| Data coverage testing employs automated test generation to systematically generate increasing test data, set size. The research goal of data coverage testing is to define and prove for a program under test that there exists a sufficiently large N associated with a data set of the program such that testing with a data set larger than N does not reveal any new faults. The current research is applied to a set of C++ Standard Template Library (STL) programs, sometimes called collection programs. The combined information from both program specification and program implementation are used to establish the value of N for a given program.; A number of experiments have been conducted that confirm this theoretical finding, that is, no more faults are detected after N is reached. These experiments validate the theoretical analysis for data coverage, confirming the predicted sufficiently large N for each program. An experimental comparison of testing efficiency is also conducted for data coverage, statement coverage and random testing. Hypothesis testing shows that data coverage has a statistically significant larger probability of detecting a fault than statement coverage. In a large program, data coverage has a statistically significant larger probability to detect a fault than random testing, and also requires fewer test inputs than random testing.; A new area of research called specification-based data coverage is introduced. In this case, data coverage testing is generated only from the specification. Additional assumptions are needed for a program to be tested in this way. A general methodology is given to demonstrate that data coverage testing can be applied to test any program put forward, claiming to meet the specification. |
