为什么我必须显式继承插槽的对象才能像宣传的那样工作？

Question

A colleague of mine recently showed me the following session:

>>> class Foo:
...     __slots__ = ['x']
...     def __init__(self):
...             self.x = "x"
... 
>>> f = Foo()
>>> f.x
'x'
>>> f.y = 1
>>> class Bar(object):
...     __slots__ = ['x']
...     def __init__(self):
...             self.x = "x"
... 
>>> b = Bar()
>>> b.x
'x'
>>> b.y = 1
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
AttributeError: 'Bar' object has no attribute 'y'

According to the Python documentation , defining __slots__ should make it impossible to assign any other variables than the ones specified in slots, unless the user provides a dict instance manually:

The documentation says nothing about an explicit need to inherit from object like Bar does.

Why does this happen?

Answer 1

It does say it, just not very explicitly:

The default can be overridden by defining __slots__ in a new-style class definition.

In Python 2, when you inherit from object , you're creating a new-style class. If you don't, it's an old-style class.

Answer 2

As Jamie pointed out, you want a new-style class. here's why:

In [5]: import sys

In [6]: sys.getsizeof(f)
Out[6]: 36

In [7]: sys.getsizeof(b)
Out[7]: 28

In [8]: sys.getsizeof(f.__dict__)
Out[8]: 140

In [9]: sys.getsizeof(b.__dict__)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-9-891152de459e> in <module>()
----> 1 sys.getsizeof(b.__dict__)

AttributeError: 'Bar' object has no attribute '__dict__'

In other words, your Foo class doesn't actually use slots, it still uses __dict__ to store member variables, thus total size of f is 176 bytes and b 28 bytes (figures for 32-bit arch). It is also why you can assign fy = 1 .

Presumably you want __slots__ to reduce memory usage or to force users of your objects to touch only slotted variables. Bar does both. Foo does neither.