PRMS Report

Abstract

• This report developes the idea of a visual graphical editor for Epsilon and the process of developing it.
• A fully working epsilon workflow builder and visualizer is also presented that helps users to compose Epsilon workflows and construct visual representations from textual ones.

Statement of Ethics

• Informal user survey for requirements.
• Formal user survey to evaluate the tool ?

Introduction

Motivation

• Why EWE ? What is the need ?
• Workflow inherently visual - flow of order of tasks.
• e.g. BPM tools all provide editors.
• MDE is essentially a sequence of steps from start to finish.
• Other model management languages, frameworks tools provide it as well. (Atlas, OpenAW, etc)
• Therefore Epsilon, needs one too.

Method

• Documents the process of building such a tool using MDE techniques and Epsilon itself.

Literature Review

• Introduce ALL concepts and technologies that are critical to understanding the tool.
• Model driven engineering - tasks involved. Tasks constitute a workflow.
• Workflows in general - BPM workflows, Model driven engineering workflows.JBPMN
• Build tools - Visualize Build Tools - ANT, VizAnt, Ant Explorer, Ant Utility
• Other MDE Workflows - OpenArchitectureWare, ModelBus Atlas MegaModel Management
• How does Epsilon handle it ? Internal architecture of Epsilon Workflow.
• Talk about EMC Layer - this will be useful later - to explain how powerful it is to directly use EMF API or DOM API.
• ANT, Java Modules, Eclipse Launch configuration, Working outside Eclipse.
• Provide a general introduction to build management, concept of dependency, task ordering, conditional ordering, task failures, transactions.
• How ANT provides some of these features as a build tool. How Epsilon uses ANT as an architecture - quote from Epsilon ANT paper.
• Introduce idea of domain modeling. Separate the concern of design and architecture. This is the CORE idea.
• Eclipse EMF technologies. How EMF supports Domain Modeling. Give a simple Library example.
  • Library domain is expressed. An Editor for the domain is generated - but uses unique architecture. e.g. Adapter, Factory and Command patterns - which WE never specified while modeling the domain.
• Make sure this concept is well understood. MDE in general and how EMF tools support MDE with practical examples.
• Eclipse graphical technologies GEF -> SWT, JFace.
• Eclipse GMF as a bridge between GEF and EMF.
• How GMF is Model based and Generative
• How it fits well for Epsilon since both are based on EMF.
• Introduction to EuGENia.

Requirements

• Initial project description
• Asking Epsilon users to comment on forum
• Discussion with Supervisor
• Self proposed requirements

Design

• Eclipse vs NonEclipse
• EMF in built editor vs (GMF/Graphiti)
• GMF vs Graphiti
• GMF - EuGENia support (Generate GMF vs manual GMF model creation), in house expertise with GMF models, EuGENia authors based in York, customization.
• EuGENia GMF support for EVL and error/warning markers.

ECore metamodel design

• Represeting ANT concepts as GMF nodes for e.g. Project, Target.
• Using concept of depends and follows.
• Accurate modeling of ANT structure - allows operations to be performed at the higher Task level as opposed to lower levels - such as EOLTask etc.

Scripting design choices

• Using Java vs using Epsilon languages
• Epsilon :- reflective features. Easy model navigation.
• Generate Java Module code ? Or Generate ANT XML Code ?
• ANT XML - To support legacy use case and avoid complexity of Java module code.
• EEF Model manipulation - Instead of editing code, was done using EOL scripting.
• Using EGL:
  • Protected regions: Merge engine
  • XML Formatter
  • Full reflective support.

Editor Design

• Task to GMF node mapping design choices (Which elements in the metamodel should be visualized)
• Compartments vs Property views - Too boxy. So rejected.
• Property Sheet
• Property Sheet implementation choices : Custom SWT code (rejected/knowledge of SWT - Time constraints: Used by Ecore editor)
• EEF as the choice.
• EEF: new technology, never been integrated with GMF before.
• Challenges in customising EEF - Poorly documented code base.

Parsing Design

• XML Parsers vs EOL vs ETL

Implementation

• Development methodology used ? - Justify.
• ECore diagram editor was used to specify the metamodels.
• GMF annotations were made in the editor.
• What all GMF annotations were used.
• How GMF models were customized after generation.
• EEF implementation.
• How EEF was plugged-in.
• Customizing code-generation properties - Namespace, package structure, documentation.
• Generated code customization: For e.g. disabling unwanted creation tools.

• EGL implementation.
• EVL implementation.
• Parsing implementation
  • EOL vs ETL
  • ETL was chosen - why ?
  • ETL implementation choice - Reflective vs Declarative (Balance dealing with future change vs ease of code)

Evaluation

• Where to list identified enhancements for the project ?
• EWE itself - support conditional statements.
• EuGENia enhancements identified - bugs.
• Integrate EEF within EuGENia MDE process.
• Contribution of EWE itself <- In Evaluation section ?

Project Description

• Evaluation will proceed by applying the new contributions to the examples of existing workflows, and demonstrating differences with respect to functionality and development process.
  • Difference in functionality between automated and manual methods.
  • Ease of usability in composing workflows.
  • Ease of modifiability of existing manual workflows using the visual editor.
  • Ease of use for first time users in building workflows and understand Epsilon Workflow language.

• • Development process refers to what ?
• An analysis of the ways in which Epsilon users construct and use workflows.
• Three improvements to the prototypical Epsilon workflow language to support visualisation and automatic code generation.
• A comparison of the prototypical and improved versions of the Epsilon workflow language.

Evaluating the Artefact

• Test cases as a method of evaluation?

• Manual testing and verifications against requirements.
• Black box testing for evaluating against user requirements.
• Test case with a description against each FR and NFR. Process of executing, screenshots etc.

• White box testing:
• Cite EUnit paper about Difficulty in performing white box testing for MDE tasks.
• Develop some Sample EUnit cases.

User Evaluation

• Similar to Case Study?
• Explain why it makes more sense for an actual user evaluation rather than a self performed case study.
  • Self case study : usability aspects difficult to evaluate due to intimacy with tool.
• Informal live User Evaluation: Perform series of exercises designed to evaluate usability of new tool.
• Discuss results of experiments ? (Describe experiments/exercies here as well).
• Ethical considerations.

Evaluating the metamodel design

• Expectation
• Result
• Evaluation

Evaluating the development process

• Huge productivity benefits.
• Understanding complex UI frameworks avoided.
• No of classes and code gneerated enormous.
• Focues on higher level abstraction - code gen, viz, validation would not have been possible.

• Significant time was spent - tweaking model, generating code.
• Finding out where in code modifications need to take place.
• Significant time was spent in researching EEF integration with EuGENia generated GMF editor. Trial/error, forum posting

Conclusions

• Provide a concise summary of the whole project.
• Making accurate predictions of results that would be obtained by applying the present work in new areas.
  • Extend it as a GMF editor for any complex application that needs to use Ant as a basis..g. JBPM - JBoss.
  • Use it to visualise complex Ant builds within Eclipse IDE.
  • More ?

• FURTHER WORK here or in Evaluation.
• Lessons learnt, personal reflection.