| Achieving complete testing of software is infeasible. Both behavioral (also termed as black box or functional) testing and code-based (also termed as white box or structural) testing encounter the infinite characteristics of the process of testing. Testing all input combinations or all code paths in a program is almost always impossible, because programs almost always contain an infinite number of input combinations and paths. Thus, a practical path selection criterion can specify only a finite subset of a program's paths and inputs to be exercised during testing. As a result, researchers have focused on constructing structural groups of criteria, which exercise the various statements in a program. A number of path selection criteria called test data adequacy criteria have been proposed which are based on the source code of a program. However, testing every statement of the source code does not guarantee that software is free of faults and failures. On the other hand, behavioral testing was less popular in the past. The main reason of its low popularity was the absence of a model that can represent it compared to the code-based testing which uses a program graph as its model. Recently, a finite-state discrete parameter Markov Chain was introduced to represent the behavior model of software. Markov chains are analytically very powerful and have been successfully used in practice to model input populations for software.; This dissertation shows new criteria based on the structure of external program behavior. Such criteria are called Markov chain-based test data adequacy criteria. This new family of test data adequacy criteria is constructed to traverse the Markov chain behavioral model of software under test. Taking inspiration from white box data-flow criteria, it is designed to focuses on data oriented aspects of the model, instead of model structure. In the case of Markov Chain, the data of interest is the operational modes that define states of the model.; In order to provide proof-of-concept, laboratory experiments have been conducted in which a software system with known faults was tested by structurally-identical Markov chains using three different test selection strategies: random path, structure-guided path and path selection guided by these new criteria. Results have shown that testing by this new family of test data adequacy criteria outperforms all other types of testing strategies in both the number of faults uncovered and time needed for testing. Additionally, these positive findings came out of the laboratory experiments have confirmed the theoretical proof-of-concept which has been provided. |
