To execute system tests, two fundamentally different execution techniques exist: manual and automated execution. For each system test suite, one must decide how to employ those techniques (this strategy is called execution mode). Despite general conditions such as fixed testing strategies or development philosophies, almost all projects permit a wide range of possible execution modes to choose from. In industry, execution techniques are often chosen by experts based on rules of thumb, experience and best practices. Although the results are mostly tolerable, they may be not cost-effective. In retrospect, it is often unclear on what basis those decisions were made, making it difficult to assess whether they are still valid. Finally, it is hard to predict the costs for test execution beforehand. We introduce a cost model to estimate the economic impact of execution modes. Our cost model is based on expert estimations and gives additional input for testing experts in balancing pros and cons of execution modes at hand. Furthermore, it helps documenting and persists decisions during the life time of a test suite. Additionally, we report on a first case study, applying our cost model in industry.

Tests are central artifacts of software systems and play a crucial role for software quality. In system testing, a lot of test execution is performed manually using tests in natural language. However, those test cases are often poorly written without best practices in mind. This leads to tests which are not maintainable, hard to understand and inefficient to execute. For source code and unit tests, so called code smells and test smells have been established as indicators to identify poorly written code. We apply the idea of smells to natural language tests by defining a set of common Natural Language Test Smells (NLTS). Furthermore, we report on an empirical study analyzing the extent in more than 2800 tests of seven industrial test suites.

Software systems contain unnecessary code. Its maintenance causes unnecessary costs. We present tool-support that employs dynamic analysis of deployed software to detect unused code as an approximation of unnecessary code, and static analysis to reveal its changes during maintenance. We present a case study on maintenance of unused code in an industrial software system over the course of two years. It quantifies the amount of code that is unused, the amount of maintenance activity that went into it and makes the potential benefit of tool support explicit, which informs maintainers that are about to modify unused code.

Tests are central artifacts of software systems. Therefore, understanding tests is essential for activities such as maintenance, test automation, and efficient execution. Redundancies in tests may significantly decrease their understandability. Clone detection is a technique to find similar parts in software artifacts. We suggest using this technique to gain a better understanding of tests and to provide guidance for testing activities. We show the capabilities as well as the limits of this approach by conducting a case study analyzing more than 4000 tests of seven industrial software systems.

Traditionally, most work in program comprehension focuses on understanding the inner workings of software systems. However, for many software maintenance tasks, not only a sound understanding of a system's implementation but also comprehensive and accurate information about the way users actually use a system's features is of crucial importance. Such information e.g. helps to determine the impact that a specific change has on the users of a system. In practice, however, this information is often not available. We propose an approach called feature profiling as a means to efficiently gather usage information to support maintenance tasks that affect the user interface of a software system. Furthermore, we present tool support for feature profiling and report on a case study in the insurance domain. In this study, we profiled the features of an application that is used by 150 users in 10 countries over a period of five months.