JSART: JavaScript assertion-based regression testing

Abstract

Web 2.0 applications rely heavily on JavaScript and client-side runtime manipulation of the DOM tree. One way to provide assurance about the correctness of such highly evolving and dynamic applications is through regression testing. However, JavaScript is loosely typed, dynamic, and notoriously challenging to analyze and test. We propose an automated technique for JavaScript regression testing, which is based on on-the-fly JavaScript source code instrumentation and dynamic analysis to infer invariant assertions. These obtained assertions are injected back into the JavaScript code to uncover regression faults in subsequent revisions of the web application under test. Our approach is implemented in a tool called Jsart. We present our case study conducted on nine open source web applications to evaluate the proposed approach. The results show that our approach is able to effectively generate stable assertions and detect JavaScript regression faults with a high degree of accuracy and minimal performance overhead. © 2012 Springer-Verlag.