Finite state machine field for sqlalchemy

sqlalchemy-fsm adds declarative states management for sqlalchemy models. Instead of adding some state field to a model, and manage its values by hand, you could use FSMState field and mark model methods with the transition decorator. Your method will contain the side-effects of the state change.

The decorator also takes a list of conditions, all of which must be met before a transition is allowed.

Usage

Add FSMState field to you model

from sqlalchemy_fsm import FSMField, transition

class BlogPost(db.Model):
    state = db.Column(FSMField, nullable = False)

Use the transition decorator to annotate model methods

@transition(source='new', target='published')
def published(self):
    """
    This function may contain side-effects,
    like updating caches, notifying users, etc.
    The return value will be discarded.
    """

source parameter accepts a list of states, or an individual state. You can use * for source, to allow switching to target from any state.

@transition- annotated methods have the following API:

1. <SqlAlchemy table class>.method() - returns an SqlAlchemy filter condition that can be used for querying the database (e.g. session.query(BlogPost).filter(BlogPost.published()))
2. <SqlAlchemy table class>.method.is_(<bool>) - same as <SqlAlchemy table class>.method() == <bool>
3. <SqlAlchemy record>.method() - returns boolean value that tells if this particular record is in the target state for that method() (e.g. if not blog.published():)
4. <SqlAlchemy record>.method.set(*args, **kwargs) - changes the state of the record object to the transitions' target state (or raises an exception if it is not able to do so)
5. <SqlAlchemy record>.method.can_proceed(*args, **kwargs) - returns True if calling .method.set(*args, **kwargs) (with same *args, **kwargs) should succeed.

You can also use None as source state for (e.g. in case when the state column in nullable). However, it is not possible to create transition with None as target state due to religious reasons.

Transition can be also used on a class object to create a group of handlers for same target state.

@transition(target='published')
class PublishHandler(object):

    @transition(source='new')
    def do_one(self, instance, value):
        instance.side_effect = "published from new"

    @transition(source='draft')
    def do_two(self, instance, value):
        instance.side_effect = "published from draft"


class BlogPost(db.Model):
    ...
    published = PublishHandler

The transition is still to be invoked by calling the model's published.set() method.

An alternative inline class syntax is supported too:

@transition(target='published')
class published(object):

    @transition(source='new')
    def do_one(self, instance, value):
        instance.side_effect = "published from new"

    @transition(source='draft')
    def do_two(self, instance, value):
        instance.side_effect = "published from draft"

If calling published.set() succeeds without raising an exception, the state field will be changed, but not written to the database.

def publish_view(request, post_id):
    post = get_object__or_404(BlogPost, pk=post_id)
    if not post.published.can_proceed():
         raise Http404;

    post.published.set()
    post.save()
    return redirect('/')

If your given function requires arguments to validate, you need to include them when calling can_proceed as well as including them when you call the function normally. Say publish.set() required a date for some reason:

if not post.published.can_proceed(the_date):
    raise Http404
else:
    post.publish(the_date)

If your code needs to know the state model is currently in, you can just call the main function function.

if post.deleted():
    raise Http404

If you require some conditions to be met before changing state, use the conditions argument to transition. conditions must be a list of functions that take one argument, the model instance. The function must return either True or False or a value that evaluates to True or False. If all functions return True, all conditions are considered to be met and transition is allowed to happen. If one of the functions return False, the transition will not happen. These functions should not have any side effects.

You can use ordinary functions

def can_publish(instance):
    # No publishing after 17 hours
    if datetime.datetime.now().hour > 17:
       return False
    return True

Or model methods

def can_destroy(self):
    return not self.is_under_investigation()

Use the conditions like this:

@transition(source='new', target='published', conditions=[can_publish])
def publish(self):
    """
    Side effects galore
    """

@transition(source='*', target='destroyed', conditions=[can_destroy])
def destroy(self):
    """
    Side effects galore
    """

You can also use FSM handlers to query the database. E.g.

session.query(BlogCls).filter(BlogCls.publish())

will return all "Blog" objects whose current state matches "publish"'es target state.

Events

Sqlalchemy-fsm integrates with sqlalchemy's event system. The library exposes two events before_state_change and after_state_change that are fired up at the expected points of state's lifecycle.

You can subscribe event listeners via standard SQLAlchemy interface of listens_for or listen.

from sqlalchemy.event import listens_for

@listens_for(Blog, 'before_state_change')
def on_state_change(instance, source, target):
    ...

Or

from sqlalchemy import event

def on_state_change(instance, source, target):
    ...

event.listen(Blog, 'after_state_change', on_state_change)

It is possible to de-register an event listener call with sqlalchemy.event.remove() method.

How does sqlalchemy-fsm diverge from django-fsm?

• Can't commit data from within transition-decorated functions

• Does support arguments to conditions functions