Speeding From Sketchpad To Smooth Code
Creating error-free software remains time consuming and labor intensive. A major European research effort has developed a system that speeds software development from the drawing board to high-quality, platform-independent code.
According to Piotr Habela, technical coordinator of the VIDE (for VIsualize all moDel drivEn programming) project, software developers have many good ideas about how to visualize, develop, debug and modify software, plus standards to guide them. The problem is that the design and development process has always been fragmented.
He explains that methods for visualizing or flowcharting how a program should work do not lead directly to computer code.
Software written in one programming language may be difficult to translate into another. No matter how carefully programmers work, complex software almost always includes errors that are difficult to diagnose and fix. Because of the lack of precise links between a program’s features and the software that implements them, updating or modifying a program often turns out to be time-consuming and costly.
“What we attempted that was quite distinct,” says Habela, “was to make the development of executable software a single process, a single tool-chain, rather than a sequence of separate activities.”
It took two-and-a-half years of intensive effort by VIDE’s ten academic and industrial research partners, funded by the European Union, but the result is a software design and development toolkit that promises to make creating well-functioning, easily-modified software ““ for example for small businesses ““ significantly smoother, faster, and less expensive.
Model driven architecture
A key part of VIDE’s approach was to build on the idea of Model Driven Architecture, a programming methodology developed by an international consortium, the Object Management Group.
The idea is that each stage of software development requires its own formal model. The VIDE team realized that by creating and linking those models in a rigorous way, they could automate many of the steps of software development.
A software developer might start by working with a domain expert ““ for example a business owner ““ to determine what a new program needs to do. Those inputs, outputs and procedures would be formalized in what is called a computation independent model (CIM), a model that does not specify what kinds of computation might be used to carry it out ““ it lays out what the program will do rather than how it will do it.
“Models are usually considered just documents,” says Habela. “Our goal was to make the models serve as production tools.”
In the case of VIDE, much of that modeling is visual, in the form of flowcharts and other diagrams that are intuitive enough for the domain expert to understand, but which are sufficiently formalized to serve as the inputs to the next stage of the software development process.
To carry out these first modeling steps, the researchers created a domain analysis tool and a programming language called VCLL, for VIDE CIM Level Language.
From CIM to PIM to program
Once they have produced a formal CIM of the program they want to implement, it’s time to move a step closer to a functioning program by translating it into a platform independent model, or PIM.
For the VIDE team, a PIM is a model that specifies precisely what a program needs to do, but at an abstract level that does not depend on any particular programming language.
The researchers developed several software tools to produce a usable, error-free PIM. These include an executable modelling language and environment, a defect-detection tool, and finally a program that translates their final model into an executable Java program.
Luckily, the researchers did not have to build their system from the ground up. They were able to rely to a large extent on a pre-existing modeling language called UML, for Unified Modeling Language. UML provides a systematic way to visualize and describe a software system.
“We now have a kind of prototyping capability built into the development process,” says Habela. “You can design a model, specify its behavioral details, run it with sample data to see how it behaves, and then check with the domain expert to see if it is in fact the behavior they expected.”
Several of the consortium members are implementing the VIDE toolkit in specific areas, for example web services, database management, and a variety of business processes.
Habela cautions that reaching VIDE’s goal of smoothly automating the entire software design and development process requires more work. Because of the broad scope of the project and the fundamental changes they are making, they are not yet ready to deploy the complete system.
However, he says, they have gone a long way towards clearing up “the muddy path from requirements to design.”
The VIDE project received funding from the ICT strand of the EU’s Sixth Framework Program for research.
On The Net: