Generating meaningful test databases

Testing is one of the most time-consuming and cost-intensive tasks in software development projects today. A recent report of the NIST [RTI02] estimated the costs for the economy of the Unites States of America caused by software errors in the year 2000 to range from $22.2 to $59.5 billion. Consequently, in the past few years, many techniques and tools have been developed to reduce the high testing costs. Many of these techniques and tools are devoted to automate various testing tasks (e.g., test case generation, test case execution, and test result checking). However, almost no research work has been carried out to automate the testing of database applications (e.g., an E-Shop application) and relational database management systems (DBMSs). The testing of a database application and of a DBMS requires different solutions because the application logic of a database application or of a DBMS strongly depends on the contents of the database (i.e., the database state). Consequently, when testing database applications or DBMSs new problems arise compared to traditional software testing. This thesis focuses on a specific problem: the test database generation. The test database generation is a crucial task in the functional testing of a database application and in the testing of a DBMS (also called test object further on). In order to test a certain behavior of the test object, we need to generate one or more test databases which are adequate for a given set of test cases. Currently, a number of academic and commercial test database generation tools are available. However, most of these generators are general-purpose solutions which create the test databases independently from the test cases that are to be executed on the test object. Hence, the generated test databases often do not comprise the necessary data characteristics to enable the execution of all test cases. In this thesis we present two innovative techniques (Reverse Query Processing and Symbolic Query Processing), which tackle this problem for different applications (i.e, the functional testing of database applications and DBMSs). The idea is to let the user specify the constraints on the test database individually for each test case in an explicit way. These constraints are then used directly to generate one or more test databases which exactly meet the needs of the test cases that are to be executed on the test object.