Python: always use __new__ instead of __init__?

夙愿已清 提交于 2019-11-28 06:48:55

So, the class of a class is typically type, and when you call Class() the __call__() method on Class's class handles that. I believe type.__call__() is implemented more or less like this:

def __call__(cls, *args, **kwargs):
    # should do the same thing as type.__call__
    obj = cls.__new__(cls, *args, **kwargs)
    if isinstance(obj, cls):
        obj.__init__(*args, **kwargs)
    return obj

The direct answer to your question is no, the things that __init__() can do (change / "initialize" a specified instance) is a subset of the things that __new__() can do (create or otherwise select whatever object it wants, do anything to that object it wants before the object is returned).

It's convenient to have both methods to use, however. The use of __init__() is simpler (it doesn't have to create anything, it doesn't have to return anything), and I believe it is best practice to always use __init__() unless you have a specific reason to use __new__().

One possible answer from guido's post (thanks @fraca7):

For example, in the pickle module, __new__ is used to create instances when unserializing objects. In this case, instances are created, but the __init__ method is not invoked.

Any other similar answers?


I'm accepting this answer as a 'yes' to my own question:

I'm wondering if there is anything __init__ can do that __new__ cannot?

Yes, unlike __new__, actions that you put in the __init__ method will not be performed during the unpickling process. __new__ cannot make this distinction.

Well, looking for __new__ vs __init__ on google showed me this.

Long story short, __new__ returns a new object instance, while __init__ returns nothing and just initializes class members.

EDIT: To actually answer your question, you should never need to override __new__ unless you are subclassing immutable types.

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!