MDD Case-Study: RCP Code Generation in an Airline Company

I was recently asked to translate into english  the case-study presented by BritAir during MD Day 2010. So, here is a summary.

BritAir, a french airline company, is a subsidiary of Air France which transported about 3,5 millions of passengers last year.

Another activity of BritAir is the training of pilots, commercial staff and maintenance technicians of other companies.

Despite BritAir has no developer, the company wanted to keep the control and the quality of its applications:

• Normalize the user interface of the applications
• Normalize the way the applications are designed
• Improve the development productivity
• Document the applications

To achive these goals, Olivier Corbel, the IS Manager, decided to put in place a new development process based on a Model-Driven approach. This project was conducted with Mia-Software and Sodifrance.

The principle of the BritAir approach is based on two main ideas:

• A model describing the business data (entities, identifiers, attributes, child/parent relationships)
• An automatic generation of graphical interfaces (Eclipse RCP) for classic functions (cread, read, update, delete, search and export to Excel)

In addition, the business logic is also described in the model: a concrete syntax has been defined to declare some rules in french. It allows to declare variables, to set values, to test, to loop, etc.

The model is created with MagicDraw, an UML modeling tool, and translated into code (Eclipse RCP and Hibernate) with Mia-Studio, the model transformation engine developped by Mia-Software.

The generated applications are complete and don't require any additional code (the business rules are also transleted into Java). They are deployed on servers with Tomcat, Xfire and Oracle.

It took 300 days (total effort between march 2008 and july 2009) to setup the process, define the target architecture, adapt the tools and train BritAir's staff. The team was composed of three people of Mia-Software (a project manager, a MDA enginer and a JEE architect) and one architect of BritAir (requirements and validation).

Several applications have been developped in parallel, in order to challenge and test the new process and tools.

Now, each time a specific development starts, it uses this approach. Until now, six applications have been developped:

• small maintenance of aircrafts
• management of aircraft events
• cost accounting
• financial forecasts
• management of passengers
• preparation of flights. 

Olivier Corbel has identified several benefits of this MDD process:

• Industrialization of the development activity: normalization of modeling and coding, and better documentation.
• Independance of the models from the target architecture (neither RCP nor Hibernate concepts in the models).
• Design is mandatory, in order to generate the application.
• Better quality of code, since it has been qualified first and then converted into generation rules
• Better productivity: during its talk, Olivier Corbel said that the development managers have evaluated to 10 the productivity ratio on these projects!
• User experience: the applications are homogeneous and always provide the same ergonomy to the users
• Prototyping: it is easier to provide early versions to users and take their feedback into account to improve the application

For BritAir, the next step now is to bring evolutions to the platform:

• Support authorization management
• Extend the expressiveness of the business rules language
• Improve the integration of new applications with existing ones (generated or COTS)
• Add a new generation target (web client)

MD Day 2010

Like every year, the french modeling community has met during the MD Day last week in Paris.

It was the fourth edition, with a format which makes the success of this event: each partner (11 this year) must come with one of its customers who presents a real-life project where model-driven engineering has been used. Once again, despite optimistic forecasts, we still ran out of place, and we have been forced to close the registrations three weeks before the event ! Finally there have been 275 attendees (234 last year).

This year, the conference started with Steve Cook's keynote "UML: Past, Present and Future". 

Stevecook uml mdday2010

I already knew the past: I've worked a while with Booch notation in the 90' (remember these little clouds...) ;-) 

What I found really interesting was the part about the UML dilemnas:

• What is the real value proposition of UML ? There is often a misandurstanding about why to create UML models.
• Should DSLs be based on UML ? UML should be more reusable.
• Is UML appropriate to describe non software domains ? It was not the initial intention of this metamodel.
• Can UML be a pivot to model semantics ? New versions of OO languages integrate concepts which are not supported by UML.

Regarding the future of UML, Steve presented the main initiatives:

• UML is too complex: it is going to be simplified, in order to facilitate its modularization and isolate the parts which are really used.
• MOF is going to disapear: UML will be defined by itself, in order to facilitate the manipulation of metamodels with any UML compliant tool.
• UML Diagrams are going to be formally defined: a Diagram Definition initiative has started

These problems are real, but as Eclipse Modeling Project already proposes solutions to solve this kind of problems, I hope that UML team is not going to reinvent the wheel.

After the keynote, the conference continued in two tracks, with partners and users talks.

At Mia-Software we had invited Olivier Corbel, the IS manager of BritAir, an airline company, who presented  its model-driven development process.

Britair mdday2010

BritAir uses MagicDraw to create very simple UML class diagrams which represent the domain model of each new application (entities and relations).

This UML model is used to fully generate RCP applications which allow to manipulate the data (CRUD). The generation rules are defined and executed with Mia-Studio.

When some business logic is required, it is defined with a textual DSL (business rules in french) and then automatically translated into Java.

This approach and the corresponding tools are now used each time BritAir has to develop a specific application. The code is 100% generated. For now, six applications have been developped (small maintenance of aircrafts, management of aircraft events, cost accounting, financial forecasts, management of passengers, preparation of flights). Olivier Corbel said that the development managers have evaluated to 10 the productivity ratio on these projects!

After this presentation, I had decided to talk (in french again) about Model-Driven Development and how to reconcile developers with Model-Driven engineering.

Mia-Software at MD Day 2010

In the first part of my presentation I constated that, in 2010, Model-Driven Engineering adoption continues to increase:

• A lot of successful projects have already proven the benefit of Model-Driven Development. In many domains (bank, insurance, defence, aeronautics, transportation, etc) and for various technologies (Java, C#, Flex, Smalltalk, Ada, C, Cobol, etc).
• Model-Driven Engineering is not only dedicated to the development of new applications. It is also used to maintain the existing ones and facilitate their modernization. The MoDisco project is an example of such initiative.
• The offer is very rich! Just have a look at the Eclipse Modeling Project: during the last nine months, there have been 14M lines of code committed on 45 active projects, by 113 committers of 20 different companies ! In addition, standards such as UML2 (even if the specification could be simplified) is now supported by a lot of mature tools.
• The MD Day reflects this adoption with a continuous augmentation (since 2007) of both partners and attendees.

Despite this fact, we must also admit that Model-Driven Engineering still faces strong resistance. 

It is easy to challenge the most current arguments:

• Complexity of UML? UML is complex (even OMG admitted it by launching the UML Specification Simplification RFP), but 20% of UML (class diagram) is sufficient for 80% of needs. Do you really think that a class diagram is more complex than all the web technologies (WSDL/SOAP, Hibernate, Spring, etc) that most developers are familiar with ? And if it's still too complex, you can use a DSL with only your vocabulary: Model-Driven Engineering is not only UML!
• Too much abstractness? MDD implies that the model should be more abstract than the code to develop (this is what makes the model more sustainable than only the source code). But the level of abstraction is free: it's up to you to define the good level of abstraction depending on the maturity of the team.
• Complicated Process? MDD implies two new activities (the production of the model and the generation). But these activities can be integrated to any kind of process (from agile to waterfall). And modeling usually corresponds to the design of the application. For those who still think that design is not compatible with agility, just read principle #9 of Agile Manifesto: "Continuous attention to technical excellence and good design enhances agility".

Nevertheless, one argument againts MDD is a real one: the difficulty to keep in sync the model and the code. If you loose the synchronization, you loose the benefit of MDD. But if you put all the energy of your team in maintaining this synchronization, you loose in agility.

Hopefully there are solutions to avoid this pitfall: 

• Model at runtime: the model is not transformed into code, it is directlty deployed and interpreted by a runtime engine. This approach is similar to declarative languages based on XML, for example to describe GUIs (XUL, XAML, etc). The drawback is that you need to implement a runtime engine.
• Model compilation: the model is transformed into a complete source code, where no additional coding is required (this is the solution adopted by BritAir). This approach is similar to the compilation of a 3GL language (Java, C++, C#, etc) into C or byte code. The drawback is that you need to create comprehensive models and develop complex generators where all the cases must be supported.

But in most cases, you can't describe all your system with a model. Then the generated code has to be completed manually. In this situation, to avoid the desynchronization between the model and the code, a solution consists in applying the Agile principles to modeling and code generation (we can call this approach Agile MDD).

Agile Modeling consists in:

• Useful modeling: only model what is really helpful for the understanding of the system, and what will be exploited by the generator. Avoid modeling any information which is redundant with the code (it would rapidely be out of sync).
• Simple modeling: if the system is complex, the model can be complex, but it should always be more simple than the code.
• Viewpoints modeling: decompose the model into submodels describing viewpoints on the system
• Incremental modeling: the model should be continuously challenged by other people on the project and/or by the generator.

Scott Ambler has defined several other core principles that you can read on http://www.agilemodeling.com/principles.htm

To produce code, these principles should be combined with other Agile principles related to generation. The objective of these Agile Generation is to reduce the gap between the model and the code:

• Targeted generation: you must be able to select only what you need to (re)generate, depending on what has changed (in the model, or in the generator). If each re-generation is global and takes too much time, you may want to delay the modification in the model and directly modify the code.
• Incremental generation: if the generated code has to be completed manually, these modifications must be protected by the next generations (now, all the generators have this functionality with protected areas).
• Traceability: the generator should keep the trace between the model and the code to allow to easily navigate from a file to the corresponding model element.
• Impact analysis: each time you click on the "generate" button, you must be informed of what exactly happened. Which are the new generated files ? Which parts of the existing files have been modified ?
• Assisted coding: after a generation, you should have a direct access to the parts of the code which have to be manually completed.
• Adaptability: the generated code reflects architectural and coding choices which can evolve during the project. Templates engine already allow to easily modify the generator. But to facilitate the modification of the right template, traceability mechanisms should give a direct access from the generated code to the corresponding template definition.

Of course, these Agile MDD principles are not exclusive: they should be combined, when it is possible, with Model at Runtime and Model Compilation approaches.

To illustrate Agile MDD, I ended my talk with a presentation of the Mia-Studio tool suite which supports these Agile Generation principles. 

Then Tony Rostren shown a short demonstration of the tool:

• Modeling of the cinematic of an application in UML with MagicDraw
• Generation of an executable application based on Struts
• Edition of the generated code in Eclipse
• Navigation from Java source code editor to corresponding element in MagicDraw
• Modification of the model and regeneration of the corresponding code
• Access to a template from a selected fragment of generated code

The full demonstration is available here:

http://www.mia-software.com/html/miaStudio/download/demo/mdday2010.htm

I spent the rest of the conference talking with customers and partners and demonstrating our tools. I could only attend one other presentation. It was the Sodius one, with Yann Lebeaupin talking with EADS about tools interoperability to support system engineering. 

You can find all the other presentations on slideshare: http://www.slideshare.net/MDDAY/presentations

Architecture-Driven Modernization Case-Studies

William Ulrich and Philip Newcomb have recently published "Information Systems Transformation, Architecture-Driven Modernization Case-Studies", a reference book on Software Modernization.

This book has been written by two of the most active members of the Architecture-Driven Modernization task force, the OMG initiative which aims at defining standard specifications for the modernization of existing software systems. They first introduce Architecture-Driven Modernization technologies, standards and approaches. Then they have compiled ten detailled case-studies on real modernization projects on various business domains (bank, administration, tourism, air-trafic management, combat system) and technologies (COBOL, VB6, Powerbuilder, ...).

With Gabriel Barbier (Mia-Software), Yves Lennon (Sodifrance/SoftMaint) , Hugo Brunelière and Frédéric Jouault (INRIA/AtlanMod) we have written one of these case-studies.

In this chapter we first describe the modernization process and tools used by Sodifrance to migrate Software Systems. The approach has been imagined and protyped thanks to a collaboration established in 1993 bewteen Sodifrance and Jean Bezivin, from University of Nantes, who was working on the representation of existing software systems with a technology based on sNets, a first generation of model engineering platform.

The sNets technology was immediately used by Sodifrance to develop a semantic discovery tool, named Semantor, to analyse any COBOL program and provide a fine-grained level of information about its internal structure and data. This tool is still evolving and has been renamed Mia-Mining.

In parallel, at the end of 1998, based on the experience gained in the rebuilding of an insurance company’s contract management system, where modelling and tailored code-generators were successfully used, Sodifrance started developing its own model transformation technology. From this work, in association with Jean Bezivin who brought his knowledge about early work at OMG on MOF, was born Mia-Studio, a model transformation tool to develop and run model-to-model transformation rules and model-to-text generation templates.

With these tools, Sodifrance has progressively built a chain which can be used on architecture migration projects, to transform existing applications from client-server to n-tiers and SOA architectures.

This chain is composed of three main steps:

1. Extraction of a comprehensive model (the initial model) of the existing application from its assets (source code, configuration files, development repositories, etc).
2. Transformation of the model of the existing application into a comprehensive model of the target application (the target model).
3. Generation of the source code of the target application from the model of the target application.

To illustrate how this chain can be used, the chapter describes a project conducted by Sodifrance to migrate an application from VB6 to JEE for Amadeus Hospitality the leader in IT solutions provided for the tourism and travel industry.

The initial application, named RMS (Revenue Management System), was developed in VB6, and was performing queries on an Oracle database. It was composed of 300 screens, 200 of them displaying charts (pie charts, bar charts or line graphs). The VB6 code was composed of 306,000 source lines of code, VB6 code in 216 classes and 261 modules.

The migration project was completed by Sodifrance in 1,600 man-days with ten engineers over a year. The transformation of all the VB6 code (access to data, business rules and interface) was 80% automated, while the definition of the screens (Forms) was only 50% automated, due to the necessity to redesign them for a web mode. The new version of RMS is now composed of about 300,000 lines of code in 1,000 Java classes and 310 Jsp (Java Server Pages).

In the last part of the chapter, we present MoDisco, the Eclipse project dedicated to software modernization. This project has been created by AtlanMod during Modelplex, a research project funded by the European Community.

Because of the widely different nature and technological heterogeneity of legacy systems, there are several different ways to extract models from such systems. MoDisco proposes a generic and extensible metamodel-driven approach to model discovery. A basic framework, including implementations of OMG standards such as KDM or SMM, and a set of guidelines are provided to the Eclipse contributors to bring their own solutions to discover models in a variety of legacy systems.

One of the first industrial use cases using MoDisco has been the understanding of a Large Scale Data Intensive Geological system for WesternGeco, a geophysical services company

JEE, Flex and MDSD in Tunis

Last week I was in Tunis.

The weather was not as sunny as expected, but it was not the reason of my travel. I was there to setup a MDSD (Model-Driven Software Development) process for a tunisian bank.

This bank has planned to redevelop its Core Banking System with new technologies (JEE and Flex). In order to facilitate the development and provide both flexibility and quality to the future system, they have decided to adopt a Model-Driven approach.

My mission consists in helping the IT team to put this approach in place.

Developing a generator for a customer is an activity which I usually decompose in four steps:

Development of a reference application

In every industrialization process, the first step consists in identifying the scope of what can be beindustrialized. In a MDSD process, the better way to define this scope is to manually develop a reference application: a subset of the future application which contains examples of each coding patterns.

During a first stay, few weeks ago, we had specified and started to develop a reference application based on a cash withdrawal scenario. Behind this scenario, we had defined several services to invoke, the corresponding business objects (BO) and data transfer objects (DTO), and the existing Oracle tables and stored procedures which have to be reused. Based on the languages and frameworks selected by the customer (Flex with the CairnGorm framework, JEE with the Spring and Hibernate frameworks), we had designed the reference application and defined the coding patterns to use.

This week, when I began my second stay, the customer had finished the development of the reference application, and it was running.

Identification of the Generation Scope

Once a reference application exists, the second step consists in analysing its source code to identify the variability factor of each line of code:

• What is the minimal information required to be able to produce this line of code?
• Is this information specific to the reference application or generic ?
• Can we produce other lines of code with the same information?
• What is the ratio between the effort to declare this information and the effort to manually produce all the corresponding code?

The answers to these questions, coupled with a discussion with the customer, help defining the generation scope:

• which code can be produced automatically?
• which code needs to be developped manually?

Definition of Modeling Rules

Once the generation scope is identified, then we need to define how the information required to generate the code can be defined within a model. There are three possibilities:

1. Defining a Profile in a UML Modeler
2. Developing a Domain-Specific Modeler
3. Developing a Domain-Specific Concrete Syntax

For my customer in Tunis, I have proposed the first option and defined a UML profile containing a first set of about 20 stereotypes (application, service, bo, vo, dao, table, ...). With this profile I used MagicDraw (which provides very powerful extensibility and customization mechanisms) integrated in Eclipse to create a model of the reference application.

Development of Generation Templates

The fourth step is the easiest: templates can be developped from the reference application by copy/pasting fragments of code. The fragments parts which are generic remain in the templates, while variable parts are replaced by calls to the model (using EMF APIs).

Last week, it tooks me one day to develop the Mia-Studio templates for the presentation layer. From the EMF model of the reference application, the templates have regenerated 6 MXML files and 19 ActionScript files (Commands, Events, Service Delegates, Front Controller, and Value Objects). The MXML files contain the graphical definition of the GUI: they will be generated only once, just to provide a first application which can be executed. Then they will be edited and maintained with a WYSIWYG designer.

The templates for the two other layers (Business and Data) will be developed with a colleague during a third stay in Tunis.

Then our role in the project will be to assist the team in modeling and developing the first application and adapt the MDSD process to integrate the unforseen cases.