__post_init vs new vs init__: When Order Matters

#python #dataclasses #objectoriented #initialization

The Hook That Breaks Dataclasses

Here's a test: what does this print?

from dataclasses import dataclass

@dataclass
class User:
    name: str
    email: str

    def __init__(self, name, email):
        print(f"__init__ called: {name}")
        self.name = name.upper()
        self.email = email

    def __post_init__(self):
        print(f"__post_init__ called: {self.name}")
        self.name = self.name.lower()

u = User("Alice", "alice@example.com")
print(u.name)

If you guessed alice, you're wrong. If you guessed ALICE, you're also wrong. The answer is TypeError: __init__() missing 2 required positional arguments. The dataclass decorator generates its own __init__, and your manual one gets overridden. But if you remove the manual __init__, __post_init__ runs after the generated one, giving you alice.

This is the kind of thing you learn when you've spent an afternoon debugging why your validation logic runs in the wrong order.