A systems approach to automated object-oriented integration testing

Integration testing of embedded systems developed using object-oriented technology can be a difficult task due to complex interactions among objects, the distribution of system state information among many objects, and encapsulation that prevents easy access to the state of the system. This thesis describes a method and an architecture providing automated support for validating domains (well-bounded software components) during integration testing. The method compares the behavior of the domain against an oracle, a requirements-level model expressed in an executable formal methods language (LOTOS).; The domains are developed using object-oriented techniques and are expressed using the Unified Modeling Language (UML), an industry standard notation. The models use well-defined executable semantics and allow code generators to produce the implementation code; other design languages could be substituted for UML if they meet the requirements for executable semantics. A test architecture is defined within which code generators can insert test instrumentation code to enable integration testing at the UML model level. The test code allows the internal state of the software to be examined directly for correctness without depending upon a distinguishing sequence of external stimulus to propagate the error to the output. The architecture provides a generic test harness in which test case execution and data collection can be automated, which in turn provides the required test information to the oracle.; UML models serve as the basis for implementation through code generation, and another model, the oracle, serves as the basis for test result validation. The oracle is a requirements-level model of the domain, expressed using formal methods, which describes its behavior in response to external stimuli. An specification models, allowing the internal state of the system to be included as part of the test results with the input-output relationship. Testing the internal system state reduces test generation effort because it does not depend on a distinguishing sequence of inputs to trigger and propagate an error to the system output.