Mutation Analysis of Relational Database Schemas

The schema is the key artefact used to describe the structure of a relational database, specifying how data will be stored and the integrity constraints used to ensure it is valid. It is therefore surprising that to date little work has addressed the problem of schema testing, which aims to identify mistakes in the schema early in software development. Failure to do so may lead to critical faults, which may cause data loss or degradation of data quality, remaining undetected until later when they will prove much more costly to fix. This thesis explores how mutation analysis – a technique commonly used in software testing to evaluate test suite quality – can be applied to evaluate data generated to exercise the integrity constraints of a relational database schema. By injecting faults into the constraints, modelling both faults of omission and commission, this enables the fault-finding capability of test suites generated by different techniques to be compared. This is essential to empirically evaluate further schema testing research, providing a means of assessing the effectiveness of proposed techniques. To mutate the integrity constraints of a schema, a collection of novel mutation operators are proposed and implementation described. These allow an empirical evaluation of an existing data generation approach, demonstrating the effectiveness of the mutation analysis technique and identifying a configuration that killed 94% of mutants on average. Cost-effective algorithms for automatically removing equivalent mutants and other ineffective mutants are then proposed and evaluated, revealing a third of mutation scores to be mutation adequate and reducing time taken by an average of 7%. Finally, the execution cost problem is confronted, with a range of optimisation strategies being applied that consistently improve efficiency, reducing the time taken by several hours in the best case and as high as 99% on average for one DBMS.