Checking for member existence in Python

Question

I regularly want to check if an object has a member or not. An example is the creation of a singleton in a function. For that purpose, you can use hasattr like this:

class Foo(object):
    @classmethod
    def singleton(self):
        if not hasattr(self, 'instance'):
            self.instance = Foo()
        return self.instance

But you can also do this:

class Foo(object):
    @classmethod
    def singleton(self):
        try:
            return self.instance
        except AttributeError:
            self.instance = Foo()
            return self.instance

Is one method better of the other?

Edit: Added the @classmethod ... But note that the question is not about how to make a singleton but how to check the presence of a member in an object.

Edit: For that example, a typical usage would be:

s = Foo.singleton()

Then s is an object of type Foo, the same each time. And, typically, the method is called many times.

Answer 1

It depends on which case is "typical", because exceptions should model, well, atypical conditions. So, if the typical case is that the instance attribute should exist, then use the second code style. If not having instance is as typical as having instance, then use the first style.

In the specific case of creating a singleton, I'm inclined to go with the first style, because creating a singleton the initial time is a typical use case. :-)

Answer 2

I just tried to measure times:

class Foo(object):
    @classmethod
    def singleton(self):
        if not hasattr(self, 'instance'):
            self.instance = Foo()
        return self.instance



class Bar(object):
    @classmethod
    def singleton(self):
        try:
            return self.instance
        except AttributeError:
            self.instance = Bar()
            return self.instance



from time import time

n = 1000000
foo = [Foo() for i in xrange(0,n)]
bar = [Bar() for i in xrange(0,n)]

print "Objs created."
print


for times in xrange(1,4):
    t = time()
    for d in foo: d.singleton()
    print "#%d Foo pass in %f" % (times, time()-t)

    t = time()
    for d in bar: d.singleton()
    print "#%d Bar pass in %f" % (times, time()-t)

    print

On my machine:

Objs created.

#1 Foo pass in 1.719000
#1 Bar pass in 1.140000

#2 Foo pass in 1.750000
#2 Bar pass in 1.187000

#3 Foo pass in 1.797000
#3 Bar pass in 1.203000

It seems that try/except is faster. It seems also more readable to me, anyway depends on the case, this test was very simple maybe you'd need a more complex one.

Answer 3

I have to agree with Chris. Remember, don't optimize until you actually need to do so. I really doubt checking for existence is going to be a bottleneck in any reasonable program.

I did see http://code.activestate.com/recipes/52558/ as a way to do this, too. Uncommented copy of that code ("spam" is just a random method the class interface has):

class Singleton:
    class __impl:
        def spam(self):
            return id(self)
    __instance = None
    def __init__(self):
        if Singleton.__instance is None:
            Singleton.__instance = Singleton.__impl()
        self.__dict__['_Singleton__instance'] = Singleton.__instance
    def __getattr__(self, attr):
        return getattr(self.__instance, attr)
    def __setattr__(self, attr, value):
        return setattr(self.__instance, attr, value)

Answer 4

These are two different methodologies: №1 is LBYL (look before you leap) and №2 is EAFP (easier to ask forgiveness than permission).

Pythonistas typically suggest that EAFP is better, with arguments in style of "what if a process creates the file between the time you test for it and the time you try to create it yourself?". This argument does not apply here, but it's the general idea. Exceptions should not be treated as too exceptional.

Performance-wise in your case —since setting up exception managers (the try keyword) is very cheap in CPython while creating an exception (the raise keyword and internal exception creation) is what is relatively expensive— using method №2 the exception would be raised only once; afterwards, you just use the property.

Answer 5

A little off-topic in the way of using it. Singletons are overrated, and a "shared-state" method is as effective, and mostly, very clean in python, for example:

class Borg:
    __shared_state = {}
    def __init__(self):
        self.__dict__ = self.__shared_state
    # and whatever else you want in your class -- that's all!

Now every time you do:

obj = Borg()

it will have the same information, or, be somewhat the same instance.