Repository navigation
Event(delay=..., internal=...) are dropped when the event is declared on a class body #647
Description
Event(..., delay=...) and Event(..., internal=...) are accepted at declaration time and then
discarded, so the resulting event has delay=0 and internal=False.
Reproduction
from statemachine import Event, State, StateChart
class Beacon(StateChart):
dark = State(initial=True)
lit = State(final=True)
light = Event(dark.to(lit), delay=500, internal=True)
print(Beacon.light.delay) # 0 , expected 500
print(Beacon.light.internal) # False , expected Truesm.light() fires immediately instead of after 500 ms, and goes to the external queue instead of
the internal one.
Cause
StateMachineMetaclass.add_from_attributes rebuilds the declared event to give it the real id,
and the new instance is constructed without the two fields:
new_event = Event(
transitions=value._transitions,
id=event_id,
name=value.name,
)delay and internal are declared on Event and honored everywhere downstream
(Event.put reads self.internal, the delayed-event machinery reads delay); they are simply
lost in this one rebuild.
Why the existing test does not catch it
tests/test_statechart_delayed.py::TestDelayedEvents::test_delayed_event_on_event_definition
declares light = Event(dark.to(lit), delay=50) but then triggers the machine with an event
object it builds itself:
BoundEvent(id="light", delay=50, _sm=sm)so it never reads the declared BeaconsOfGondor.light and passes regardless of what the
metaclass did with delay. A regression test for this issue should assert on the declared
event (SM.light.delay) and drive the machine through it.
Scope
Reproduced on the current develop. Affects the top-level declaration form; the nested
(State.Compound) form drops the two fields as well, for the same reason.
Activity
fgmacedo commentedon Sep 13, 2026
Fixed by #645 (4058570 on
develop). The reproduction now prints the expected values:add_from_attributespasses both fields through the rebuild, and the nested(
State.Compound) form preserves them too, since it now handsEvents.addanEventinstanceinstead of a plain id string.
The regression test this issue asked for is in place:
tests/test_events.py::TestExplicitEvent::test_accept_delay_and_internalasserts on thedeclared event and on the bound one.
tests/test_statechart_delayed.py::TestDelayedEvents::test_delayed_event_on_event_definitionno longer builds its own
BoundEvent(id="light", delay=50); it triggers the declared eventwith
sm.light.put(), so it now actually exercises the bug it was named for.Two remainders, each with its own issue:
internalis preserved on the declaration but never consumed at trigger time, becauseEvent.__get__does not pass it toBoundEventandsend()does not read it. Sosm.light()still goes to the external queue. internal declared on an Event has no effect at trigger time #655.error_,done_state_,done_invoke_), the declareddelay,internalandnameare still dropped:_expand_event_idproduces two ids in onespace-separated string, which is the public format for "N distinct events", so each one is
rebuilt and names itself. Expanded event ids travel inside a space-separated string, so error_ never worked as a method call #654.