How to implement a Decorator with non-local equality?

Question

Greetings, currently I am refactoring one of my programs, and I found an interesting problem.

I have Transitions in an automata. Transitions always have a start-state and an end-state. Some Transitions have a label, which encodes a certain Action that must be performed upon traversal. No label means no action. Some transitions have a condition, which must be fulfilled in order to traverse this condition, if there is no condition, the transition is basically an epsilon-transition in an NFA and will be traversed without consuming an input symbol.

I need the following operations:

• check if the transition has a label
• get this label
• add a label to a transition
• check if the transition has a condition
• get this condition
• check for equality

Judging from the first five points, this sounds like a clear decorator, with a base transition and two decorators: Labeled and Condition. However, this approach has a problem: two transitions are considered equal if their start-state and end-state are the same, the labels at both transitions are equal (or not-existing) and both conditions are the same (or not existing). With a decorator, I might have two transitions Labeled("foo", Conditional("bar", Transition("baz", "qux"))) and Conditional("bar", Labeled("foo", Transition("baz", "qux"))) which need a non-local equality, that is, the decorators would need to collect all the data and the Transition must compare this collected data on a set-base:

class Transition(object):
    def __init__(self, start, end):
        self.start = start
        self.end = end
    def get_label(self):
        return None
    def has_label(self):
        return False
    def collect_decorations(self, decorations):
        return decorations
    def internal_equality(self, my_decorations, other):
        try:
            return (self.start == other.start
                    and self.end == other.end
                    and my_decorations = other.collect_decorations())
    def __eq__(self, other):
        return self.internal_equality(self.collect_decorations({}), other)

class Labeled(object):
    def __init__(self, label, base):
        self.base = base
        self.label = label
    def has_label(self):
        return True
    def get_label(self):
        return self.label
    def collect_decorations(self, decorations):
        assert 'label' not in decorations
        decorations['label'] = self.label
        return self.base.collect_decorations(decorations)
    def __getattr__(self, attribute):
        return self.base.__getattr(attribute)

Is this a clean approach? Am I missing something?

I am mostly confused, because I can solve this - with longer class names - using cooperative multiple inheritance:

class Transition(object):
    def __init__(self, **kwargs):
        # init is pythons MI-madness ;-)
        super(Transition, self).__init__(**kwargs)
        self.start = kwargs['start']
        self.end = kwargs['end']
    def get_label(self):
        return None
    def get_condition(self):
        return None
    def __eq__(self, other):
        try:
            return self.start == other.start and self.end == other.end
        except AttributeError:
            return False

class LabeledTransition(Transition):
    def __init__(self, **kwargs):
        super(LabeledTransition).__init__(**kwargs)
        self.label = kwargs['label']
    def get_label(self):
        return self.label
    def __eq__(self):
        super_result = super(LabeledTransition, self).__eq__(other)
        try:
            return super_result and self.label == other.label
        except AttributeError:
            return False

class ConditionalTransition(Transition):
    def __init__(self, **kwargs):
        super(ConditionalTransition, self).__init__(**kwargs)
        self.condition = kwargs['condition']

    def get_condition(self):
        return self.condition

    def __eq__(self, other):
        super_result = super(ConditionalTransition, self).__eq__(other)
        try:
            return super_result and self.condition = other.condition
        except AttributeError:
            return False

# ConditionalTransition about the same, with get_condition
class LabeledConditionalTransition(LabeledTransition, ConditionalTransition):
    pass

the class LabledConditionalTransition behaves exactly as expected - and having no code in there is appealing and I do not thing MI is confusing at this size.

Of course, the third option would be to just hammer everything into a single transition class with a bunch of in has_label/has_transition.

So... I am confused. Am I missing something? Which implementation looks better? How do you handle similar cases, that is, objects which look like a Decorator could handle them, but then, such a non-local method comes around?

EDIT: Added the ConditionalTransition-class. Basically, this kinda behaves like the decorator, minus the order created by the order of creating the decorators, the transition checks for start and end being correct, the LabeledTransition-class checks for label being correct and ConditionalTransition checks for condition being correct.

Answer 1

From the code that was posted, the only difference between Transition and Labeled Transition is the return of get_lable() and has_label(). In which case you can compress these two a single class that sets a label attribute to None and

return self.label is not None

in the has_label() function.

Can you post the code for the ConditionalTransition class? I think this would make it clearer.

Answer 2

I think its clear that nobody really understands your question. I would suggest putting it in context and making it shorter. As an example, here's one possible implementation of the state pattern in python, please study it to get an idea.

class State(object):
    def __init__(self, name):
        self.name = name

    def __repr__(self):
        return self.name

class Automaton(object):
    def __init__(self, instance, start):
        self._state = start
        self.transitions = instance.transitions()

    def get_state(self):
        return self._state

    def set_state(self, target):
        transition = self.transitions.get((self.state, target))
        if transition:
            action, condition = transition
            if condition:
                if condition():
                    if action:
                        action()
                    self._state = target
            else:
                self._state = target
        else:
            self._state = target

    state = property(get_state, set_state)

class Door(object):
    open = State('open')
    closed = State('closed')

    def __init__(self, blocked=False):
        self.blocked = blocked

    def close(self):
        print 'closing door'

    def do_open(self):
        print 'opening door'

    def not_blocked(self):
        return not self.blocked

    def transitions(self):
        return {
            (self.open, self.closed):(self.close, self.not_blocked),
            (self.closed, self.open):(self.do_open, self.not_blocked),
        }

if __name__ == '__main__':
    door = Door()
    automaton = Automaton(door, door.open)

    print 'door is', automaton.state
    automaton.state = door.closed
    print 'door is', automaton.state
    automaton.state = door.open
    print 'door is', automaton.state
    door.blocked = True
    automaton.state = door.closed
    print 'door is', automaton.state

the output of this programm would be:

door is open
closing door
door is closed
opening door
door is open
door is open