FlakiMe: Laboratory-Controlled Test Flakiness Impact Assessment. A Case Study on Mutation Testing and Program Repair

TL;DR

This paper introduces FlakiMe, a platform for controlled assessment of test flakiness, revealing its significant impact on mutation testing and program repair, with insights into mitigation strategies.

Contribution

We present FlakiMe, a novel laboratory platform enabling controlled study of test flakiness effects on testing and repair techniques, providing new insights and mitigation approaches.

Findings

5% flakiness affects mutation score modestly (2-4%)

Flakiness can completely hinder program repair in 50% of cases

Minimal user feedback can reduce flakiness impact

Abstract

Much research on software testing makes an implicit assumption that test failures are deterministic such that they always witness the presence of the same defects. However, this assumption is not always true because some test failures are due to so-called flaky tests, i.e., tests with non-deterministic outcomes. Unfortunately, flaky tests have major implications for testing and test-dependent activities such as mutation testing and automated program repair. To deal with this issue, we introduce a test flakiness assessment and experimentation platform, called FlakiMe, that supports the seeding of a (controllable) degree of flakiness into the behaviour of a given test suite. Thereby, FlakiMe equips researchers with ways to investigate the impact of test flakiness on their techniques under laboratory-controlled conditions. We use FlakiME to report results and insights from case studies that assesses the impact of flakiness on mutation testing and program repair. These results indicate that a 5% of flakiness failures is enough to affect the mutation score, but the effect size is modest (2% - 4% ), while it completely annihilates the ability of program repair to patch 50% of the subject programs. We also observe that flakiness has case-specific effects, which mainly disrupts the repair of bugs that are covered by many tests. Moreover, we find that a minimal amount of user feedback is sufficient for alleviating the effects of flakiness.