Implementation : excerpt of Kermeta code 

Context in source publication

Context 1

... to access to the metamodel The language or library for implementing metrics should include an easy way to navigate through the model and to access to the φ 1 to subclasses by polymorphism. NumberOfCityCon- metamodel. For instance, considering a model element, nectedToA5Metric overrides φ 1 to compute the number there should be a way to access to the referenced ele- of cities connected to the road named A5 . ments as well as the metaclass and its attributes. Kermeta satisfies all the issues cited above, except the first one. Even if a filtering function written in Kermeta Non-intrusivity We have experienced that it is often tempt- is syntactically close to its semantic, a domain expert ing to pollute models with metrics concerns. For exam- can not feel comfortable with writing pieces of Kermeta ple, to add an attribute marked : boolean to the root class code. It is outside of the scope of this contribution to of the domain model, to mark visited objects. However, specify visual or textual syntax usable by the domain this practice violates the separation of concerns principle. expert. A good metrics design practice is totally non-intrusive The prototype involves two main abstract classes Re- with respect to the semantic of the model. In the pre- flectiveWalk and SigmaMetric . To define a σ model met- vious example, the need to mark visited objects implies ric, the user just need to create a new class inheriting the use of M ap : Object → Boolean . from SigmaMetric and to write the associated filtering function. Our implementation One can find in the literature several proposals for the implementation of model metrics (see section 2) e.g., Java, SQL, Python, Xquery, XML- 4 Applications of the σ metric based DSL, OCL, and a graphical language. Proposals mix some of these languages. These proposals do not In this section, we present three case studies so as to adress all the issues cited above. We based our approach illustrate the genericity and the feasibility of the σ met- on the model-oriented programming language Kermeta ric. We first show that the σ metric allows the compu- [38]. tation of logical lines of code in usual languages such as Kermeta is a language based upon the EMF API. Java, the value-added of this approach is the ease of use This facilitates the accesses to models and the navigation compared to equivalent existing approaches. Then, we go through models and metamodels. Furthermore, our in- beyond software metrics and consider non code centric dustrial partners generally use Eclipse Modeling Frame- artifacts such as requirements and system engineering work (EMF) models. Kermeta features closures, which metrics. The application of the σ metric in these various are of great help for the filtering functions. Indeed, a fil- domains shows the genericity of the σ metric. Finally, tering function written in Kermeta is syntactically close since metamodels can be considered as models too, we to the underlying semantic hence very concise. present results of the σ metric at computing metamodel Polymorphism is useful to express modularized met- metrics in a full section. rics. An example of code is shown on figure 2. An abstract class SigmaMetric encapsulates the generic code of the generic σ metric. A class NumberOfCityMetric is defined 4.1 Case study: lines of code (LOC) as a subclass of the SigmaMetric class. NumberOfCityMetric implements the filtering function φ and poten- Kan states that “the lines of code (LOC) metric is any- tially delegates the definition of a subfiltering function thing but simple” [39] (p.88). Indeed, there are ...