Software Factory versus MDA

En TheServerSide.net existe una discusión en curso de interés sobre el valor del estándar de Arquitectura conducida por Modelado (MDA). Sólo en varios días podré acotar opiniones sobre el conjunto de la discusión, particularmente en cuanto al "versus", la factoría. Por el momento, quisiera solamente remarcar algunos aspectos señalados por Jack Greenfield, como puntos débiles del estándar. Jack no avala en general MDA, aportando argumentos que en mi criterio no veo como opuestos. Prometo releer varias veces sus razones, hasta encontrar por qué algunos de sus puntos pudieran considerarse "contradictorios". Por ejemplo, Jack cuestiona la idea de independencia de la plataforma:

The primary stated goal of MDA is platform independence. While some insulation from platform technology changes is no doubt achievable, I do not believe it is feasible to produce secure, usable, reliable, performing and supportable software from specifications that take no platform dependencies of any kind into account

Sin embargo, esta afirmación parece ignorar la existencia de dos niveles de modelado, el Platform Independent Model (PIM), y el Platform Specific Model (PSM). Creo que es cuestión de tiempo el que un PSM pueda generar código suficientemente específico para una plataforma dada, como para satisfacer los reclamos de Greenfield acerca de la necesidad de especificar el código. Pero, por si lo leí mal, lo releeré diez veces en estos días...
De todas formas, más allá de estas críticas, que releeré, están sus observaciones, que podrían tomarse como una lista de tareas a realizar:
Relacionado con el ciclo de vida:

• MDA says nothing about how models should be integrated with requirements, architecture, frameworks, patterns, code, configuration files, and other key development artifacts in the context of a development process. Surely, the failures of CASE in the 80s and 90s should tell us that genuinely seamless integration between code and models is an absolute requirement for success in model driven development. Any methodology that claims to explain how models should be used to develop software must prescribe robust solutions at the critical interfaces between models and other mediums of specification and implementation.
• MDA says nothing about how models or metadata derived from models should be used to automate specific tasks across the software life cycle, such as debugging, testing, deployment, version control and configuration management, operations and maintenance. Surely, these activities are as important, if not more so, than generating implementations? To be effective, a model driven development methodology should describe a holistic approach to automating the rest of the software life cycle.

Sobre el UML:

According to the OMG's official definition of MDA, MDA is a way to develop applications in UML. (There are two other competing but unofficial definitions of MDA, as described in a blog posting by Steve Cook, one that advocates the use of MOF based languages, not UML, as the medium of model driven development, and one that advocates the use of UML as a first class programming language that is compiled directly into executables. Of the three definitions, the MOF based one is the closest to the approach that we have taken with software factories and DSLs.). This is a large topic that I will take up in subsequent blog postings, even though it is already covered at length in the book. The point, for those who can’t wait, is that as a methodology for developing software in UML, the effectiveness of MDA is limited to the effectiveness of the UML as a language for model driven development, a task it was not designed to support.

Este asunto continuará...