Metaprogramming

The final part is the bottom of the language: where attribute access, class creation, and the rules of inheritance themselves become things you can program against. Most code never needs this. The libraries you depend on — dataclasses, attrs, ORM column descriptors, framework auto-registration — were written with these tools.

• 34  Dynamic Attributes and Properties covers the runtime side of attribute access: __getattr__, __setattr__, @property, and the precedence rules that decide which one fires.

• 35  Attribute Descriptors is the deepest single chapter in Fluent Python. A descriptor is an object that lives on the class and intercepts attribute access on the instance. Once you understand descriptors, you understand @property, @classmethod, @staticmethod, and almost every “magic” attribute in the standard library.

• 36  Class Metaprogramming closes the book with the controlled use of type-as-metaclass and the lighter-weight alternative, __init_subclass__ (PEP 487). The chapter ends with the same advice it begins with: don’t reach for metaclasses unless you’ve already failed with the simpler tools.