UML22 — Section: 15.3.12

The UML 2.0 Specification states at 15.1 that "The state machine formalism described in this section is an object-based variant of Harel statecharts." However, there is a big semantical discrepancy between the Harel statecharts as described in D. Harel and M. Politi, Modeling Reactive Systems with Statecharts: The STATEMATE Approach, (with M. Politi), McGraw-Hill, 1998 and the UML 2.0 specification. The major difference is in the priority of transitions when multiple transitions are enabled in case of a nested (hierarchical) state machine. Harel states (6.3.1 (pages 99-100)): "The criterion for priority of transitions prefers the transition whose source and target have a higher common ancestor state, if possible. If the common ancestors of both transitions are identical, then non-determinism indeed occurs." (i.e. it prefers global, higher-level transitions over local ones) UML 2.0 (15.3.12 page 618) imposes almost a reveres-ed order on the priority of the transitions, by looking up from the current nested leaf state and taking the first enabled transition. The impact of the UML definition is that the author can not only "refine" a high-level state in its descendants, he/she can override the global transitions thus violating the global (high-level) contract of the state machine. This becomes even more dangerous when using submachine state, i.e. the nested state is actually drawn in a separate diagram. Example: imagine an electrical device, which can be in one of 2 top-level states: ON, OFF and having two transitions power_on, power_off. The ON state can have multiple sub-states describing a particular state of the operation. Using the UML 2.0 semantics, one can effectively override the global power_off transition locally in on of the ON's children, forcing the electrical device to keep working, even if the power has been shut down - ignoring the signal using e.g. a self-transition.

To avoid confusion, add a clarification