Change of behavior in 1.72: different order if process_event() in on_entry()

[BrukerJWD]

I wrote a state machine with Boost 1.64. As it seemed most convenient for me, I put all actions into the on_entry() functions. For error handling I created a ErrorState that all states have a transition to if an ErrorEvent is signalled. This event is mostly sent internally, e.g. if a calculation over the measured data produces invalid results.

Now I changed to Boost 1.72 and my unit tests start to fail, because the ErrorEvent is now queued and only processed after the transition table. I do have some immediate transactions to separate the logic which now happen although the ErrorEvent is sent.

I wrote a small reproducer on wandbox. It has 3 states (State1, 2 and 3) + the ErrorState. According to the transition table, State1 transitions to State2 on a StartEvent, State2 immediately to State3 and State3 immediatly back to State1. State2 also has an entry for an ErrorEvent to transition to ErrorState", and inon_entry()ofState2theErrorEventis emitted. Until Boost 1.71 this put the state machine into theErrorState`, but since 1.72 the transition table is processed first.

If this program is executed with 1.71, the output is:

State1::on_entry after N5boost3msm4back13state_machineI2SmNS_9parameter5void_ES5_S5_S5_E9InitEventE
State2::on_entry after 10StartEvent
ErrorState::on_entry after 10ErrorEvent

whereas with 1.72 it is:

State1::on_entry after N5boost3msm4back13state_machineI2SmNS_9parameter5void_ES5_S5_S5_E9InitEventE
State2::on_entry after 10StartEvent
State3::on_entry after N5boost3msm5front4noneE
State1::on_entry after N5boost3msm5front4noneE
no transition from state State1 (0) on event 10ErrorEvent

I am not sure whether this is intended behavior.

[henry-ch]

Sorry for the very long dealy :( The behavior in 1.72 is correct. According to the UML Standard, completion events (none) have the highest priority, so that the event queue can only be processed after yóur FSM transitions back to State1.