We need to have comprehensive knowledge about when, where and how we should use multi-level modeling methodologies and frameworks. There are previous work that introduce patterns and con rm the applicability of the methodologies. However, it is still difficult to select frameworks for various kinds of applications. In this paper, we focus on graphical editing tools as an application of multi-level modeling frameworks, and evaluates the capability for modeling and extensibility of multi-level modeling frameworks. We introduce a data ow model as an example model of graphical editing tools. The evaluation result shows that Melanee can describe the data ow model more accurately and can extend metamodels and metametamodels with less changes to existing elements.
for extending models. However, in EMF, we can access only two levels, model and its metamodel, and it is difficult to de ne our own metametamodel. This problem makes limitations for extending graphical editing tools. For example, adding a new concept of diagrams by third-party developers is difficult due to metametamodel.
Multi-level modeling is an approach that can de ne and manipulate the
arbitrary number of levels of metamodels. There are several methodologies for
multilevel modeling: orthogonal classi cation architecture (OCA)[
We need to have comprehensive knowledge about when, where and how we
should use those multi-level modeling methodologies and frameworks. Lara et al.
[
This paper focuses on graphical editing tool as an application of multi-level modeling frameworks, and evaluates the capability for modeling and extensibility of multi-level modeling frameworks. Our evaluation contributes to select the framework that is the most appropriate for graphical editing tools.
The remainder of this paper is organized as follows. Section 2 describes a data ow model as an example of graphical editing tools. Section 3 evaluate frameworks by de ning and extending the data ow model on them. In Sect. 4 we discuss the result, and our conclusions are presented in Sect. 5. 2
In this section, we describe a model of typical graphical editing tools and introduce a data ow model for using in the evaluation of the paper.
A typical graphical editing tool consists of a palette and a canvas. The palette displays icons of building blocks for composing a diagram. On the canvas, we put icons from the palette and create connections between icons for de ning a diagram.
We consider the typical graphical editing tool has a ve-level model. Table 1 shows the overview of the ve-level model of the graphical editing tool. Models M3 M2 M1 M0
ProcessType
Process
Element port residing in level M4 is the metametamodel of the metamodeling architecture such as MOF. Beware that de nitions of each level of the ve-level model differs from the original MOF de nitions such that the MOF metametamodel resides in level M3. The model only allows instantiation relations between elements of different levels in the model. Metametamodels of the diagram edited on the tool reside in level M3. Metamodels that de ne types of elements displayed in the palette reside in level M2. Models residing in level M1 are diagram instances edited on the canvas, and level M0 is the real-world software de ned by diagram instances in level M1.
Now, we introduce a data ow model as an example of models of
graphical editing tools. Figure 1 shows the data ow model. The data ow model is
used to de ne data processing applications[
In order to help selecting the framework that is most appropriate for models of graphical editing tools, we apply multi-level modeling frameworks to de ne the data ow model5 and evaluate the frameworks from the following viewpoints. Capability for modeling: how well we can represent the data ow model as it is, Extensibility: how easy we can extend metamodels and metametamodels of the data ow model.
We select EMF and two multi-level modeling frameworks, Melanee and MetaDepth6,
which are still actively maintained and support MDE tools such as model
transformation from the list of tools in the multi-level modeling wiki [
EMF is a two-level modeling framework based on MOF. Regarding Fig. 1,
if we use EMF, we have to de ne level M3 and M2 in the same level. There
are several methods for de ning two levels in one level[
Element
in, out Duplication Epsilon languages9 for model transformation, model validation and so on. 7 http://projects.eclipse.org/projects/modeling.mdt.ocl
We de ned the data ow model shown in Fig. 1 on each framework, and conrmed existing limitations for de ning the original model by each framework. Table 3 shows the number of limitations for de ning the original model by each framework.
Figures 2 and 3 show the data ow model de ned by EMF. We found there are three limitations; 1) inheritance relations from ProcessType and DataType should be used instead of instantiation relations, 2) association between ProcessType and DataType cannot be created, and 3) the associations from Duplication and AddTimestamp cannot be de ned as the instances of the association of ProcessType.
Figure 4 shows the data ow model de ned by Melanee. Melanee can completely de ne the original data ow model as it is, because it can represent instantiation relations between levels and instantiate references of level M3 at level M2.
Figure 5 and 6 show the data ow model de ned by MetaDepth. We found there is a limitation; Although MetaDepth can instantiate references of level M3 at level M2, the references cannot be de ned as unidirectional relations. This means Data and Table also have references to Duplication and AddTimestamp, respectively, and it causes violations of the original connective rules between types of Process and Data. 3.2 Extensibility We consider that there are two kinds of extentions of graphical editing tools: adding a new building block to palette and introducing a new notation in the diagrams. In accordance with the two kinds of extentions, we used the following two scenarios.
A) Adding a new type in level M2: as shown in Fig. 7, we added process type Filter in level M2. Filter is the process type that picks up data records satisfying the conditions given by attributes of the process. B) Adding a new conceptual type in level M3: as shown in Fig. 8, we added ControlType in level M3. ControlType is a conceptual type for con guring execution of real-world software and has references to ProcessType and DataType. We also added Control and Schedule in level M2. They are instances of ControlType, and Schedule is the type of elements that schedule the execution of the referenced process.
Table 4 shows the number of modi cations of existing parts as a result of extending the model for each framework. We found we do not have to modify anything in both scenarios using EMF. Regarding Melanee, we just had to
Filter +fieldName: String +equals: String
ProcessType
Process <<port>> in, out
Duplication
Element
port modify the element representing the inheritance relation among Process and its subclasses in scenario A. There is only one modi cation of the inheritance relation among Element and its subclasses in Scenario B as well. Regarding MetaDepth, scenario B requires four modi cations of bidirectional references between process types and data types. 4
We could say that Melanee is more suitable than the other two frameworks to de ne models of graphical editing tools. Regarding the capability of modeling, Melanee can de ne the original data ow model most precisely. Regarding the extensibility, EMF has the least modi cations in both scenarios. Melanee also has the least modi cations next to EMF.
The data ow model having two conceptual types and a unidirectional relation between the types in its metametamodel has the generality as the models of graphical editing tools, because there are various kinds of graphical editing tools for de ning procedures such as data ow, work ow and process ow.
We consider that the results of conducting two scenarios for extending models also have the generalities because of the following facts. Regarding the scenario process <<process>> target
ProcessType
Process
data
Element port
A, the evaluation result basically depends on the number of additional types in level M2. Adding process types or data types just causes the changes of inheritance relations. Regarding the scenario B, the evaluation result depends on the number of additional relations between conceptual types in level M3. If Melanee needs modi cations of relations, MetaDepth generally needs much more modi cations than Melanee.
We consider that there is a limitation in our conclusion. The data ow model
contains only two of the ve patterns shown in [
In this paper, we have evaluated the following frameworks regarding the capability for modeling and their extensibility: EMF, Melanee and MetaDepth. We have used the data ow model as an example of graphical editing tools, which is ve-level metamodels including MOF at the top level.
The evaluation results show that Melanee is more suitable than the other three frameworks for graphical editing tools. Melanee can represent the data ow model most precisely and has less modi cations for extending metamodels and metametamodels. However, we consider that the multi-level modeling frameworks including Melanee still have difficulities regarding developing graphical editing tools and need more compatibilities with MDE tools to facilitate the tool developments.