design a profile to support them, and 3) it is sometimes impossible to define the associations between stereotypes to ensure the necessary decoupling.

2. Use ElementImport to select only the stereotypes you need, then subclass to minimize the coupling: This can work, but it results in complex profiles with possibly a lot of subclasses simply to integrate with other profiles. For example, UPDM couldn't use ViewPoint directly, it would have to create a subclass, either coming up with a new name, or putting its ViewPoint in a different Package so that it wouldn't collide with SysML. This is confusing, and results in stereotypes with either the same meaning but different names, or two stereotypes with the same name in different packages. This also requires both profiles to exist, even though the both don't need to be applied. This is again an undesirable side-effect of too much coupling.

Both of these approaches end up inhibiting profile integration and reuse resulting in limited integration between OMG submissions. UPMS had wanted to include integrations with many other submissions including RAS, BPDM, BPMN, ODM, QoS, and BMM. However we could not determine a practical way to do this with current technologies and did not include many of these integrations because of the resulting risk, complexity and coupling. This is a particular problem when we consider the OMG specifications, profiles, and metamodels in an enterprise architecture context where the relationships between the parts are critical to delivering value.

UML2 provides a solution to this problem for extensions created using MOF metamodels to model capabilities. PackageMerge can be used to merge metaclasses with the same name from different capabilities in order to mixin their capabilities. What is needed is a similar capability for UML2 profiles.

A proposed solution would be to extend UML2 Profiles to include similar merge semantics when multiple profiles containing the same classes or stereotypes are applied to the same model. When a Profile is applied to a Package, the Classes and Stereotypes in the Profile would be merged with Classes and Stereotypes of other Profiles that have already been applied. The rules for PackageMerge can be used to define how this merge is done as they already apply to Class, and can equally apply to Stereotype which is a specialization of Class. Conflicts resulting from the merge could be considered defects against the profiles that could be handled in an RTF.

Consider the same example above; both UPDM and SysML define ViewPoint.

3. Profile Merge: The UPDM submitters would be careful to use ViewPoint is a manner that is semantically consistent with SysML since SysML already existed. However UPDM conuld extend ViewPoint with additional properties and associations for its purposes. The UPDM submission could note to users that ViewPoint is a stereotype in UPDM that represents a "placeholder" to ViewPoint in SysML. Users could then apply UPDM to a model, and get UPDM's ViewPoint capabilities without any coupling or need for SysML. Later on, another user could decide to apply SysML to the same model in order to use its modeling capabilities. The SysML::ViewPoint would be merged with the UPDM::ViewPoint allowing the shared semantics to be supported without making any changes to the existing model. Similarly, users could have started with SysML and later applied UPDM to achieve the same effect.

This is a significant change to UML2, but may be an urgent issue due to the number of other profiles and submissions looking for a solution to this problem.

No Data Available