.. currentmodule:: xworkflows
The XWorkflow library has two main aspects:
- Defining a workflow;
- Using a workflow on an object.
A workflow is defined by subclassing the :class:`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:
- :attr:`~Workflow.states` is defined as a list of two-tuples and converted into a :class:`~base.StateList`
- :attr:`~Workflow.transitions` is defined as a list of three-tuples and converted into a :class:`~base.TransitionList`
- :attr:`~Workflow.initial_state` is defined as the :attr:`~base.State.name` of the initial :class:`~base.State` of the :class:`Workflow` and converted into the appropriate :class:`~base.State`
Accessing :class:`Workflow` states and transitions
The :attr:`~Workflow.states` attribute, a :class:`~base.StateList` instance, provides a mixed dictionary/object API:
>>> MyWorkflow.states.init State('init') >>> MyWorkflow.states.init.title u"Initial state" >>> MyWorkflow.states['ready'] State('ready') >>> 'active' in MyWorkflow.states True >>> MyWorkflow.states.init in MyWorkflow.states True >>> list(MyWorkflow.states) # definition order is kept [State('init'), State('ready'), State('active'), State('done'), State('cancelled')]
The :attr:`~Workflow.transitions` attribute of a :class:`Workflow` is a :class:`~base.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 True >>> 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')]
The process to apply a :class:`Workflow` to an object is quite straightforward:
- Inherit from :class:`WorkflowEnabled`
- Define one or more class-level attributes as
foo = SomeWorkflow()
These attributes will be transformed into :class:`~base.StateProperty` objects, acting as a wrapper around the :class:`~base.State` held in the object's internal :attr:`__dict__`.
For each transition of each related :class:`Workflow`, the :class:`~base.WorkflowEnabledMeta` metaclass will add or enhance a method for each transition, according to the following rules:
- If a class method is decorated with :func:`transition('XXX') <transition>` where
XXX
is the name of a transition, that method becomes the :class:`~base.ImplementationWrapper` for that transition - For each remaining transition, if a method exists with the same name and is decorated with the :func:`~transition` decorator, it will be used for the :class:`~base.ImplementationWrapper` of the transition. Methods with a transition name but no decorator will raise a :exc:`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 :func:`~base.noop` implementation. :exc:`TypeError` is raised if a non-callable attribute already exists for a transition name.
For a :class:`WorkflowEnabled` object, each <attr> = SomeWorkflow()
definition
is translated into a :class:`~base.StateProperty` object, which adds a few functions
to a plain attribute:
It checks that any value set is a valid :class:`~base.State` from the related :class:`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 :attr:`~Workflow.initial_state` of the :class:`Workflow` if no value was set:
>>> obj = MyObject() >>> obj.state State('init')
It wraps retrieved values into a :class:`~base.StateWrapper`, which adds a few extra attributes:
Access to the related workflow:
>>> obj.state.workflow <Workflow: MyWorkflow>
List of accessible transitions:
>>> obj.state.transitions [Transition('accept')]
Easy testing of the current value:
>>> obj.state.is_init True >>> obj.state.is_ready False
Native equivalence to the :attr:`state's name <base.State.name>`:
>>> obj.state == 'init' True >>> obj.state == 'ready' False >>> obj.state in ['init', 'ready'] True
Note
This behavior should only be used when accessing the :class:`~base.State` objects from the :attr:`Workflow.states` list is impossible, e.g comparison with external data (URL, database, ...).
Using :class:`~base.State` objects or the :attr:`is_XXX` attributes protects from typos in the code (:exc:`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 True >>> # Setting from a state name is also possible >>> obj.state = 'ready' >>> obj.state.is_ready True
Note
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.
In order to link a state change with specific code, a :class:`WorkflowEnabled` object must simply have a method decorated with the :func:`transition` decorator.
If that method cannot be defined with the name of the related :class:`~base.Transition`, the name of that :class:`~base.Transition` should be passed as first argument to the :func:`transition` decorator:
class MyObject(xworkflows.WorkflowEnabled): state = MyWorkflow() @xworkflows.transition() def accept(self): pass @xworkflows.transition('cancel') def do_cancel(self): pass
Once decorated, any call to that method will perfom the following steps:
- Check that the current :class:`~base.State` of the object is a valid source for the target :class:`~base.Transition` (raises :exc:`InvalidTransitionError` otherwise);
- Checks that all optional :func:`transition_check` hooks, if defined, returns
True
(raises :exc:`ForbiddenTransition` otherwise); - Run optional :func:`~before_transition` and :func:`~on_leave_state` hooks
- Call the code of the function;
- Change the :class:`~base.State` of the object;
- Call the :func:`Workflow.log_transition` method of the related :class:`Workflow`;
- Run the optional :func:`~after_transition` and :func:`~on_enter_state` hooks, if defined.
Transitions for which no implementation was defined will have a basic :func:`~base.noop` implementation.
According to the order above, preventing a :class:`~base.State` change can be done:
- By returning
False
in a custom :func:`~transition_check` hook; - By raising any exception in a custom :func:`~before_transition` or :func:`~on_leave_state` hook;
- By raising any exception in the actual implementation.
Additional control over the transition implementation can be obtained via hooks. 5 kinds of hooks exist:
- :func:`transition_check`: those hooks are called just after the :class:`~base.State` check, and should
return
True
if the transition can proceed. No argument is provided to the hook. - :func:`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). - :func:`after_transition`: those hooks are called just after the :class:`~base.State` has been updated.
It receives:
res
: the return value of the actual implementation;*args
and**kwargs
: the arguments passed to the actual implementation
- :func:`on_leave_state`: functions to call just before leaving a state, along with the :func:`before_transition` hooks. They receive the same arguments as a :func:`before_transition` hook.
- :func:`on_enter_state`: hooks to call just after entering a new state, along with :func:`after_transition` hooks. They receive the same arguments as a :func:`after_transition` hook.
The hook decorators all accept the following arguments:
A list of :class:`~base.Transition` names (for transition-related hooks) or :class:`~base.State` names (for state-related hooks); if empty, the hook will apply to all transitions:
@xworkflows.before_transition() @xworkflows.after_transition('foo', 'bar') def hook(self, *args, **kwargs): pass
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() @xworkflows.on_enter_state(field='state2') def hook(self, res, *args, **kwargs): # Only called for transitions on state2. pass
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 pass @xworkflows.before_transition('foo', priority=10) def first_hook(self, *args, **kwargs): # Will be called first pass
Hook decorators can also be stacked, in order to express complex hooking systems:
@xworkflows.before_transition('foobar', priority=4) @xworkflows.on_leave_state('baz') def hook(self, *args, **kwargs): pass
The order in which hooks are applied is computed based on the following rules:
- Build the list of hooks to apply
- When testing if a transition can be applied, use all :func:`transition_check` hooks
- Before performing a transition, use all :func:`before_transition` and :func:`on_leave_state` hooks
- After performing a transition, use all :func:`after_transition` and :func:`on_enter_state` hooks
- Sort that list from higher to lower priority, and in alphabetical order if priority match
In the following code snippet, the order is hook3, hook1, hook4, hook2
:
@xworkflows.before_transition() def hook1(self): pass @xworkflows.before_transition(priority=-1) def hook2(self): pass @xworkflows.before_transition(priority=10) def hook3(self): pass @xworkflows.on_leave_state() def hook4(self): pass
Hooks can also be bound to the implementation at the :func:`transition` level:
@xworkflows.transition(check=some_fun, before=other_fun, after=something_else) def accept(self): pass
.. deprecated:: 0.4.0 Use :func:`before_transition`, :func:`after_transition` and :func:`transition_check` instead; will be removed in 0.5.0. The old behaviour did not allow for hook overriding in inherited workflows.
Some programs may need to display available transitions, without calling them.
Instead of checking manually the :class:`state <base.State>` of the object and calling
the appropriate :func:`transition_check` hooks if defined, you should simply call myobj.some_transition.is_available()
:
class MyObject(WorkflowEnabled):
state = MyWorkflow
x = 13
@transition_check('accept')
def check(self):
return self.x == 42
def accept(self):
pass
@transition()
def cancel(self):
pass
>>> obj = MyObject()
>>> obj.accept.is_available() # Forbidden by 'check'
False
>>> obj.cancel.is_available() # Forbidden by current state
False
>>> obj.x = 42
>>> obj.accept.is_available()
True
The :func:`~Workflow.log_transition` method of a :class:`Workflow` allows logging each :class:`~base.Transition` performed by an object using that :class:`Workflow`.
This method is called with the following arguments:
transition
: the :class:`~base.Transition` just performedfrom_state
: the :class:`~base.State` in which the object was just before the transitioninstance
: the :class:`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 :mod:`logging` module) to the xworkflows.transitions
logger.
This behaviour can be overridden on a per-workflow basis: simply override the :func:`Workflow.log_transition` method.
In order to perform advanced tasks when running transitions, libraries may hook directly at the :class:`~base.ImplementationWrapper` level.
For this, custom :class:`Workflow` classes should override the :attr:`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 :class:`Workflow`