Home / Questions / 351477
Accepted
Julio Cesar
Asked: 2020-05-02 17:18:22 +0800 CST 2020-05-02 17:18:22 +0800 CST

RecursionError: maximum recursion depth exceeded

  • 772

I was doing some tests with python trying to "force", so to speak, the encapsulation in python, since in python private attributes do not exist as such:

import re

class Mixin(object):
    __a = 1

    def __getattribute__(self, name):
        if re.match(f'_{ self.__class__.__name__ }__', name):
            raise AttributeError(f"'{ self.__class__.__name__ }' object has no attribute '{ name }'")

        return super().__getattribute__(name)

Mixin()._Mixin__a

It is assumed that in this way the attribute could not be accessed __alike this: _Mixin__a.

But it gives me the following error:

Traceback (most recent call last):
  File "/home/lcteen/Documentos/Programming/Python/Practices/sss.py", line 24, in <module>
    Mixin()._Mixin__a
  File "/home/lcteen/Documentos/Programming/Python/Practices/sss.py", line 19, in __getattribute__
    if re.match(f'_{ self.__class__.__name__ }__', name):
  File "/home/lcteen/Documentos/Programming/Python/Practices/sss.py", line 19, in __getattribute__
    if re.match(f'_{ self.__class__.__name__ }__', name):
  File "/home/lcteen/Documentos/Programming/Python/Practices/sss.py", line 19, in __getattribute__
    if re.match(f'_{ self.__class__.__name__ }__', name):
  [Previous line repeated 996 more times]
RecursionError: maximum recursion depth exceeded

I do not understand what I am doing wrong, nor why the error occurs.

python

1 Answers

  1. Best Answer

    It is perhaps the most common error when reimplementing __getattribute__. The minimum code to reproduce the problem is very basic:

    class Mixin:
    def __getattribute__(self, name):
        self.__class__

Mixin().test

    Note that when implemented __getattribute__, such a method is always the entry point for any access to attributes of the instance.

    Let's analyze the previous example:

    1. Mixin().testcreates an instance of the class and tries to access the attribute test.

    2. The above causes the method to be called as __getattribute__follows:

      self.__getattribute__("test")

    3. The line is executed self.__class__, causing it to be called again __getattribute__:

      self.__getattribute__("__class__")

    4. The line is reached again self.__class__.... And so on until the stack is full and the interpreter explodes, well no, because Python limits the allowed recursive calls to 1000 by default and shows us the aforementioned error.

    The solution in these cases is always to delegate the resolution of the attribute to the parent class ( objectin this case), thus avoiding the recursive call. The normal thing is to use superfor it:

    class Mixin:
    __a = 1

    def __getattribute__(self, name):
        class_ =  super().__getattribute__("__class__")
        if re.match(f"_{ class_.__name__ }__", name):
            raise AttributeError(f"'{ class_.__name__ }' object has no attribute '{ name }'")
        return super().__getattribute__(name)

    super ()returns a proxy object that will look up any methods that can be found below in the base classes following the MRO. If no such method exists, it will fail with a AttributeErrorbut never call the original method.

    Alternatively, you can call the method __getattribute__directly objet. The C implementation of the method is always the endpoint in the MRO for attribute access and has direct access to__dict__ .

    class Mixin:
    __a = 1

    def __getattribute__(self, name):
        class_ =  object.__getattribute__(self, "__class__")
        if re.match(f"_{ class_.__name__ }__", name):
            raise AttributeError(f"'{ class_.__name__ }' object has no attribute '{ name }'")
        return super().__getattribute__(name)
    • 2