Software testing is a fundamental part of software development, but the cost of running tests can be high. As software grows in complexity and test suites get larger, software testing also takes longer to run. Existing approaches to speed up testing such as testsuite reduction or regression test selection aim to run some subset of the test suite, but these approaches run the risk of missing to run some key tests that are needed to detect faults in the code.

We propose a new technique to transform test code to speed up test running time while still running all the tests. The insight is that testing frameworks such as JUnit for Java projects allow for developers to define test fixtures, i.e., methods that run before or after every test to setup or teardown test state, but these test fixtures need not be called all the time before/after each test. It may be sufficient to do the setup and teardown once at the beginning and end, respectively, of all tests. Our technique, TestBoost, transforms the test fixtures within a test class to instead run once before/after all tests in the test class, thereby running the test fixtures less while still running all tests and ensuring that tests all still pass, as they did before. Our evaluation on 697 test classes from 34 projects shows that on average we can reduce the running time per test class by 28.39% for the cases with positive significant improvement. Using these transformed test classes can result in an average 18.24% reduction per test suite running time.