Unlock Python Metaprogramming: Master Metaclasses, Singletons, and Decorators

Many people don't understand what "metaprogramming" is; this article discusses Python metaprogramming, even if it doesn't perfectly match the formal definition.

The subtitle "control everything you want to control" reflects the goal of using Python's features to make code elegant and concise, modifying abstractions at a higher level.

In Python everything is an object. Special methods, metaclasses, and other mechanisms constitute metaprogramming. Adding attributes dynamically is not considered metaprogramming in Python, though it is in static languages.

Objects exist at two levels: instance objects and class objects. Types are also objects, introducing a third level. Metaclasses are "classes of classes", adding another level.

We also distinguish ImportTime and RunTime. ImportTime occurs when a module is imported: global statements execute, function objects are created, class objects are created, but their bodies are not run. RunTime executes function and method bodies when called.

Metaclasses and classes are created at ImportTime; instances are created at RunTime, unless instantiated at module level.

To control instance creation, override __init__ in the class. To control class behavior, use a metaclass.

Singleton pattern: the simplest implementation is shown.

Using a metaclass to enforce a singleton: define a metaclass inheriting from type with custom __init__ and __call__ methods, then set metaclass=Singleton in the class definition.

The metaclass itself is a subclass of type, which is a subclass of object. Custom metaclasses must subclass type.

Decorators are often seen as difficult, but once functions are objects, writing decorators is straightforward. The @wraps decorator preserves the original function's signature.

Class decorators and decorator factories can be written similarly.

Descriptors allow sharing behavior across multiple classes without repeating code. A descriptor class defines methods like __set_name__ and __get__ . An example is shown.

Since Python 3.6, __init_subclass__ can replace metaclasses for customizing subclass creation.

Conclusion : Metaprogramming techniques such as metaclasses and descriptors are powerful but often unnecessary; they are used in many frameworks to simplify user code. To deepen understanding, consult resources like "Fluent Python", "Python Cookbook", the official descriptor HowTo, the Data Model documentation, and the Python source code.