On supporting software evolution: Automatic decomposition schemes based on static and dynamic analysis of programs.

摘要

Program decomposition is a divide-and-conquer strategy that can be used to deal with the complexities of maintaining multimodule programs. The objective of program decomposition is to identify the components of a program which reflect only a part of the semantics of the original program. Programming activities associated with software evolution (maintenance) such as comprehension, testing, debugging, and verification can be simplified if syntactic and semantic decomposition of programs are made available to software engineers.;A unified scheme for syntactic and semantic decomposition is provided. Visibility flow graphs are introduced for representing the information needed for facilitating syntactic decomposition. This formalism provides a set theoretic representation of the visibility, requirement, and definitional attributes of a multimodule program. Declarative contexts, composed of definitional and/or requirement units, constitute the nodes of the graph, while the arcs represent the flow of visibility.;Dependence relations are introduced to represent dependences among the input and output values of variables and the statements comprising a program. The dependence relations are used in producing program projections, a type of semantic program decomposition. Static analysis is used to generate these relations for primitives such as assignment, conditional, and loop statements which are then combined to produce the relations for the entire program. The results are then extended to handle multilevel exit loops, return statements, procedure calls, limited types of recursions, and arbitrary control flow arising from gotos and exceptions. Dynamic analysis is used for generating more precise dependence relations for specific program executions.;An interactive program decomposition tool called AdaAn has been built in a UNIX/C environment with a subset of Ada as the target language. AdaAn provides direct support for program comprehension and debugging activities. A novel method, based on automatic program decomposition, for supporting the maintenance of program proofs and tests is also presented. Individual program projections with respect to output variables are verified or tested during program development. After any modification, only those projections that are either new or modified need to be reverified or retested.