Measuring the Lifecycle of Web GUI Test Smells: A Repository-Mining Study of End-to-End Test Smell Evolution

Abstract

End-to-end (E2E) Web GUI tests are a critical quality gate for modern web applications, but their own maintainability is rarely measured longitudinally. Existing work has characterized the adoption and maintenance of Web GUI testing frameworks, while evidence on the lifecycle of quality problems inside E2E tests remains limited.

This paper frames Web GUI test smells as measurable maintainability-risk events and studies their incidence, evolution, and socio-technical context. We mine 358 open-source repositories from the E2EGit dataset, covering 3,435 smell-related test files, 29,830 commits, and 567,265 detected smell instances in JavaScript and TypeScript projects using Cypress, Playwright, Puppeteer, and Selenium. We operationalize five families of measures: smell incidence, file-level diffusion, transition type, distance from release dates, and developer ownership. The results show that Global Variable is the most frequent smell overall, with 149,673 instances, while Assertion Roulette is the most widespread, affecting 2,560 distinct files. Smells tend to accumulate more often than they are removed: worsening transitions outnumber improving transitions in both TypeScript and JavaScript histories. Smell-related changes are also release-centered, with 76.4% of TypeScript smell introductions and 62.9% of JavaScript smell introductions occurring within seven days before or after a release. Finally, file owners and core maintainers are deeply involved in both smell introduction and removal, and commit messages suggest that Sleepy Test, Redundant Print, and Unknown Test may reflect short-term maintenance trade-offs. These findings indicate that E2E test smells should be treated not only as static code-quality issues, but as measurable lifecycle phenomena connected to release pressure and maintenance behavior.