This paper contains a summary of the talk given by the author on the occasion of the MODELS 2013 most in uential paper award. The talk discussed the original paper as published in 2003, the research work done by others afterwards and the author's personal re ection on the award. Version Control of Software and System
There are two main usage scenarios for design models in software and system development: models as sketches, that serve as a communication aid in informal discussions, and models as formal artifacts, to be analyzed, transformed into other artifacts, maintained and evolved during the whole software and system development process.
In this second scenario, models are valuable assets that should be kept in a trusted repository. In a complex development project, these models will be updated often and concurrently by di erent developers. Therefore, there is a need for a version control system for models with optimistic locking. This is a system to compare, merge and store all versions of all models created within a development project.
We can illustrate the use of a version control system for models as follows. Let us assume that the original model shown at the top of Figure 1 is edited simultaneously by two developers. One developer has decided that the subclass B is no longer necessary in the model. Simultaneously, the other developer has decided that class C should have a subclass D. The problem is to combine the contributions of both developers into a single model. This is the model shown at the bottom of Fig. 1.
We presented the basic algorithms to solve this problem in the original paper
published in the proceedings of the UML 2003 conference [
C
Original Model A C part of Fig. 2. In this case, the di erence of the models contains negative model elements, i.e., elements that should be removed from a model.
The original UML 2013 paper contains two algorithms to calculate the
difference between two models and to merge a model with a di erence. Given two
models M1 and M2 de ned in UML or another modeling language based on the
Meta Object Facility (MOF) [
M2
M1 +
M1 = = M2
Once we know how to operate with di erences between two models, we can solve our original problem by computing the union of two versions of a model as follows:
M nal = Moriginal + (M1
Figure 3 shows an example of the application of this operation.
The de nition of these operations is complicated by the fact that two developers may have changed the same subset of a model. This is called a con ict and +
D + - B = D it may prevent us to merge all changes into the nal model. The original paper also presents an algorithm to resolve some of these con icts automatically. Automatic con ict resolution is an important component of a version control system with optimistic locking since it can reduce the need for a cumbersome manual intervention when merging models. 2
The original paper published in the UML 2003 conference contained many assumptions and limitations that were tackled by other authors after its publication.
An important goal of the original paper was to provide algorithms that are
independent of the actual modeling language. This allows us to manage
models in many di erent languages, as long as these languages are de ned using a
MOF-base metamodel. On the other hand, the original algorithms cannot use
information that is speci c to a particular modeling language to improve the
di erence, merge and con ict resolution algorithms. In contrast, other authors
have presented algorithms that are speci c to one particular modeling language.
Examples of this is the work of Nejatil et al., who presented algorithms that are
speci c to match and merge statecharts [
Also, an signi cant design decision of the algorithms published originally is
that model comparison is based on matching model elements by identity. The
algorithms assume that each model element has a universally unique identi er
(UUID) that is unique and constant. This is an adequate choice for a version
control system, where the models to compare are two versions of the same original
model. However, it is not a suitable approach to compare two arbitrary models.
Other authors have presented model di erent algorithms that mach elements by
similarity [
We should note that an important limitation in the original approach is the
assumption that the metamodel describing the modeling language used in the
artifacts to process never changes. This is not a realistic assumption, specially
when using internally developed domain speci c modeling languages. This
problem was tackled by Gruschko et al. who have studied the problem of metamodel
evolution and how to update a model when its underlying metamodel changes [
The original paper did not discuss the representation of model di erences
in any structured format. This problem was studied by other authors that
proposed for well de ned metamodels to represent model di erences. Cicchetti et al
presented a model di erence approach [
Finally, it is worth to mention that while the original paper contains the
key algorithms for a version control system for software models, it does not
actually present an implementation of the algorithms into a working system.
Other authors have implemented model management tools such as DSMDi [
The original paper published in the UML 2003 conference was selected for the 10 years most in uential paper award presented during the MODELS 2013 conference. As a researcher, I consider that achieving long lasting impact among academics and practitioners is the ultimate goal of our work. Therefore, I believe that the reader may be interested in my own personal re ections about the award and the context that made the paper become in uential to many other works.
I consider that one of the main reasons for the high impact of the original paper is that is was published timely. The modeling community was then focusing in the MDE research agenda and there was a clear need for this kind of work. The original paper presented the basic algorithms for a version control system, but it also described, sometimes implicitly, many other related problems to be solve. Another important factor that helped the paper to become highly in uential is that it was published at the right venue. In 2003, the participants of the UML conference were the right audience for this work, who quickly extended and improved it.
I think it is interesting to discuss what would be the outcome of the original paper if it had been submitted to a high quality conference such as MODELS in the year 2013. This is just my personal speculation, but I consider that the original paper would not have performed that well in the selection process of a contemporary conference. Currently, the program committee of conferences expect an extensive evaluation and validation of the research contributions presented in research papers, something that is actually missing from the original paper. I have the opinion that the requirement for the inclusion of an extensive validation helps to improve the quality of the published papers, but may be detrimental for the quick dissemination of novel ideals, as the ones contained in the original UML 2003 paper.
To conclude, I consider that ten years later the research on model
comparison and versioning is not completed yet. For those interested in this problem,
1 http://www.eclipse.org/emf/compare/
Altmanninger et al. have published a plea for more research work in this area
and a concrete research agenda [
I would like to thank and acknowledge the work of Dr. Marcus Alanen, co-author and main contributor of the original paper \Di erence and Unions of Models", published in the UML 2003 Conference.