以版本歷程與動態執行資訊產生註解文件之自動化系統

Abstract

文件撰寫工作在專案中是必要而不可或缺的過程，但我們卻常忽略其在發展過程中的必要性。對於一個組員時常變動的團隊，在沒有明確註解文件的情形下難以將程式的修改部份加入其系統。本研究期望透過動態分析的技巧以及版本資料庫的內容此二途徑，以揭露專案開發的內隱知識。我們的目標是在不需程式內特殊標註的情形下，自動產生文件(documentation)以增進人員對程式的理解，或者是以此自動產生之文件對原始碼除錯。此文件以程式常用模式(frequent pattern)表示，方法上使用sequential pattern mining演算法於動態歷程(run-time trace)上找出pattern，以出現次數(support)、機率(confidence)描述該pattern，另外使用版本資料庫之資訊，得出程式中每個識別符(identifier)的變動情況(variant)。使用上述三種參數決定pattern的品質，結果可分為三類－Frequent、Potential-Error、Unlikly。開發者可依此文件得到程式撰寫建議，或是提示原始碼中可能出錯的部份，並指出可能的正確撰寫方法。

Documentation is an essential and indispensable part in a software project but we often ignore its importance for the development process. Developers construct the software systems with incomplete comments or features left undocumented; for this reason, it’s hard for a rapid changing team to accommodate changes inherently embedded without explicit annotations. Via the techniques of dynamic analysis and revision history manipulation, we discover then implicit knowledge in the projects development. Our main objective is to generate program documents without explicit specifying related annotations in the program source. The knowledge of program understanding could be also used to debug the source code. The document is represented as program frequent patterns. We apply the sequential pattern algorithm on run-time trace to find out the patterns and describe patterns by occurrence times(support) and probability(confidence). We also process the revision history database to count the change times for each identifier and this value are named after “variant”. Patterns are ranked by the above 3 parameters. We classify the pattern into 3 categories: Frequent, Potential-Error and Unlikly。Upon this document, developers could obtain programming suggestion. In addition, our system could point out the susceptible site in the source and prompt the possibly correct coding style.