Fine-Grain Noise-Injection Heuristics for SearchBestie Infrastructure

The main topic of this thesis is testing of concurrent Java programs. Concurrency errors often manifest rarely. To increase a chance to spot such errors, a technique called noise injection can be used. This technique inserts extra instruction into the application under test to disturb the scheduler and thus to explore less common thread scheduling. Because noise injection can be parameterized in many ways, a tool called SearchBestie was created to handle noise-based testing as a search problem. SearchBestie used a tool called Contest for bytecode instrumentation and test execution. However, the development of ConTest has been discontinued and therefore an alternative tool was necessary. In the first part of this thesis, SearchBestie was connected with a tool called RoadRunner, which replaced ConTest. Afterwards new fine-grain noise-injection heuristics have been proposed and implemented. Experiments proved that these heuristics achieve better results in most cases