Mining Executable Specifications of Web Applications from Selenium IDE Tests

A common practice for system testing of web-based applications is to perform the test cases through a web browser. These tests are often recorded and managed by a record and replay tool, such as Selenium IDE. Mining specifications from such tests can be very useful for understanding, verifying, and debugging the system under test. This paper presents an approach to mining a behavior specification from a Selenium IDE test suite such that (a) it captures the behavior of the tests at a high level of abstraction, (b) the behavior can be simulated, and (c) all the tests are completely reproducible from the specification. We first identify similar test actions through context-sensitive clustering so as to normalize the given Selenium IDE tests. Then, we mine patterns of test actions that represent meaningful functions and transform Selenium IDE tests into abstract tests, which are similar to the tests used in the existing model-mining techniques. From the abstract tests, we synthesize a high-level Petri net that captures both temporal constraints and data values. For evaluation purposes, we applied our approach to eight test suites of two real-world systems, Magento (an online shopping system being used by many live stores) and Amazon. Two of the test suites are for security testing, aiming at SQL injection and XSS vulnerabilities. The result shows that our approach is effective in producing abstract yet executable specifications and reducing the complexity of the models.