Lambda the Ultimate

Building distributed applications involves many different concerns, for example application functionality, distribution structure, tolerance for partial failure, and security. Instead of trying to build a generic tool that could handle all these concerns, we will treat each concern in depth (solve it completely) before going to the next. We consider each aspect as a partial specification, and the full specification as the conjunction of all these partial specifications. The crucial property is independence: each new partial specification that is added should not invalidate the previous ones. We have been following this approach for almost a decade now in our work on the Oz language and the Mozart system. In this talk I will give some examples of how it works for real applications and then conclude with some lessons we have learned.

A first lesson is that language design is an important part of the solution: keeping independence imposes strong conditions on the language. A second lesson is that to make each aspect work in realistic conditions, we have to change the application architecture. Simple source transformations are not enough. In hindsight this is obvious, since we have no automatic technique for going from specification to program. A final lesson is regarding the limits of aspect-oriented programming. Even with the best possible AOP, complexity increases because of the large number of aspects, each of which requires changes in the application architecture. At some point we need to go beyond AOP. We propose a simple and time-honored solution: to find "good enough" solutions for certain aspects and then to hide them inside abstraction layers. We finish the talk by summarizing what we have realized in Oz and Mozart and outlining the large amount of work that still has to be done.
Posted to Software-Eng by Peter Van Roy on 5/15/04; 2:10:56 PM