🤖 AI Summary
Modeling quiescence in model-based testing (MBT) typically relies on complex timed automata (TA), hindering scalability and practical adoption.
Method: This paper proposes a lightweight temporal extension to labeled transition systems (LTS): introducing a single global clock and a user-configurable timeout threshold $M$, together with a formal $chi^M$ lifting operator to construct an equivalent timed automaton.
Contribution/Results: The approach formally captures quiescence while preserving LTS simplicity and ensuring strict equivalence between ioco and tioco$_M$ conformance relations. We prove three key properties: (1) semantic conformance equivalence; (2) order independence (commutativity) of test case generation; and (3) consistency of test verdicts. By providing a concise, reliable, and formally verifiable foundation for timeout-based quiescence detection, the method bridges a critical gap between theory and industrial practice in MBT.
📝 Abstract
Model-based testing (MBT) derives test suites from a behavioural specification of the system under test. In practice, engineers favour simple models, such as labelled transition systems (LTSs). However, to deal with quiescence - the absence of observable output - in practice, a time-out needs to be set to conclude observation of quiescence. Timed MBT exists, but it typically relies on the full arsenal of timed automata (TA).
We present a lifting operator $χ^{scriptstyle M}!$ that adds timing without the TA overhead: given an LTS, $χ^{scriptstyle M}!$ introduces a single clock for a user chosen time bound $M>0$ to declare quiescence. In the timed automaton, the clock is used to model that outputs should happen before the clock reaches value $M$, while quiescence occurs exactly at time $M$. This way we provide a formal basis for the industrial practice of choosing a time-out to conclude quiescence. Our contributions are threefold: (1) an implementation conforms under $mathbf{ioco}$ if and only if its lifted version conforms under timed $mathbf{tioco_M}$ (2) applying $χ^{scriptstyle M}!$ before or after the standard $mathbf{ioco}$ test-generation algorithm yields the same set of tests, and (3) the lifted TA test suite and the original LTS test suite deliver identical verdicts for every implementation.