Python Descriptor Protocol

Also Known As

TL;DR

Explanation

A descriptor is any object defining at least one of __get__(self, obj, objtype), __set__(self, obj, value), or __delete__(self, obj). When such an object is assigned as a class attribute, Python routes attribute access through these methods instead of returning the descriptor itself. This is fundamentally different from __getattr__/__getattribute__, which live on the instance's class and act as fallbacks for the whole object: descriptors are per-attribute and resolved during the normal attribute lookup chain. There are two kinds. A data descriptor defines __set__ or __delete__ (plus usually __get__) and takes priority over the instance __dict__ — this is why property assignments are intercepted even when an instance has a same-named entry. A non-data descriptor defines only __get__ and is shadowed by instance __dict__ entries, which is how functions become bound methods yet can be overridden per instance. The lookup order for instance.attr is: data descriptors on the type, then instance __dict__, then non-data descriptors and class attributes. Built-ins property, classmethod, staticmethod, and functions are all descriptors. Since Python 3.6, __set_name__(self, owner, name) is called automatically when the owning class is created, letting a descriptor learn the attribute name it was bound to without repeating it. Descriptors shine when you need reusable, validated, computed, or lazily-loaded attributes across many fields — ORMs (Django, SQLAlchemy), typed-field libraries, and caching decorators rely on them. The common error is storing per-instance state on the descriptor itself; because one descriptor instance is shared by every instance of the owning class, state must live in the instance __dict__ or a WeakKeyDictionary keyed by instance, not on self. For single computed attributes, property is the simpler, idiomatic choice; reach for a custom descriptor only when the same access logic must be reused across multiple attributes or classes.

Common Misconception

Why It Matters

Common Mistakes

• Storing per-instance values on the descriptor object (self.value), so every instance shares and overwrites the same state.
• Confusing data and non-data descriptors and being surprised that instance __dict__ shadows a __get__-only descriptor.
• Hardcoding the attribute name instead of using __set_name__ to learn it automatically in Python 3.6+.
• Reaching for a custom descriptor when a simple property would express a single computed attribute more clearly.
• Forgetting to handle obj is None in __get__, which breaks class-level access like MyClass.field.

Avoid When

• A single computed or validated attribute is needed — a property is simpler and more readable.
• The logic is not reused across multiple attributes or classes, so the indirection adds no value.
• Team members are unfamiliar with the data model and the implicit lookup order would obscure behavior.

When To Use

• The same access, validation, or caching logic must be reused across many attributes or classes.
• Building field abstractions for ORMs, typed-config, or lazy-loaded computed properties.
• You need data-descriptor precedence to guarantee interception even when instance __dict__ holds a same-named key.

Code Examples

class Positive:
    def __get__(self, obj, objtype=None):
        return self.value          # shared across ALL instances!
    def __set__(self, obj, value):
        if value < 0:
            raise ValueError('must be positive')
        self.value = value         # bug: stored on descriptor, not obj

class Account:
    balance = Positive()

a = Account(); a.balance = 100
b = Account(); b.balance = 5
print(a.balance)  # 5 — a and b share descriptor state

class Positive:
    def __set_name__(self, owner, name):
        self.name = name           # learn the attribute name

    def __get__(self, obj, objtype=None):
        if obj is None:
            return self            # class-level access
        return obj.__dict__[self.name]

    def __set__(self, obj, value):
        if value < 0:
            raise ValueError('must be positive')
        obj.__dict__[self.name] = value  # per-instance storage

class Account:
    balance = Positive()

a = Account(); a.balance = 100
b = Account(); b.balance = 5
print(a.balance)  # 100 — state isolated per instance

References

• https://docs.python.org/3/howto/descriptor.html
• https://docs.python.org/3/reference/datamodel.html#descriptors
• https://docs.python.org/3/reference/datamodel.html#object.__set_name__

Tags

Related Terms