Uma abordagem para testar implementações de refatoramentos estruturais e comportamentais de programas C.

Abstract

Refactoring a program is the act of changing its structure in order to improve any architectural aspect, without changing its observable behavior. Developers use tools like Eclipse and NetBeans to help refactoring their programs. These tools implement a set of conditions to ensure that the transformations performed do not change the observable behavior of the program. However, it is not easy to identify ali the necessary conditions for a refactoring to be correct due to the semantic complexity of the languages, mainly because the languages are not formally proved and neither the refactorings. In practice, developers often use collections of tests to assess the correctness of their implementations. However, there is no evidence of having systematic process in the definition of this collection. In this work, we propose a technique to test refactoring implementations of C programs. It consists of five phases. First, we generate automatically several C programs to be refactored from CDOLLY, which has been proposed based on a specification in Alloy. Then, we generate automatically a collection of unit tests of the programs via CATESTER. We then apply the refactoring through CREXECUTOR on the generated programs using the tool to be tested. After this, we execute the test collection in the refactored programs and finally we classify the compilation errors and behavioral changes in bugs. We evaluated our technique in the implementation of eight refactorings of Eclipse CDT. These refactorings not only modify the program structure, but also the body functions, such as the refactoring Extract Function. We detected 41 bugs, which introduce compilation errors in the resulting program, and six bugs related to behavioral changes. We analyzed manually these bugs in refactoring implementations of NetBeans, Xrefactory and OpenRefactory and identified 29 bugs in these tools.