Develop Reference

How can I delay the __init__ call until an attribute is accessed?

I have a test framework that requires test cases to be defined using the following class patterns:
class TestBase:
def __init__(self, params):
self.name = str(self.__class__)
print('initializing test: {} with params: {}'.format(self.name, params))
class TestCase1(TestBase):
def run(self):
print('running test: ' + self.name)
When I create and run a test, I get the following:
>>> test1 = TestCase1('test 1 params')
initializing test: <class '__main__.TestCase1'> with params: test 1 params
>>> test1.run()
running test: <class '__main__.TestCase1'>
The test framework searches for and loads all TestCase classes it can find, instantiates each one, then calls the run method for each test.
load_test(TestCase1(test_params1))
load_test(TestCase2(test_params2))
...
load_test(TestCaseN(test_params3))
...
for test in loaded_tests:
test.run()
However, I now have some test cases for which I don't want the __init__ method called until the time that the run method is called, but I have little control over the framework structure or methods. How can I delay the call to __init__ without redefining the __init__ or run methods?
Update
The speculations that this originated as an XY problem are correct. A coworker asked me this question a while back when I was maintaining said test framework. I inquired further about what he was really trying to achieve and we figured out a simpler workaround that didn't involve changing the framework or introducing metaclasses, etc.
However, I still think this is a question worth investigating: if I wanted to create new objects with "lazy" initialization ("lazy" as in lazy evaluation generators such as range, etc.) what would be the best way of accomplishing it? My best attempt so far is listed below, I'm interested in knowing if there's anything simpler or less verbose.

First Solution：use property.the elegant way of setter/getter in python.
class Bars(object):
def __init__(self):
self._foo = None
#property
def foo(self):
if not self._foo:
print("lazy initialization")
self._foo = [1,2,3]
return self._foo
if __name__ == "__main__":
f = Bars()
print(f.foo)
print(f.foo)
Second Solution:the proxy solution,and always implement by decorator.
In short, Proxy is a wrapper that wraps the object you need. Proxy could provide additional functionality to the object that it wraps and doesn't change the object's code. It's a surrogate which provide the abitity of control access to a object.there is the code come form user Cyclone.
class LazyProperty:
def __init__(self, method):
self.method = method
self.method_name = method.__name__
def __get__(self, obj, cls):
if not obj:
return None
value = self.method(obj)
print('value {}'.format(value))
setattr(obj, self.method_name, value)
return value
class test:
def __init__(self):
self._resource = None
#LazyProperty
def resource(self):
print("lazy")
self._resource = tuple(range(5))
return self._resource
if __name__ == '__main__':
t = test()
print(t.resource)
print(t.resource)
print(t.resource)
To be used for true one-time calculated lazy properties. I like it because it avoids sticking extra attributes on objects, and once activated does not waste time checking for attribute presence

Metaclass option
You can intercept the call to __init__ using a metaclass. Create the object with __new__ and overwrite the __getattribute__ method to check if __init__ has been called or not and call it if it hasn't.
class DelayInit(type):
def __call__(cls, *args, **kwargs):
def init_before_get(obj, attr):
if not object.__getattribute__(obj, '_initialized'):
obj.__init__(*args, **kwargs)
obj._initialized = True
return object.__getattribute__(obj, attr)
cls.__getattribute__ = init_before_get
new_obj = cls.__new__(cls, *args, **kwargs)
new_obj._initialized = False
return new_obj
class TestDelayed(TestCase1, metaclass=DelayInit):
pass
In the example below, you'll see that the init print won't occur until the run method is executed.
>>> new_test = TestDelayed('delayed test params')
>>> new_test.run()
initializing test: <class '__main__.TestDelayed'> with params: delayed test params
running test: <class '__main__.TestDelayed'>
Decorator option
You could also use a decorator that has a similar pattern to the metaclass above:
def delayinit(cls):
def init_before_get(obj, attr):
if not object.__getattribute__(obj, '_initialized'):
obj.__init__(*obj._init_args, **obj._init_kwargs)
obj._initialized = True
return object.__getattribute__(obj, attr)
cls.__getattribute__ = init_before_get
def construct(*args, **kwargs):
obj = cls.__new__(cls, *args, **kwargs)
obj._init_args = args
obj._init_kwargs = kwargs
obj._initialized = False
return obj
return construct
#delayinit
class TestDelayed(TestCase1):
pass
This will behave identically to the example above.

In Python, there is no way that you can avoid calling __init__ when you instantiate a class cls. If calling cls(args) returns an instance of cls, then the language guarantees that cls.__init__ will have been called.
So the only way to achieve something similar to what you are asking is to introduce another class that will postpone the calling of __init__ in the original class until an attribute of the instantiated class is being accessed.
Here is one way:
def delay_init(cls):
class Delay(cls):
def __init__(self, *arg, **kwarg):
self._arg = arg
self._kwarg = kwarg
def __getattribute__(self, name):
self.__class__ = cls
arg = self._arg
kwarg = self._kwarg
del self._arg
del self._kwarg
self.__init__(*arg, **kwarg)
return getattr(self, name)
return Delay
This wrapper function works by catching any attempt to access an attribute of the instantiated class. When such an attempt is made, it changes the instance's __class__ to the original class, calls the original __init__ method with the arguments that were used when the instance was created, and then returns the proper attribute. This function can be used as decorator for your TestCase1 class:
class TestBase:
def __init__(self, params):
self.name = str(self.__class__)
print('initializing test: {} with params: {}'.format(self.name, params))
class TestCase1(TestBase):
def run(self):
print('running test: ' + self.name)
>>> t1 = TestCase1("No delay")
initializing test: <class '__main__.TestCase1'> with params: No delay
>>> t2 = delay_init(TestCase1)("Delayed init")
>>> t1.run()
running test: <class '__main__.TestCase1'>
>>> t2.run()
initializing test: <class '__main__.TestCase1'> with params: Delayed init
running test: <class '__main__.TestCase1'>
>>>
Be careful where you apply this function though. If you decorate TestBase with delay_init, it will not work, because it will turn the TestCase1 instances into TestBase instances.

In my answer I'd like to focus on cases when one wants to instantiate a class whose initialiser (dunder init) has side effects. For instance, pysftp.Connection, creates an SSH connection, which may be undesired until it's actually used.
In a great blog series about conceiving of wrapt package (nit-picky decorator implementaion), the author describes Transparent object proxy. This code can be customised for the subject in question.
class LazyObject:
_factory = None
'''Callable responsible for creation of target object'''
_object = None
'''Target object created lazily'''
def __init__(self, factory):
self._factory = factory
def __getattr__(self, name):
if not self._object:
self._object = self._factory()
return getattr(self._object, name)
Then it can be used as:
obj = LazyObject(lambda: dict(foo = 'bar'))
obj.keys() # dict_keys(['foo'])
But len(obj), obj['foo'] and other language constructs which invoke Python object protocols (dunder methods, like __len__ and __getitem__) will not work. However, for many cases, which are limited to regular methods, this is a solution.
To proxy object protocol implementations, it's possible to use neither __getattr__, nor __getattribute__ (to do it in a generic way). The latter's documentation notes:
This method may still be bypassed when looking up special methods as the result of implicit invocation via language syntax or built-in functions. See Special method lookup.
As a complete solution is demanded, there are examples of manual implementations like werkzeug's LocalProxy and django's SimpleLazyObject. However a clever workaround is possible.
Luckily there's a dedicated package (based on wrapt) for the exact use case, lazy-object-proxy which is described in this blog post.
from lazy_object_proxy import Proxy
obj = Proxy(labmda: dict(foo = 'bar'))
obj.keys() # dict_keys(['foo'])
len(len(obj)) # 1
obj['foo'] # 'bar'

One alternative would be to write a wrapper that takes a class as input and returns a class with delayed initialization until any member is accessed. This could for example be done as this:
def lazy_init(cls):
class LazyInit(cls):
def __init__(self, *args, **kwargs):
self.args = args
self.kwargs = kwargs
self._initialized = False
def __getattr__(self, attr):
if not self.__dict__['_initialized']:
cls.__init__(self,
*self.__dict__['args'], **self.__dict__['kwargs'])
self._initialized = True
return self.__dict__[attr]
return LazyInit
This could then be used as such
load_test(lazy_init(TestCase1)(test_params1))
load_test(lazy_init(TestCase2)(test_params2))
...
load_test(lazy_init(TestCaseN)(test_params3))
...
for test in loaded_tests:
test.run()

Answering your original question (and the problem I think you are actually trying to solve), "How can I delay the init call until an attribute is accessed?": don't call init until you access the attribute.
Said another way: you can make the class initialization simultaneous with the attribute call. What you seem to actually want is 1) create a collection of TestCase# classes along with their associated parameters; 2) run each test case.
Probably your original problem came from thinking you had to initialize all your TestCase classes in order to create a list of them that you could iterate over. But in fact you can store class objects in lists, dicts etc. That means you can do whatever method you have for finding all TestCase classes and store those class objects in a dict with their relevant parameters. Then just iterate that dict and call each class with its run() method.
It might look like:
tests = {TestCase1: 'test 1 params', TestCase2: 'test 2 params', TestCase3: 'test 3 params'}
for test_case, param in tests.items():
test_case(param).run()

Overridding __new__
You could do this by overriding __new__ method and replacing __init__ method with a custom function.
def init(cls, real_init):
def wrapped(self, *args, **kwargs):
# This will run during the first call to `__init__`
# made after `__new__`. Here we re-assign the original
# __init__ back to class and assign a custom function
# to `instances.__init__`.
cls.__init__ = real_init
def new_init():
if new_init.called is False:
real_init(self, *args, **kwargs)
new_init.called = True
new_init.called = False
self.__init__ = new_init
return wrapped
class DelayInitMixin(object):
def __new__(cls, *args, **kwargs):
cls.__init__ = init(cls, cls.__init__)
return object.__new__(cls)
class A(DelayInitMixin):
def __init__(self, a, b):
print('inside __init__')
self.a = sum(a)
self.b = sum(b)
def __getattribute__(self, attr):
init = object.__getattribute__(self, '__init__')
if not init.called:
init()
return object.__getattribute__(self, attr)
def run(self):
pass
def fun(self):
pass
Demo:
>>> a = A(range(1000), range(10000))
>>> a.run()
inside __init__
>>> a.a, a.b
(499500, 49995000)
>>> a.run(), a.__init__()
(None, None)
>>> b = A(range(100), range(10000))
>>> b.a, b.b
inside __init__
(4950, 49995000)
>>> b.run(), b.__init__()
(None, None)
Using cached properties
The idea is to do the heavy calculation only once by caching results. This approach will lead to much more readable code if the whole point of delaying initialization is improving performance.
Django comes with a nice decorator called #cached_property. I tend to use it a lot in both code and unit-tests for caching results of heavy properties.
A cached_property is a non-data descriptor. Hence once the key is set in instance's dictionary, the access to property would always get the value from there.
class cached_property(object):
"""
Decorator that converts a method with a single self argument into a
property cached on the instance.
Optional ``name`` argument allows you to make cached properties of other
methods. (e.g. url = cached_property(get_absolute_url, name='url') )
"""
def __init__(self, func, name=None):
self.func = func
self.__doc__ = getattr(func, '__doc__')
self.name = name or func.__name__
def __get__(self, instance, cls=None):
if instance is None:
return self
res = instance.__dict__[self.name] = self.func(instance)
return res
Usage:
class A:
#cached_property
def a(self):
print('calculating a')
return sum(range(1000))
#cached_property
def b(self):
print('calculating b')
return sum(range(10000))
Demo:
>>> a = A()
>>> a.a
calculating a
499500
>>> a.b
calculating b
49995000
>>> a.a, a.b
(499500, 49995000)

I think you can use a wrapper class to hold the real class you want to instance, and use call __init__ yourself in your code, like(Python 3 code):
class Wrapper:
def __init__(self, cls):
self.cls = cls
self.instance = None
def your_method(self, *args, **kwargs):
if not self.instance:
self.instnace = cls()
return self.instance(*args, **kwargs)
class YourClass:
def __init__(self):
print("calling __init__")
but it's a dump way, but without any trick.

static method called instaed of instance method [duplicate]

I'd like to do something like this:
class X:
#classmethod
def id(cls):
return cls.__name__
def id(self):
return self.__class__.__name__
And now call id() for either the class or an instance of it:
>>> X.id()
'X'
>>> X().id()
'X'
Obviously, this exact code doesn't work, but is there a similar way to make it work?
Or any other workarounds to get such behavior without too much "hacky" stuff?

Class and instance methods live in the same namespace and you cannot reuse names like that; the last definition of id will win in that case.
The class method will continue to work on instances however, there is no need to create a separate instance method; just use:
class X:
#classmethod
def id(cls):
return cls.__name__
because the method continues to be bound to the class:
>>> class X:
... #classmethod
... def id(cls):
... return cls.__name__
...
>>> X.id()
'X'
>>> X().id()
'X'
This is explicitly documented:
It can be called either on the class (such as C.f()) or on an instance (such as C().f()). The instance is ignored except for its class.
If you do need distinguish between binding to the class and an instance
If you need a method to work differently based on where it is being used on; bound to a class when accessed on the class, bound to the instance when accessed on the instance, you'll need to create a custom descriptor object.
The descriptor API is how Python causes functions to be bound as methods, and bind classmethod objects to the class; see the descriptor howto.
You can provide your own descriptor for methods by creating an object that has a __get__ method. Here is a simple one that switches what the method is bound to based on context, if the first argument to __get__ is None, then the descriptor is being bound to a class, otherwise it is being bound to an instance:
class class_or_instancemethod(classmethod):
def __get__(self, instance, type_):
descr_get = super().__get__ if instance is None else self.__func__.__get__
return descr_get(instance, type_)
This re-uses classmethod and only re-defines how it handles binding, delegating the original implementation for instance is None, and to the standard function __get__ implementation otherwise.
Note that in the method itself, you may then have to test, what it is bound to. isinstance(firstargument, type) is a good test for this:
>>> class X:
... #class_or_instancemethod
... def foo(self_or_cls):
... if isinstance(self_or_cls, type):
... return f"bound to the class, {self_or_cls}"
... else:
... return f"bound to the instance, {self_or_cls"
...
>>> X.foo()
"bound to the class, <class '__main__.X'>"
>>> X().foo()
'bound to the instance, <__main__.X object at 0x10ac7d580>'
An alternative implementation could use two functions, one for when bound to a class, the other when bound to an instance:
class hybridmethod:
def __init__(self, fclass, finstance=None, doc=None):
self.fclass = fclass
self.finstance = finstance
self.__doc__ = doc or fclass.__doc__
# support use on abstract base classes
self.__isabstractmethod__ = bool(
getattr(fclass, '__isabstractmethod__', False)
)
def classmethod(self, fclass):
return type(self)(fclass, self.finstance, None)
def instancemethod(self, finstance):
return type(self)(self.fclass, finstance, self.__doc__)
def __get__(self, instance, cls):
if instance is None or self.finstance is None:
# either bound to the class, or no instance method available
return self.fclass.__get__(cls, None)
return self.finstance.__get__(instance, cls)
This then is a classmethod with an optional instance method. Use it like you'd use a property object; decorate the instance method with #<name>.instancemethod:
>>> class X:
... #hybridmethod
... def bar(cls):
... return f"bound to the class, {cls}"
... #bar.instancemethod
... def bar(self):
... return f"bound to the instance, {self}"
...
>>> X.bar()
"bound to the class, <class '__main__.X'>"
>>> X().bar()
'bound to the instance, <__main__.X object at 0x10a010f70>'
Personally, my advice is to be cautious about using this; the exact same method altering behaviour based on the context can be confusing to use. However, there are use-cases for this, such as SQLAlchemy's differentiation between SQL objects and SQL values, where column objects in a model switch behaviour like this; see their Hybrid Attributes documentation. The implementation for this follows the exact same pattern as my hybridmethod class above.

I have no idea what's your actual use case is, but you can do something like this using a descriptor:
class Desc(object):
def __get__(self, ins, typ):
if ins is None:
print 'Called by a class.'
return lambda : typ.__name__
else:
print 'Called by an instance.'
return lambda : ins.__class__.__name__
class X(object):
id = Desc()
x = X()
print x.id()
print X.id()
Output
Called by an instance.
X
Called by a class.
X

It can be done, quite succinctly, by binding the instance-bound version of your method explicitly to the instance (rather than to the class). Python will invoke the instance attribute found in Class().__dict__ when Class().foo() is called (because it searches the instance's __dict__ before the class'), and the class-bound method found in Class.__dict__ when Class.foo() is called.
This has a number of potential use cases, though whether they are anti-patterns is open for debate:
class Test:
def __init__(self):
self.check = self.__check
#staticmethod
def check():
print('Called as class')
def __check(self):
print('Called as instance, probably')
>>> Test.check()
Called as class
>>> Test().check()
Called as instance, probably
Or... let's say we want to be able to abuse stuff like map():
class Str(str):
def __init__(self, *args):
self.split = self.__split
#staticmethod
def split(sep=None, maxsplit=-1):
return lambda string: string.split(sep, maxsplit)
def __split(self, sep=None, maxsplit=-1):
return super().split(sep, maxsplit)
>>> s = Str('w-o-w')
>>> s.split('-')
['w', 'o', 'w']
>>> Str.split('-')(s)
['w', 'o', 'w']
>>> list(map(Str.split('-'), [s]*3))
[['w', 'o', 'w'], ['w', 'o', 'w'], ['w', 'o', 'w']]

"types" provides something quite interesting since Python 3.4: DynamicClassAttribute
It is not doing 100% of what you had in mind, but it seems to be closely related, and you might need to tweak a bit my metaclass but, rougly, you can have this;
from types import DynamicClassAttribute
class XMeta(type):
def __getattr__(self, value):
if value == 'id':
return XMeta.id # You may want to change a bit that line.
#property
def id(self):
return "Class {}".format(self.__name__)
That would define your class attribute. For the instance attribute:
class X(metaclass=XMeta):
#DynamicClassAttribute
def id(self):
return "Instance {}".format(self.__class__.__name__)
It might be a bit overkill especially if you want to stay away from metaclasses. It's a trick I'd like to explore on my side, so I just wanted to share this hidden jewel, in case you can polish it and make it shine!
>>> X().id
'Instance X'
>>> X.id
'Class X'
Voila...

In your example, you could simply delete the second method entirely, since both the staticmethod and the class method do the same thing.
If you wanted them to do different things:
class X:
def id(self=None):
if self is None:
# It's being called as a static method
else:
# It's being called as an instance method

(Python 3 only) Elaborating on the idea of a pure-Python implementation of #classmethod, we can declare an #class_or_instance_method as a decorator, which is actually a class implementing the attribute descriptor protocol:
import inspect
class class_or_instance_method(object):
def __init__(self, f):
self.f = f
def __get__(self, instance, owner):
if instance is not None:
class_or_instance = instance
else:
class_or_instance = owner
def newfunc(*args, **kwargs):
return self.f(class_or_instance, *args, **kwargs)
return newfunc
class A:
#class_or_instance_method
def foo(self_or_cls, a, b, c=None):
if inspect.isclass(self_or_cls):
print("Called as a class method")
else:
print("Called as an instance method")

Add attribute to python class

Consider the following code:
class Foo():
pass
Foo.entries = dict()
a = Foo()
a.entries['1'] = 1
b = Foo()
b.entries['3'] = 3
print(a.entries)
This will print:
{'1': 1, '3': 3}
because the entries is added as static attribute. Is there a way monkey patch the class definition in order to add new attributes (without using inheritance).
I managed to find the following way but it looks convoluted to me:
def patch_me(target, field, value):
def func(self):
if not hasattr(self, '__' + field):
setattr(self, '__' + field, value())
return getattr(self, '__' + field)
setattr(target, field, property(func))
patch_me(Foo, 'entries', dict)

Ordinarily, attributes are added either by the __init__() function or after instantiating:
foo = Foo()
foo.bar = 'something' # note case
If you want to do this automatically, inheritance is by far the simplest way to do so:
class Baz(Foo):
def __init__(self):
super().__init__() # super() needs arguments in 2.x
self.bar = 'something'
Note that classes don't need to appear at the top level of a Python module. You can declare a class inside a function:
def make_baz(value):
class Baz(Foo):
def __init__(self):
super().__init__() # super() needs arguments in 2.x
self.bar = value()
return Baz()
This example will create a new class every time make_baz() is called. That may or may not be what you want. It would probably be simpler to just do this:
def make_foo(value):
result = Foo()
result.bar = value()
return result
If you're really set on monkey-patching the original class, the example code you provided is more or less the simplest way of doing it. You might consider using decorator syntax for property(), but that's a minor change. I should also note that it will not invoke double-underscore name mangling, which is probably a good thing because it means you cannot conflict with any names used by the class itself.

Method Inheritance in Python

I have a parent class and two child class. The parent class is an abstract base class that has method combine that gets inherited by the child classes. But each child implements combine differently from a parameter perspective therefore each of their own methods take different number of parameters. In Python, when a child inherits a method and requires re-implementing it, that newly re-implemented method must match parameter by parameter. Is there a way around this? I.e. the inherited method can have dynamic parameter composition?

This code demonstrates that signature of overridden method can easily change.
class Parent(object):
def foo(self, number):
for _ in range(number):
print "Hello from parent"
class Child(Parent):
def foo(self, number, greeting):
for _ in range(number):
print greeting
class GrandChild(Child):
def foo(self):
super(GrandChild,self).foo(1, "hey")
p = Parent()
p.foo(3)
c = Child()
c.foo(2, "Hi")
g = GrandChild()
g.foo()

As the other answer demonstrates for plain classes, the signature of an overridden inherited method can be different in the child than in the parent.
The same is true even if the parent is an abstract base class:
import abc
class Foo:
__metaclass__ = abc.ABCMeta
#abc.abstractmethod
def bar(self, x, y):
return x + y
class ChildFoo(Foo):
def bar(self, x):
return super(self.__class__, self).bar(x, 3)
class DumbFoo(Foo):
def bar(self):
return "derp derp derp"
cf = ChildFoo()
print cf.bar(5)
df = DumbFoo()
print df.bar()
Inappropriately complicated detour
It is an interesting exercise in Python metaclasses to try to restrict the ability to override methods, such that their argument signature must match that of the base class. Here is an attempt.
Note: I'm not endorsing this as a good engineering idea, and I did not spend time tying up loose ends so there are likely little caveats about the code below that could make it more efficient or something.
import types
import inspect
def strict(func):
"""Add some info for functions having strict signature.
"""
arg_sig = inspect.getargspec(func)
func.is_strict = True
func.arg_signature = arg_sig
return func
class StrictSignature(type):
def __new__(cls, name, bases, attrs):
func_types = (types.MethodType,) # include types.FunctionType?
# Check each attribute in the class being created.
for attr_name, attr_value in attrs.iteritems():
if isinstance(attr_value, func_types):
# Check every base for #strict functions.
for base in bases:
base_attr = base.__dict__.get(attr_name)
base_attr_is_function = isinstance(base_attr, func_types)
base_attr_is_strict = hasattr(base_attr, "is_strict")
# Assert that inspected signatures match.
if base_attr_is_function and base_attr_is_strict:
assert (inspect.getargspec(attr_value) ==
base_attr.arg_signature)
# If everything passed, create the class.
return super(StrictSignature, cls).__new__(cls, name, bases, attrs)
# Make a base class to try out strictness
class Base:
__metaclass__ = StrictSignature
#strict
def foo(self, a, b, c="blah"):
return a + b + len(c)
def bar(self, x, y, z):
return x
#####
# Now try to make some classes inheriting from Base.
#####
class GoodChild(Base):
# Was declared strict, better match the signature.
def foo(self, a, b, c="blah"):
return c
# Was never declared as strict, so no rules!
def bar(im_a_little, teapot):
return teapot/2
# These below can't even be created. Uncomment and try to run the file
# and see. It's not just that you can't instantiate them, you can't
# even get the *class object* defined at class creation time.
#
#class WrongChild(Base):
# def foo(self, a):
# return super(self.__class__, self).foo(a, 5)
#
#class BadChild(Base):
# def foo(self, a, b, c="halb"):
# return super(self.__class__, self).foo(a, b, c)
Note, like with most "strict" or "private" type ideas in Python, that you are still free to monkey-patch functions onto even a "good class" and those monkey-patched functions don't have to satisfy the signature constraint.
# Instance level
gc = GoodChild()
gc.foo = lambda self=gc: "Haha, I changed the signature!"
# Class level
GoodChild.foo = lambda self: "Haha, I changed the signature!"
and even if you add more complexity to the meta class that checks whenever any method type attributes are updated in the class's __dict__ and keeps making the assert statement when the class is modified, you can still use type.__setattr__ to bypass customized behavior and set an attribute anyway.
In these cases, I imagine Jeff Goldblum as Ian Malcolm from Jurassic Park, looking at you blankly and saying "Consenting adults, uhh, find a way.."

Python and the Singleton Pattern [duplicate]

This question already has answers here:
Creating a singleton in Python
(38 answers)
Closed 4 years ago.
There seem to be many ways to define singletons in Python. Is there a consensus opinion on Stack Overflow?

I don't really see the need, as a module with functions (and not a class) would serve well as a singleton. All its variables would be bound to the module, which could not be instantiated repeatedly anyway.
If you do wish to use a class, there is no way of creating private classes or private constructors in Python, so you can't protect against multiple instantiations, other than just via convention in use of your API. I would still just put methods in a module, and consider the module as the singleton.

Here's my own implementation of singletons. All you have to do is decorate the class; to get the singleton, you then have to use the Instance method. Here's an example:
#Singleton
class Foo:
def __init__(self):
print 'Foo created'
f = Foo() # Error, this isn't how you get the instance of a singleton
f = Foo.instance() # Good. Being explicit is in line with the Python Zen
g = Foo.instance() # Returns already created instance
print f is g # True
And here's the code:
class Singleton:
"""
A non-thread-safe helper class to ease implementing singletons.
This should be used as a decorator -- not a metaclass -- to the
class that should be a singleton.
The decorated class can define one `__init__` function that
takes only the `self` argument. Also, the decorated class cannot be
inherited from. Other than that, there are no restrictions that apply
to the decorated class.
To get the singleton instance, use the `instance` method. Trying
to use `__call__` will result in a `TypeError` being raised.
"""
def __init__(self, decorated):
self._decorated = decorated
def instance(self):
"""
Returns the singleton instance. Upon its first call, it creates a
new instance of the decorated class and calls its `__init__` method.
On all subsequent calls, the already created instance is returned.
"""
try:
return self._instance
except AttributeError:
self._instance = self._decorated()
return self._instance
def __call__(self):
raise TypeError('Singletons must be accessed through `instance()`.')
def __instancecheck__(self, inst):
return isinstance(inst, self._decorated)

You can override the __new__ method like this:
class Singleton(object):
_instance = None
def __new__(cls, *args, **kwargs):
if not cls._instance:
cls._instance = super(Singleton, cls).__new__(
cls, *args, **kwargs)
return cls._instance
if __name__ == '__main__':
s1 = Singleton()
s2 = Singleton()
if (id(s1) == id(s2)):
print "Same"
else:
print "Different"

A slightly different approach to implement the singleton in Python is the borg pattern by Alex Martelli (Google employee and Python genius).
class Borg:
__shared_state = {}
def __init__(self):
self.__dict__ = self.__shared_state
So instead of forcing all instances to have the same identity, they share state.

The module approach works well. If I absolutely need a singleton I prefer the Metaclass approach.
class Singleton(type):
def __init__(cls, name, bases, dict):
super(Singleton, cls).__init__(name, bases, dict)
cls.instance = None
def __call__(cls,*args,**kw):
if cls.instance is None:
cls.instance = super(Singleton, cls).__call__(*args, **kw)
return cls.instance
class MyClass(object):
__metaclass__ = Singleton

See this implementation from PEP318, implementing the singleton pattern with a decorator:
def singleton(cls):
instances = {}
def getinstance():
if cls not in instances:
instances[cls] = cls()
return instances[cls]
return getinstance
#singleton
class MyClass:
...

The Python documentation does cover this:
class Singleton(object):
def __new__(cls, *args, **kwds):
it = cls.__dict__.get("__it__")
if it is not None:
return it
cls.__it__ = it = object.__new__(cls)
it.init(*args, **kwds)
return it
def init(self, *args, **kwds):
pass
I would probably rewrite it to look more like this:
class Singleton(object):
"""Use to create a singleton"""
def __new__(cls, *args, **kwds):
"""
>>> s = Singleton()
>>> p = Singleton()
>>> id(s) == id(p)
True
"""
it_id = "__it__"
# getattr will dip into base classes, so __dict__ must be used
it = cls.__dict__.get(it_id, None)
if it is not None:
return it
it = object.__new__(cls)
setattr(cls, it_id, it)
it.init(*args, **kwds)
return it
def init(self, *args, **kwds):
pass
class A(Singleton):
pass
class B(Singleton):
pass
class C(A):
pass
assert A() is A()
assert B() is B()
assert C() is C()
assert A() is not B()
assert C() is not B()
assert C() is not A()
It should be relatively clean to extend this:
class Bus(Singleton):
def init(self, label=None, *args, **kwds):
self.label = label
self.channels = [Channel("system"), Channel("app")]
...

As the accepted answer says, the most idiomatic way is to just use a module.
With that in mind, here's a proof of concept:
def singleton(cls):
obj = cls()
# Always return the same object
cls.__new__ = staticmethod(lambda cls: obj)
# Disable __init__
try:
del cls.__init__
except AttributeError:
pass
return cls
See the Python data model for more details on __new__.
Example:
#singleton
class Duck(object):
pass
if Duck() is Duck():
print "It works!"
else:
print "It doesn't work!"
Notes:
You have to use new-style classes (derive from object) for this.
The singleton is initialized when it is defined, rather than the first time it's used.
This is just a toy example. I've never actually used this in production code, and don't plan to.

I'm very unsure about this, but my project uses 'convention singletons' (not enforced singletons), that is, if I have a class called DataController, I define this in the same module:
_data_controller = None
def GetDataController():
global _data_controller
if _data_controller is None:
_data_controller = DataController()
return _data_controller
It is not elegant, since it's a full six lines. But all my singletons use this pattern, and it's at least very explicit (which is pythonic).

The one time I wrote a singleton in Python I used a class where all the member functions had the classmethod decorator.
class Foo:
x = 1
#classmethod
def increment(cls, y=1):
cls.x += y

Creating a singleton decorator (aka an annotation) is an elegant way if you want to decorate (annotate) classes going forward. Then you just put #singleton before your class definition.
def singleton(cls):
instances = {}
def getinstance():
if cls not in instances:
instances[cls] = cls()
return instances[cls]
return getinstance
#singleton
class MyClass:
...

There are also some interesting articles on the Google Testing blog, discussing why singleton are/may be bad and are an anti-pattern:
Singletons are Pathological Liars
Where Have All the Singletons Gone?
Root Cause of Singletons

I think that forcing a class or an instance to be a singleton is overkill. Personally, I like to define a normal instantiable class, a semi-private reference, and a simple factory function.
class NothingSpecial:
pass
_the_one_and_only = None
def TheOneAndOnly():
global _the_one_and_only
if not _the_one_and_only:
_the_one_and_only = NothingSpecial()
return _the_one_and_only
Or if there is no issue with instantiating when the module is first imported:
class NothingSpecial:
pass
THE_ONE_AND_ONLY = NothingSpecial()
That way you can write tests against fresh instances without side effects, and there is no need for sprinkling the module with global statements, and if needed you can derive variants in the future.

The Singleton Pattern implemented with Python courtesy of ActiveState.
It looks like the trick is to put the class that's supposed to only have one instance inside of another class.

class Singleton(object[,...]):
staticVar1 = None
staticVar2 = None
def __init__(self):
if self.__class__.staticVar1==None :
# create class instance variable for instantiation of class
# assign class instance variable values to class static variables
else:
# assign class static variable values to class instance variables

class Singeltone(type):
instances = dict()
def __call__(cls, *args, **kwargs):
if cls.__name__ not in Singeltone.instances:
Singeltone.instances[cls.__name__] = type.__call__(cls, *args, **kwargs)
return Singeltone.instances[cls.__name__]
class Test(object):
__metaclass__ = Singeltone
inst0 = Test()
inst1 = Test()
print(id(inst1) == id(inst0))

OK, singleton could be good or evil, I know. This is my implementation, and I simply extend a classic approach to introduce a cache inside and produce many instances of a different type or, many instances of same type, but with different arguments.
I called it Singleton_group, because it groups similar instances together and prevent that an object of the same class, with same arguments, could be created:
# Peppelinux's cached singleton
class Singleton_group(object):
__instances_args_dict = {}
def __new__(cls, *args, **kwargs):
if not cls.__instances_args_dict.get((cls.__name__, args, str(kwargs))):
cls.__instances_args_dict[(cls.__name__, args, str(kwargs))] = super(Singleton_group, cls).__new__(cls, *args, **kwargs)
return cls.__instances_args_dict.get((cls.__name__, args, str(kwargs)))
# It's a dummy real world use example:
class test(Singleton_group):
def __init__(self, salute):
self.salute = salute
a = test('bye')
b = test('hi')
c = test('bye')
d = test('hi')
e = test('goodbye')
f = test('goodbye')
id(a)
3070148780L
id(b)
3070148908L
id(c)
3070148780L
b == d
True
b._Singleton_group__instances_args_dict
{('test', ('bye',), '{}'): <__main__.test object at 0xb6fec0ac>,
('test', ('goodbye',), '{}'): <__main__.test object at 0xb6fec32c>,
('test', ('hi',), '{}'): <__main__.test object at 0xb6fec12c>}
Every object carries the singleton cache... This could be evil, but it works great for some :)

My simple solution which is based on the default value of function parameters.
def getSystemContext(contextObjList=[]):
if len( contextObjList ) == 0:
contextObjList.append( Context() )
pass
return contextObjList[0]
class Context(object):
# Anything you want here

Being relatively new to Python I'm not sure what the most common idiom is, but the simplest thing I can think of is just using a module instead of a class. What would have been instance methods on your class become just functions in the module and any data just becomes variables in the module instead of members of the class. I suspect this is the pythonic approach to solving the type of problem that people use singletons for.
If you really want a singleton class, there's a reasonable implementation described on the first hit on Google for "Python singleton", specifically:
class Singleton:
__single = None
def __init__( self ):
if Singleton.__single:
raise Singleton.__single
Singleton.__single = self
That seems to do the trick.

Singleton's half brother
I completely agree with staale and I leave here a sample of creating a singleton half brother:
class void:pass
a = void();
a.__class__ = Singleton
a will report now as being of the same class as singleton even if it does not look like it. So singletons using complicated classes end up depending on we don't mess much with them.
Being so, we can have the same effect and use simpler things like a variable or a module. Still, if we want use classes for clarity and because in Python a class is an object, so we already have the object (not and instance, but it will do just like).
class Singleton:
def __new__(cls): raise AssertionError # Singletons can't have instances
There we have a nice assertion error if we try to create an instance, and we can store on derivations static members and make changes to them at runtime (I love Python). This object is as good as other about half brothers (you still can create them if you wish), however it will tend to run faster due to simplicity.

In cases where you don't want the metaclass-based solution above, and you don't like the simple function decorator-based approach (e.g. because in that case static methods on the singleton class won't work), this compromise works:
class singleton(object):
"""Singleton decorator."""
def __init__(self, cls):
self.__dict__['cls'] = cls
instances = {}
def __call__(self):
if self.cls not in self.instances:
self.instances[self.cls] = self.cls()
return self.instances[self.cls]
def __getattr__(self, attr):
return getattr(self.__dict__['cls'], attr)
def __setattr__(self, attr, value):
return setattr(self.__dict__['cls'], attr, value)