The XWorkflow library has two main aspects:

  • Defining a workflow;
  • Using a workflow on an object.

Defining a workflow

A workflow is defined by subclassing the Workflow class, and setting a few specific attributes:

class MyWorkflow(xworkflows.Workflow):

    # The states in the workflow
    states = (
        ('init', _(u"Initial state")),
        ('ready', _(u"Ready")),
        ('active', _(u"Active")),
        ('done', _(u"Done")),
        ('cancelled', _(u"Cancelled")),

    # The transitions between those states
    transitions = (
        ('prepare', 'init', 'ready'),
        ('activate', 'ready', 'active'),
        ('complete', 'active', 'done'),
        ('cancel', ('ready', 'active'), 'cancelled'),

    # The initial state of objects using that workflow
    initial_state = 'init'

Those attributes will be transformed into similar attributes with friendlier APIs:

Accessing Workflow states and transitions

The states attribute, a StateList instance, provides a mixed dictionary/object API:

>>> MyWorkflow.states.init
>>> MyWorkflow.states.init.title
u"Initial state"
>>> MyWorkflow.states['ready']
>>> 'active' in MyWorkflow.states
>>> MyWorkflow.states.init in MyWorkflow.states
>>> list(MyWorkflow.states)  # definition order is kept
[State('init'), State('ready'), State('active'), State('done'), State('cancelled')]

The transitions attribute of a Workflow is a TransitionList instance, exposing a mixed dictionary/object API:

>>> MyWorkflow.transitions.prepare
Transition('prepare', [State('init')], State('ready'))
>>> MyWorkflow.transitions['cancel']
Transition('cancel', [State('ready'), State('actuve')], State('cancelled'))
>>> 'activate' in MyWorkflow.transitions
>>> MyWorkflow.transitions.available_from(MyWorkflow.states.ready)
[Transition('activate'), Transition('cancel')]
>>> list(MyWorkflow.transitions)  # Definition order is kept
[Transition('prepare'), Transition('activate'), Transition('complete'), Transition('cancel')]

Using a workflow

The process to apply a Workflow to an object is quite straightforward:

  • Inherit from WorkflowEnabled
  • Define one or more class-level attributes as foo = SomeWorkflow()

These attributes will be transformed into StateProperty objects, acting as a wrapper around the State held in the object’s internal __dict__.

For each transition of each related Workflow, the WorkflowEnabledMeta metaclass will add or enhance a method for each transition, according to the following rules:

  • If a class method is decorated with transition('XXX') where XXX is the name of a transition, that method becomes the ImplementationWrapper for that transition
  • For each remaining transition, if a method exists with the same name and is decorated with the transition() decorator, it will be used for the ImplementationWrapper of the transition. Methods with a transition name but no decorator will raise a TypeError – this ensures that all magic is somewhat explicit.
  • For all transitions which didn’t have an implementation in the class definition, a new method is added to the class definition. They have the same name as the transition, and a noop() implementation. TypeError is raised if a non-callable attribute already exists for a transition name.

Accessing the current state

For a WorkflowEnabled object, each <attr> = SomeWorkflow() definition is translated into a StateProperty object, which adds a few functions to a plain attribute:

  • It checks that any value set is a valid State from the related Workflow:

    >>> obj = MyObject()
    >>> obj.state = State('foo')
    Traceback (most recent call last):
        File "<stdin>", line 1, in <module>
    ValueError: Value State('foo') is not a valid state for workflow MyWorkflow.
  • It defaults to the initial_state of the Workflow if no value was set:

    >>> obj = MyObject()
    >>> obj.state
  • It wraps retrieved values into a StateWrapper, which adds a few extra attributes:

    • Access to the related workflow:

      >>> obj.state.workflow
      <Workflow: MyWorkflow>
    • List of accessible transitions:

      >>> obj.state.transitions
    • Easy testing of the current value:

      >>> obj.state.is_init
      >>> obj.state.is_ready
    • Native equivalence to the state's name:

      >>> obj.state == 'init'
      >>> obj.state == 'ready'
      >>> obj.state in ['init', 'ready']


      This behavior should only be used when accessing the State objects from the Workflow.states list is impossible, e.g comparison with external data (URL, database, …).

      Using State objects or the is_XXX attributes protects from typos in the code (AttributeError would be raised), whereas raw strings provide no such guarantee.

    • Easily setting the current value:

      >>> obj.state = MyWorkflow.states.ready
      >>> obj.state.is_ready
      >>> # Setting from a state name is also possible
      >>> obj.state = 'ready'
      >>> obj.state.is_ready


      Setting the state without going through transitions defeats the goal of xworkflows; this feature should only be used for faster testing or when saving/restoring objects from external storage.

Using transitions

Defining a transition implementation

In order to link a state change with specific code, a WorkflowEnabled object must simply have a method decorated with the transition() decorator.

If that method cannot be defined with the name of the related Transition, the name of that Transition should be passed as first argument to the transition() decorator:

class MyObject(xworkflows.WorkflowEnabled):

    state = MyWorkflow()

    def accept(self):

    def do_cancel(self):

Once decorated, any call to that method will perfom the following steps:

  1. Check that the current State of the object is a valid source for the target Transition (raises InvalidTransitionError otherwise);
  2. Checks that all optional transition_check() hooks, if defined, returns True (raises ForbiddenTransition otherwise);
  3. Run optional before_transition() and on_leave_state() hooks
  4. Call the code of the function;
  5. Change the State of the object;
  6. Call the Workflow.log_transition() method of the related Workflow;
  7. Run the optional after_transition() and on_enter_state() hooks, if defined.

Transitions for which no implementation was defined will have a basic noop() implementation.

Controlling transitions

According to the order above, preventing a State change can be done:


Additional control over the transition implementation can be obtained via hooks. 5 kinds of hooks exist:

  • transition_check(): those hooks are called just after the State check, and should return True if the transition can proceed. No argument is provided to the hook.
  • before_transition(): hooks to call just before running the actual implementation. They receive the same *args and **kwargs as passed to the actual implementation (but can’t modify them).
  • after_transition(): those hooks are called just after the State has been updated. It receives:
    • res: the return value of the actual implementation;
    • *args and **kwargs: the arguments passed to the actual implementation
  • on_leave_state(): functions to call just before leaving a state, along with the before_transition() hooks. They receive the same arguments as a before_transition() hook.
  • on_enter_state(): hooks to call just after entering a new state, along with after_transition() hooks. They receive the same arguments as a after_transition() hook.

The hook decorators all accept the following arguments:

  • A list of Transition names (for transition-related hooks) or State names (for state-related hooks); if empty, the hook will apply to all transitions:

    @xworkflows.after_transition('foo', 'bar')
    def hook(self, *args, **kwargs):
  • As a keyword field= argument, the name of the field whose transitions the hook applies to (when an instance uses more than one workflow):

    class MyObject(xworkflows.WorkflowEnabled):
        state1 = SomeWorkflow()
        state2 = AnotherWorkflow()
        def hook(self, res, *args, **kwargs):
            # Only called for transitions on state2.
  • As a keyword priority= argument (default: 0), the priority of the hook; hooks are applied in decreasing priority order:

    class MyObject(xworkflows.WorkflowEnabled):
        state = SomeWorkflow()
        @xworkflows.before_transition('*', priority=-1)
        def last_hook(self, *args, **kwargs):
            # Will be called last
        @xworkflows.before_transition('foo', priority=10)
        def first_hook(self, *args, **kwargs):
            # Will be called first

Hook decorators can also be stacked, in order to express complex hooking systems:

@xworkflows.before_transition('foobar', priority=4)
def hook(self, *args, **kwargs):

Hook call order

The order in which hooks are applied is computed based on the following rules:

In the following code snippet, the order is hook3, hook1, hook4, hook2:

def hook1(self):

def hook2(self):

def hook3(self):

def hook4(self):

Old-style hooks

Hooks can also be bound to the implementation at the transition() level:

@xworkflows.transition(check=some_fun, before=other_fun, after=something_else)
def accept(self):

Deprecated since version 0.4.0: Use before_transition(), after_transition() and transition_check() instead; will be removed in 0.5.0.

The old behaviour did not allow for hook overriding in inherited workflows.

Checking transition availability

Some programs may need to display available transitions, without calling them. Instead of checking manually the state of the object and calling the appropriate transition_check() hooks if defined, you should simply call myobj.some_transition.is_available():

class MyObject(WorkflowEnabled):
    state = MyWorkflow
    x = 13

    def check(self):
        return self.x == 42

    def accept(self):

    def cancel(self):
>>> obj = MyObject()
>>> obj.accept.is_available()  # Forbidden by 'check'
>>> obj.cancel.is_available()  # Forbidden by current state
>>> obj.x = 42
>>> obj.accept.is_available()

Logging transitions

The log_transition() method of a Workflow allows logging each Transition performed by an object using that Workflow.

This method is called with the following arguments:

  • transition: the Transition just performed
  • from_state: the State in which the object was just before the transition
  • instance: the object to which the transition was applied
  • *args: the arguments passed to the transition implementation
  • **kwargs: the keyword arguments passed to the transition implementation

The default implementation logs (with the logging module) to the xworkflows.transitions logger.

This behaviour can be overridden on a per-workflow basis: simply override the Workflow.log_transition() method.

Advanced customization

In order to perform advanced tasks when running transitions, libraries may hook directly at the ImplementationWrapper level.

For this, custom Workflow classes should override the Workflow.implementation_class attribute with their custom subclass and add extra behaviour there.

Possible customizations would be:

  • Wrapping implementation call and state update in a database transaction
  • Persisting the updated object after the transition
  • Adding workflow-level hooks to run before/after the transition
  • Performing the same sanity checks for all objects using that Workflow