Recca Tsai

Posted on Apr 2 • Originally published at recca0120.github.io

polyfactory: Stop Hand-Writing Test Fixtures, Let Type Hints Do It

#productivity #python #testing #tooling

Originally published at recca0120.github.io

Preparing fake data for tests is its own kind of overhead.

To test a user system, you need to build a User object — fill in id, email, name, created_at. If you're testing orders, you need an Order, which contains a list of OrderItem. Hand-crafting all of that often takes longer than writing the actual test logic.

polyfactory solves this: give it a class with type hints, and it generates conforming fake data automatically.

Install

pip install polyfactory

With Pydantic:

pip install polyfactory pydantic

Basic Usage: dataclass

from dataclasses import dataclass
from polyfactory.factories import DataclassFactory

@dataclass
class User:
    id: int
    name: str
    email: str
    is_active: bool

class UserFactory(DataclassFactory):
    __model__ = User

user = UserFactory.build()
# User(id=42, name='vDjhqXt', email='KpLmn@example.com', is_active=True)

Every build() call produces different values — id is a random int, email is a random string, is_active is random True/False.

Pydantic v2

from pydantic import BaseModel
from polyfactory.factories.pydantic_factory import ModelFactory

class Order(BaseModel):
    id: int
    amount: float
    status: str
    items: list[str]

class OrderFactory(ModelFactory):
    __model__ = Order

order = OrderFactory.build()
# Order(id=7, amount=3.14, status='aBcD', items=['x', 'y'])

Pydantic validators still run — polyfactory won't generate data that fails validation.

Overriding Specific Fields

Most of the time you only care about a few fields — just pass them in:

# only set status, everything else is auto-filled
order = OrderFactory.build(status="paid")

# or set defaults in the factory class
class PaidOrderFactory(OrderFactory):
    status = "paid"
    amount = 100.0

This is the pattern I use most: base factory handles the bulk of the data, override the fields that matter for the specific test. No need to hardcode the whole object every time.

Batch Generation

users = UserFactory.batch(10)
# gives you 10 different User objects

Useful when testing a list or pagination logic.

Combining With pytest Fixtures

# conftest.py
import pytest
from polyfactory.factories import DataclassFactory

@pytest.fixture
def user():
    return UserFactory.build()

@pytest.fixture
def active_user():
    return UserFactory.build(is_active=True)

@pytest.fixture
def users():
    return UserFactory.batch(5)

Fixtures return factory-generated objects. The test doesn't need to know what fields User has — only the fields actually relevant to that test need to be specified.

def test_deactivate_user(active_user):
    deactivate(active_user)
    assert not active_user.is_active

def test_list_users(users):
    result = get_user_list(users)
    assert len(result) == 5

TypedDict and attrs

from typing import TypedDict
from polyfactory.factories import TypedDictFactory

class Config(TypedDict):
    host: str
    port: int
    debug: bool

class ConfigFactory(TypedDictFactory):
    __model__ = Config

config = ConfigFactory.build()
# {'host': 'abc', 'port': 8080, 'debug': False}

Nested Objects

polyfactory handles nested types recursively:

@dataclass
class Address:
    city: str
    country: str

@dataclass
class User:
    name: str
    address: Address  # nested, auto-generated

class UserFactory(DataclassFactory):
    __model__ = User

user = UserFactory.build()
# user.address is also auto-generated

No need to build AddressFactory separately and pass it in.

polyfactory vs Faker

Faker also generates fake data, but you tell it what you want:

from faker import Faker
fake = Faker()
email = fake.email()
name = fake.name()

polyfactory is "give me a class, I'll fill it":

user = UserFactory.build()

They're not competing — they serve different needs:

Faker: You care what each field looks like (realistic emails, real city names, etc.)
polyfactory: You just need the types to be correct, the specific values don't matter

Test logic usually doesn't care if an email looks real — it just needs to be a string. For that case, polyfactory is simpler.

Summary

polyfactory's core idea is one thing: types are the specification, factories generate data according to types.

Combined with pytest fixtures, getting test data down to one or two lines is straightforward — you can focus on what the test is actually checking.

DEV Community