Program-input generation for testing database applications using existing database states

Abstract

Testing is essential for quality assurance of database applications. Achieving high code coverage of the database applications is important in testing. In practice, there may exist a copy of live databases that can be used for database application testing. Using an existing database state is desirable since it tends to be representative of real-world objects' characteristics, helping detect faults that could cause failures in real-world settings. However, to cover a specific program-code portion (e.g., block), appropriate program inputs also need to be generated for the given existing database state. To address this issue, in this paper, we propose a novel approach that generates program inputs for achieving high code coverage of a database application, given an existing database state. Our approach uses symbolic execution to track how program inputs are transformed before appearing in the executed SQL queries and how the constraints on query results affect the application's execution. One significant challenge in our problem context is the gap between program-input constraints derived from the program and from the given existing database state; satisfying both types of constraints is needed to cover a specific program-code portion. Our approach includes novel query formulation to bridge this gap. We incorporate the data-instantiation component in our framework to deal with the case that no effective program input values can be attained. We determine how to generate new records and populate them in the new database state such that the code along the path can be covered. We also extend our approach of program-input generation to test database applications including canonical queries and group-by queries. Our approach is loosely integrated into Pex, a state-of-the-art white-box testing tool for .NET from Microsoft Research. Empirical evaluations on two real database applications show that our approach assists Pex to generate program inputs that achieve higher code coverage than the program inputs generated by Pex without our approach's assistance.