A Framework for Detecting and Diagnosing Configuration Faults in Web Applications

Abstract

Software portability is a key concern when target operational environments are highly configurable; variations in configuration settings can significantly impact software correctness. While portability is key for a wide range of software types, it is a significant challenge in web application development. The client configuration used to navigate and interact with web content is known to be an important factor in the subsequent quality of deployed web applications. With the widespread use of diverse, heterogeneous web client configurations, the results of web application deployment can vary unpredictably among users. Given existing approaches and limited development resources, attempting to develop web applications that are viewable, functional, and portable for the vast web configuration space is a significant undertaking. As a result, faults that only surface in precise configurations, termed configuration faults, have the potential to escape detection until web applications are fielded.This dissertation presents an automated, model-based framework that uses static analysis to detect and diagnose web configuration faults. This approach overcomes the limitations of current techniques by featuring an extensible model of the configuration space that enables efficient portability analysis across the vast array of client environments. The basic idea behind this approach is that source code fragments (i.e., HTML tags and CSS rules) embedded in web application source code adversely impact portability of web applications when they are unsupported in target client configurations; without proper support, the source code is either processed incorrectly or ignored, resulting in configuration faults. Using static analysis, configuration fault detection is performed by applying a model of the web application source against knowledge of support criteria; any unsupported source code detected is considered an index to potential configuration faults. In the effort to fully exploit this approach, improve practicality, and maximize fault detection efficiency, manual and automated approaches to knowledge acquisition have been implemented, variations of web application and client support knowledge models have been investigated, and visualization of configuration fault detection results has been explored. To optimize the automated acquisition of support knowledge, alternate learning strategies have been empirically investigated and provisions for capturing tag interaction have been integrated into the process.