Descriptors - The Magic behind python - Raymond Hettinger

What is a descriptor

It is like a magic dot…

Reading the dot, invokes its magic. has all the gory details.


An object that defines __get__, __set__ and __delete__.

Must be put into the class dictionary, not in the instance dictionary. If you put it in the instance dictionary, it will not be invoked. It is only invoked when it is put into the class dictionary.


  • __get__ is invoked when you read the

  • __set__ when something is assigned to the object.

  • __del__ when the object is deleted?


Property is a descriptor.

It is trivially easy to write your own variants.

The following two __getattribute__ methods are not the same:

  • A.x translates to type.__getattribute__(A, 'x')

  • a.x translates to object.__getattribute__(a, 'x')

If we override __getattribute__ you can create your own new types of magic for dotted access.

Every time you see a . think __getattribute__.

Super provides it’s own __getattribute__. It’s special trick is to search the __mro__ during dotted access.


Functions are descriptors, running dir(f) shows that functions have a __get__ methods.

If you put a method in a class dictionary, the __get__ method will activate upon dotted access.


When a class is created, space is pre-allocated for each slot.

How python works

Dotted attribute access like A.x or a.x calls the __getattribute__ method.

Useful Methods (nothing to do with the talk)


How to describe __

  • double under

  • under, under

  • dunder