Unwoven Aspect Analysis

Various languages and tools supporting advanced separation of concerns (such as aspect-oriented programming) provide a software developer with the ability to separate functional and non-functional programmatic intentions. Once these separate pieces of the software have been specified, the tools automatically handle interaction points between separate modules, relieving the developer of this chore and permitting more understandable, maintainable code. Many approaches have left traditional compiler analysis and optimization until after the composition has been performed; unfortunately, analyses performed after composition cannot make use of the logical separation present in the original program. Further, for modular systems that can be configured with different sets of features, testing under every possible combination of features may be necessary and time-consuming to avoid bugs in production software. To solve this testing problem, we investigate a feature-aware compiler analysis that runs during composition and discovers features strongly independent of each other. When the their independence can be judged, the number of feature combinations that must be separately tested can be reduced. We develop this approach and discuss our implementation. We look forward to future programming languages in two ways: we implement solutions to problems that are conceptually aspect-oriented but for which current aspect languages and tools fail. We study these cases and consider what language designs might provide even more information to a compiler. We describe some features that such a future language might have, based on our observations of current language deficiencies and our experience with compilers for these languages