AuMi-Tests: Automated Generation of a Regression Test Suite for Libraries by Mining Tests from Open Source Client Projects

Abstract

Regression testing is one of the testing practices used by developers to ensure that their program runs as expected. As the program evolves, it takes more work and effort for developers to update and maintain a regression test suite. Researchers have developed techniques to automate the generation of tests to help reduce that workload. However, the generated tests are often semantically meaningless and do not show the canonical usage of a class under test. In this thesis, we propose AuMi-Tests, a novel technique that generates a regression test suite for a given library by mining developer-written tests from open source projects which use that library. This technique leverages the popularity of open source development in the last decade. Because developers wrote them, the tests from a test suite generated with our approach are likely to be semantically meaningful. They will demonstrate real-world usage of the library to its developers as well. We implemented our technique into a prototype tool and tested it on 22 different subject libraries to evaluate its effectiveness. Our findings showed that a test suite generated for a library through this technique could complement the existing library tests and tests generated through stateof- the-art automated test suite generators for that library to a certain degree.