| Program Understanding (PU) or Program Comprehension (PC) is often viewed as the process of recognizing and extracting system design concepts from the source code. This activity takes place throughout the whole lifecycle of software systems. It's widely accepted that software maintenance spent about 35～80% cost of software systems inside which about 30～90% cost is spent on PU including program maintenance, debugging, framework reconstructing, source code reusing, error checking and document coordinating. PU and its related technologies have been developed and researched actively.There are three main strategies in PU: Top-down, Bottom-up and their hybrid approach. Bottom-up strategy is like building a pyramid: it constructs upper-level concepts from basic components step by step until the whole concept of selected program is achieved. Almost all automatic PU methods apply Bottom-up strategy and are based on Plans which is the template of program. Although this strategy is feasible in theory, it really encounters two problems: 1) the construction and organization of Plan library; 2) the efficiency of exhaustive matching algorithms.In this paper we focus our research on CS-based PU. By applying characteristic symbols (CS) in PU domain we construct our CS-based Plan matching framework to solve the two main problems listed above by: 1) constructing hierarchy and constraint-oriented CS-Plan library; 2) applying layered matching mechanism to improve the efficiency of exhaustive searching; 3) applying comparability analysis on the results to ensure the veracity while using CS as the matching agent.Our chief innovations include: 1) A layered matching algorithm is proposed which includes stages of rough-matching and refined-matching. Rough-matching is used to reduce the response time while searching on total Plan library. Refined-matching is used to reduce the scope of possible-fit Plan set by applying more characteristic symbols; 2) A comparability analysis algorithm based on fuzzy...
|
PDF Full Text Request