Beware of keyword's default values

Daniele Esposti's Blog
, in 26 March 2013

Look this class definition, do you see what’s is wrong and potentially generate big problems in you code?

class Foo(object):
    
    def __init__(self, bars=[]):
        self.bars = bars

<!–more–!>

Before jumping straight to the solution lets play with some instances of that class:

>>> foo1=Foo([1, 2, 3])
>>> foo2=Foo([3, 4, 5])
>>> foo1.bars
[1, 2, 3]
>>> foo2.bars
[3, 4, 5]
>>> foo1.bars.append(10)
>>> foo2.bars.append(20)
>>> foo1.bars
[1, 2, 3, 10]
>>> foo2.bars
[3, 4, 5, 20]

Nothing really interesting here: I created two instances of Foo class passing two different list of numbers and added a value to both bars attributes.

Boring and useless.

I’ll try now using the default value for bars attribute defined in the class’ constructor:

>>> foo3=Foo()
>>> foo4=Foo()
>>> foo3.bars
[]
>>> foo4.bars
[]
>>> foo3.bars.append(10)
>>> foo4.bars.append(20)
>>> foo3.bars
[10, 20]
>>> foo4.bars
[10, 20]

Wait a sec! I’ve added a single number to each attribute to each class but why both of them have the same content?

Looking back at the definition of the constructor the bar keyword is optional and has a empty list as default value; but the list instance is created only once when the class is parsed by the interpreter.

So both parameters in foo3 and foo4 instances are using the same list instance and not a copy, that’s why modifying the attribute in one instance alters the attribute the other one.

As a rule of thumb when you are defining default values for methods/function keywords pay attention at the type of the default value. It’s safe to use immutable types (strings, integers, floats, None) but mutable types can lead to bugs in your application difficult to track down.

Coming back to my example here a safer way to implement the class:

class Foo(object):
    
    def __init__(self, bars=None):
        self.bars = bars if bars is None else []

This way a new list instance will be created in case the bars keyword is null.