Know your neighbor: fast static prediction of test flakiness

Flaky tests plague regression testing in Continuous Integration environments by slowing down change releases, wasting development effort, and also eroding testers trust in the test process. We present FLAST, the rst static approach to akiness detection using test code similarity. Our extensive evaluation on 24 projects taken from repositories used in three previous studies showed that FLAST can identify aky tests with up to 0.98 Median and 0.92 Mean precision. For six of those projects it could already yield 0.98 average precision values with a training set containing less than 100 tests. Besides, where known aky tests are classied according to their causes, the same approach can also predict a aky test category with alike precision values. The cost of the approach is negligible: the average train time over a dataset of 1,700 test methods is less than one second, while the average prediction time for a new test is less than one millisecond.