Inferential checking for mutants modifying database states

Several recent research projects have developed new testing techniques targeted to database application programs, a widely used class of programs with some unique characteristics. Mutation testing, focused on interactions between an application and a database, presents a promising approach to evaluating test adequacy and to evaluating the effectiveness of other test generation and test adequacy criteria. Prior work on database application mutation techniques has focused exclusively on SELECT statements, which extract data without modifying the database state. Extending the techniques to updating statements, INSERT, DELETE, and UPDATE, is challenging because it's necessary to compare the behavior of mutants when executed against the same database state, but each mutant execution changes the database state. This paper presents inferential checking, a new technique for determining whether mutants of an updating statement are killed, without actually executing the mutants. Queries are automatically derived from updating statements and their mutants, characterizing the state change that would occur if the statement were executed. Results of these queries are compared to determine which mutants are killed. The paper presents the technique, its implementation for Java/JDBC applications, and empirical evaluation demonstrating that it is efficient, precise, and useful