The adoption of Model-Driven Engineering (MDE) as software development paradigm, is recently growing. The creation of Domain Speci c Modelling Languages (DSMLs) is a frequent task, because it has several bene ts to describe a particular domain. However, the construction of graphical DSMLs is a technically challenging task, and the generated environments do not scale well for large models. In order to improve this situation, in this thesis, we propose borrowing modularity concepts from programming languages. In this manner, DSMLs will be enriched with fragmentation strategies, which will produce environments able to seamlessly partition models across folders and les. To facilitate the de nition of the graphical syntax, we propose a guided, heuristics-based approach, which is capable to pro t from the de ned fragmentation strategy.
The Model-Driven Engineering (MDE) paradigm, proposes software
development by the use of models of higher level of abstraction than code [
Current applications become complex and MDE aims to reduce production
costs. However, while models have a higher level of abstraction than code, for
large systems, they may become large and unwieldy as well. There are di erent
tools that focus on MDE, being Eclipse Modelling Framework (EMF) [
With respect to models, there is an absence of a system that facilitates the modularity, therefore, the composition, extension and re-utilization, becomes di cult. Eclipse JDT for example, basically proposes the organization of Java projects in packages, fragments and compilation units. I propose to organise a model like a programming project, fragmenting it into folders and les. Once the fragmentation is applied, there is a need to de ne scoping rules for references, visibility rules for model elements and mechanisms for e cient check of OCL constraints. The modularity in models would bring as a bene t, the creation of libraries of abstractions, that provide standard solutions to already resolved problems.
A further di culty is the de nition of graphical editors for DSMLs. Graphical Frameworks do not scale well and do not support scalability mechanisms, just barely propose layers and hierarchical drill down, to describe the di erent levels of the model. Thus, I propose a guided, heuristics-based approach, to help users with the de nition of the graphical syntax and afterwards the generation of these environments. Accordingly, the generated editor will be capable to pro t from the de ned fragmentation strategy.
The proposal is to establish modularity mechanisms for DSMLs, which provide scalability for DSMLs. In addition, facilitate the generation of graphical modelling environments that supports this modularity. 2
This research attempts to provide scalability through modularity techniques to build modelling environments for DSMLs. Thereby, we are going to focus on works dealing with the need of process large models. In addition to this, we will review the use of frameworks for the generation of graphical editors. Our approach consists in generating structured model editors from meta-models and speed up the generation of graphical environments through patterns and wizards.
There are some works that use di erent techniques to process large models
and obtain some performance gains. Among them, the work of Scheidgen and
Zubow [
On the other hand, some works decompose models into sub-models for
enhancing their comprehensibility. In [
There are a collection of tools that propose MDE solutions for the development of graphical editors compatible with EMF. There are some Eclipse plug-ins to construct diagram editors, such as: GMF 1 , Eugenia2, Graphiti3 and Sirius 4. Graphiti provides a Java API for coding, the other frameworks are model-based. Sirius, is the graphical modelling framework that is currently becoming popular in the MDE community. However, we need some expert knowledge to use these frameworks, sometimes they require the con guration of complex environment speci cations. Besides that, the constructed editors do not scale well, because it is based on build monolithic models. Our approach is to generate the scalable environment with the required settings. 3
We are going to base our tool in DSLtao, which provides an environment to
design DSMLs through patterns. DSLtao proposes a catalogue of patterns
divided into domain, design, concrete syntax, dynamic semantics and
infrastructure patterns [
We plan to create the environment for DSMLs according to the scheme in Fig. 1. Our tool will take as input a meta-model, with instances of patterns, such as: Modularity, Filter, Scoping, Visibility and its corresponding graphical representation. The interaction of the DSML designer with the tool, would be through wizards, that facilitate the creation of diagram editors, proposing efcient and exible schemes that may be adapted to di erent problems. After applying the patterns to the DSML meta-model, the modelling environment will be automatically generated.
In terms of fragmentation, the modularity mechanism that we currently
support [
Using this modular organization, we plan to add scoping mechanism, to allow
objects being visible or not, between model elements that belong to di erent
locations in the le system or satisfy certain query. We intend to use Epsilon
Object Language (EOL) 5 and Hawk [
In the case of graphical syntax, we propose a wizard to de ne the concrete
syntax. Here, we are going to develop a dedicated wizard to apply the graphical
pattern over the elements of the DSMLs meta-model. This wizard will contain
some heuristics, that generate a graphical representation model, from which the
environment is synthesized, using di erent graphical editor frameworks. Using
this model, we can improve the cognitive e ectiveness, using Moody's principles
for the evaluation of graphical notations [
Finally, we are going to do some research in Model Transformations (MTs).
Taking advantage that the model is fragmented, it could be possible to put
forward a formal description which explains how we can take advantage of the
splitted model and improve the performance of MTs, perhaps in connection with
distributed models for transformation execution [
We have developed EMF Splitter 6, which works both as an independent Eclipse
plug-in, and as a plug-in of DSLtao. EMF Splitter supports both, the de nition
of a modular structure for EMF models [
First of all, we de ne a modular structure based on how Eclipse JDT
organized projects. This Modularity Pattern has three main concepts Project,
Package and Unit, from which we generate an Eclipse plug-in that enables the editing
of models according to this structure [
Additionally, we explored how to combine fragmentation strategies with
visualization mechanisms. The feasibility of this combination is con rmed based
on an evaluation over a synthetic models, and the model sets of the GraBaTs'09
contest. In [
The construction of modelling environments to create and edit models, is
usually an ad-hoc, technically di cult process. In [
We will provide developers with advanced tools, to facilitate the work with large models, facilitating distributed development, through division of models in more manageable chunks.
We are also implementing a pattern based approach for the automatic generation of graphical DSMLs. We are currently working on new features and developing further services like layers and abstractions. In the future, we hope to contribute to the generation of advanced graphical and textual environments. In addition, the approach could be used to the modernization of existing editors. In this moment, the generation of editors is very costly for developers. For that reason, the creation of the wizard to de ne the concrete syntax aims to reduce this e ort.
In this research we will generate the plug-ins that produce the environment using the de ned patterns. These patterns may be applied in combination, then the developer can reuse the functionality implemented. Our goal is to generate scalable modelling environments. 6
We plan some experiments to test the performance of the generated tools and
also validate the construction process. We already made some test with the
GraBaTs'09 models, in which we evaluate the performance of model fragmentation.
In the future, we plan to choose other case studies like the Knowledge
Discovery Metamodel (KDM) 8 models, which provides a representation related to
the existing software assets. The companies use these technology to understand
the functionality and the data of their legacy systems [
One of the main purpose of this approach is to make easier the tasks to developers. For that reason, it is necessary to test the tool with users have some experience in the creation of environments for DSMLs. We will evaluate the
6
Fig. 2: Gantt Diagram-Research Schedule. process performing a case study and additionally a test to assess the quality of our tool. For instance, the wizard for de ning the concrete syntax has been partially assessed in the MONDO project. 7
At this moment, we are developing EMF Splitter that enables developers to build models in a structured way. We are in an early stage of development, but already we made some experiments with large models and how to visually explore them. We also provide a wizard for a semi-automatic generation of graphical editors. Currently, we only support Sirius.
We aim to continue working on this tool, to provide more advanced features in the generation of modelling environments using patterns. The tasks are summarized in Figure 2, with an approximation of the estimated time. We plan to nish the thesis by the end of 2018.