Automatic Refactoring for Renamed Clones in Test Code

Abstract

Unit testing plays an essential role in software development and maintenance, especially in Test-Driven Development. Conventional unit tests, which have no input parameters, often exercise similar scenarios with small variations to achieve acceptable coverage, which often results in duplicated code in test suites. Test code duplication hinders comprehension of test cases and maintenance of test suites. Test refactoring is a potential tool for developers to use to control technical debt arising due to test cloning. In this thesis, we present a novel tool, JTestParametrizer, for automatically refactoring method-scope renamed clones in test suites. We propose three levels of refactoring to parameterize type, data, and behaviour differences in clone pairs. Our technique works at the Abstract Syntax Tree level by extracting a parameterized template utility method and instantiating it with appropriate parameter values. We applied our technique to 5 open-source Java benchmark projects and conducted an empirical study on our results. Our technique examined 14,431 test methods in our benchmark projects and identified 415 renamed clone pairs as effective candidates for refactoring. On average, 65% of the effective candidates (268 clone pairs) in our test suites are refactorable using our technique. All of the refactored test methods are compilable, and 94% of them pass when executed as tests. We believe that our proposed refactorings generally improve code conciseness, reduce the amount of duplication, and make test suites easier to maintain and extend.