Transferring Context-Dependent Test Inputs
André Reichstaller, Alexander Knapp · 2017
We consider the question of how to treat existing, context-based test inputs when contextual conditions change. Simply ignoring the voided inputs reduces confidence in the correctness of the system under test (SuT). Instead, we suggest to adjust the parameters of those inputs to the new conditions in a way that retains their original intention. This often comprises behavioral assumptions, e.g., because of coverage or risk considerations. Transferred test inputs should consequently trigger similar behavior of the SuT within the new environment as the original ones did in the old. We formalize this claim by a distance function on test inputs which compares the expected reactions of the SuT. The more similar the responses, the closer the test inputs. The proposed metric can thus be used for guiding test input transfer. In addition to a recursive definition, we present an algorithm that utilizes neural models to estimate the metric by simply observing a given simulation which sketches the intended behavior of the SuT. As this approach seems to specifically match the prerequisites when testing proactive systems, motivation and first experiments consider a simplified instance of those: an exemplary smart vacuum system.