The initialization process of a group of classes that share a common parent can be divided into three parts:
- Common initialization
- Subclass-specific initialization
- Common post-initialization
Currently the first two parts are called from the __init__ method of each child class, but the final post-initialization part has to be called separately, for example
class BaseClass:def __init__(self):print 'base __init__'self.common1()def common1(self):print 'common 1'def finalizeInitialization(self):print 'finalizeInitialization [common2]'class Subclass1(BaseClass):def __init__(self):BaseClass.__init__(self)self.specific()def specific(self):print 'specific'if __name__ == '__main__':s = Subclass1() # Don't forget to finalize the initializations.finalizeInitialization() # now the object is fully initializedIs there a way to not to have to call finalizeInitialization()? Or one can transfer the call to finalizeInitialization() into Subclass1's __init__ (as in S.Lott's answer). This makes the life easier, but still one has to remember to complete the initialization, this time inside the "constructor". Either way there is no way to enforce full initialization, which is what I'm looking for.
Best Answer
Template Method Design Pattern to the rescue:
class BaseClass:def __init__(self, specifics=None):print 'base __init__'self.common1()if specifics is not None:specifics()self.finalizeInitialization()def common1(self):print 'common 1'def finalizeInitialization(self):print 'finalizeInitialization [common2]'class Subclass1(BaseClass):def __init__(self):BaseClass.__init__(self, self.specific)def specific(self):print 'specific'You can use metaclasses (and updated to Python3 code):
class PostInitCaller(type):def __call__(cls, *args, **kwargs):obj = type.__call__(cls, *args, **kwargs)obj.__post_init__()return objclass BaseClass(metaclass=PostInitCaller): def __init__(self):print('base __init__')self.common1()def common1(self):print('common 1')def finalizeInitialization(self):print('finalizeInitialization [common2]')def __post_init__(self): # this is called at the end of __init__self.finalizeInitialization()class Subclass1(BaseClass):def __init__(self):super().__init__()self.specific()def specific(self):print('specific')s = Subclass1() Version 1 - delegate everything.
class Subclass1(BaseClass):def __init__(self):super( Subclass1, self ).__init__()self.specific()super( Subclass1, self ).finalizeInitialization()Version 2 - delegate just one step
class BaseClass:def __init__(self):print 'base __init__'self.common1()self.specific()self.finalizeInitialization()def common1(self):print 'common 1'def finalizeInitialization(self):print 'finalizeInitialization [common2]'def specific( self ):# two choices:# if this is "abstract": raise an exception# if this is "concrete": passSimilar to S. Lott's approach, except there's no way (short of overriding __init__) for the derived classes to override (or even call) the common methods:
class BaseClass:def __init__(self):def common():print "common initialization..."def final():print "common finalization..."common()self.specific()final()def final_init(self):print "BaseClass.final_init"class Subclass1(BaseClass):def specific(self):print "Subclass1.specific"You might want to provide a default implementation of specific in BaseClass if it's not okay to raise an AttributeError when you create an instance of any subclass that doesn't provide its own implementation.
What's wrong with calling finalInitilazation from the Subclass's init?
class BaseClass:def __init__(self):print 'base __init__'self.common1()def common1(self):print 'common 1'def finalizeInitialization(self):print 'finalizeInitialization [common2]'class Subclass1(BaseClass):def __init__(self):BaseClass.__init__(self)self.specific()BaseClass.finalizeInitialization(self)def specific(self):print 'specific'if __name__ == '__main__':s = Subclass1() #Don't forget to finalize the initializations.finalizeInitialization() # now the object is fully initializedIf you have to call multiple methods in a specific order it typically means that the design has problems to begin with (it's leaking implementation detail). So I would try to work from that end.
On the other hand if people derive from the class and have to modify the initialisation they should be aware of the implications - it is not something you would want to have in your normal API. Alternatively you could be defensive about the final initialization and check that it has been called in methods that depend on it (to call it or raise an exception if not).