Test coverage of impacted code elements for detecting refactoring faults: An exploratory study

- An analysis on test suites' capacity of revealing refactoring faults.
- Identification of which code elements are most likely to reveal refactoring faults.
- Guidelines to help developers to decide when it is safe to perform a refactoring.
- A model for predicting a suite's likelihood of detecting refactoring faults.

Refactoring validation by testing is critical for quality in agile development. However, this activity may be misleading when a test suite is insufficiently robust for revealing faults. Particularly, refactoring faults can be tricky and difficult to detect. Coverage analysis is a standard practice to evaluate fault detection capability of test suites. However, there is usually a low correlation between coverage and fault detection. In this paper, we present an exploratory study on the use of coverage data of mostly impacted code elements to identify shortcomings in a test suite. We consider three real open source projects and their original test suites. The results show that a test suite not directly calling the refactored method and/or its callers increases the chance of missing the fault. Additional analysis of branch coverage on test cases shows that there are higher chances of detecting a refactoring fault when branch coverage is high. These results give evidence that a combination of impact analysis with branch coverage could be highly effective in detecting faults introduced by refactoring edits. Furthermore, we propose a statistic model that evidences the correlation of coverage over certain code elements and the suite's capability of revealing refactoring faults.