litestar-org/litestar-skills/polyfactory

polyfactory

Polyfactory is a typed mock-data factory library: declare ModelFactory[T] (or DataclassFactory, MsgspecFactory, AttrsFactory, TypedDictFactory) and .build() returns a populated instance of T. The matching factory base inspects the model's annotations and supported constraints so tests do not need hand-written happy-path fixtures.

In Litestar projects, polyfactory's pytest plugin is the canonical way to feed TestClient.post(...) / AsyncTestClient.put(...) payloads. The companion skill litestar:litestar-testing covers the request side.

Code Style Rules

• PEP 604 unions: T | None, never Optional[T].
• One factory per model. Don't reuse a factory across unrelated models — clarity beats DRY when the test fails at 3am.
• Pick the right factory base by backend: ModelFactory (Pydantic), DataclassFactory, MsgspecFactory, AttrsFactory, TypedDictFactory. A mismatched base raises ConfigurationException.
• Prefer register_fixture over hand-rolled @pytest.fixture wrappers. Call it in a pytest-discovered test module or conftest.py so the injected fixture is collected.

Quick Reference

Picking the right factory base

polyfactory.factories.__all__ only exports BaseFactory, DataclassFactory, and TypedDictFactory. Always import backend-specific factories (ModelFactory, MsgspecFactory, AttrsFactory, SQLAlchemyFactory, BeanieDocumentFactory, OdmanticModelFactory) from their concrete submodules (polyfactory.factories.pydantic_factory, polyfactory.factories.msgspec_factory, polyfactory.factories.sqlalchemy_factory, etc.), register_fixture from polyfactory.pytest_plugin, and post_generated from polyfactory.decorators so runtime imports and mypy/pyright checks succeed.

Defining a factory

from dataclasses import dataclass
from polyfactory.factories.dataclass_factory import DataclassFactory
@dataclass
class Order:
    id: int
    customer_email: str
    total_cents: int
    status: str
class OrderFactory(DataclassFactory[Order]):
    pass
# Use it
one = OrderFactory.build()
many = OrderFactory.batch(10)

The single concrete generic argument lets Polyfactory infer __model__. build() returns one instance, batch(n) returns list[T], coverage() yields the smallest set of instances that covers the model's supported variants, create_sync() / create_batch_sync(n) and create_async() / create_batch_async(n) build and persist instances, seed_random(seed) re-seeds the factory PRNG and Faker instance, and add_provider(type_, fn) registers a custom type generator.

Customizing fields

from polyfactory import Ignore, Require, Use
from polyfactory.decorators import post_generated
from polyfactory.factories import DataclassFactory
class OrderFactory(DataclassFactory[Order]):
    # Plain literal — every build returns this exact value
    status = "pending"
    # Callable — re-evaluated per build
    customer_email = Use(lambda: "test@example.com")
    # Random choice — re-evaluated per build
    total_cents = Use(DataclassFactory.__random__.randint, 100, 10_000)
    # Skip generation so the model's default / default_factory runs
    internal_note = Ignore()
    # Force callers to supply this kwarg on build() / batch() / coverage()
    tenant_id = Require()
    # Derive a field from already-generated fields on the same instance
    @post_generated
    @classmethod
    def reference(cls, id: int) -> str:
        return f"order-{id}"

Use(callable, *args, **kwargs) is re-invoked on every build(), so each generated instance gets a fresh value. Ignore() and Require() must be instantiated; omitting a Require() field at build time raises MissingBuildKwargException.

For standalone post-generation callbacks, use PostGenerated(fn, *args, **kwargs) where fn has signature (field_name, generated_values, *args, **kwargs):

from typing import Any
from polyfactory import PostGenerated
def order_reference(name: str, values: dict[str, Any], prefix: str) -> str:
    return f"{prefix}-{values['id']}"
class OrderFactory(DataclassFactory[Order]):
    reference = PostGenerated(order_reference, "order")

Determinism and configuration

class OrderFactory(DataclassFactory[Order]):
    __random_seed__ = 42  # seeds cls.__random__ and cls.__faker__ at class creation
OrderFactory.seed_random(42)  # re-seed at runtime before a test or build

Set __random_seed__ (or call Factory.seed_random(seed) / __faker__.seed_instance(seed) on a custom __faker__ = Faker(locale="en_US")) when test assertions depend on the exact generated values. Never pass Faker(seed=...) — Faker.__init__ ignores a seed keyword argument without seeding.

Core BaseFactory configuration attributes:

• __model__: target model class (inferred from a single concrete generic argument when omitted).
• __check_model__ = True: validates that Use, PostGenerated, Ignore, and Require attribute names exist on __model__.
• __random_seed__: integer seed passed to cls.seed_random(...) during class creation.
• __use_defaults__ = False: when True, prefers model field defaults / default_factory over generated values.
• __allow_none_optionals__ = True: when True, optional fields (T | None) randomly produce None; set False to always populate T.
• __set_as_default_factory_for_type__ = False: when True, registers the factory globally for nested fields of type __model__.
• __faker__ and __random__: the factory's Faker and random.Random instances.

Default factory registration

class CustomerFactory(DataclassFactory[Customer]):
    __set_as_default_factory_for_type__ = True
@dataclass
class Order:
    id: int
    customer: Customer  # automatically populated by CustomerFactory.build()
class OrderFactory(DataclassFactory[Order]):
    pass

When __set_as_default_factory_for_type__ = True, polyfactory uses that factory whenever the type appears as a field on another model — no manual nesting required.

Pytest fixture from a factory

from polyfactory.pytest_plugin import register_fixture
from polyfactory.factories.dataclass_factory import DataclassFactory
@register_fixture
class OrderFactory(DataclassFactory[Order]):
    pass
def test_order_total(order_factory: type[OrderFactory]) -> None:
    order = order_factory.build()
    assert order.total_cents >= 0

@register_fixture returns the class unchanged and injects a pytest fixture into the caller's module. The fixture name is the snake-cased class name and its value is the factory class.

Cross-model relationships

from polyfactory import Use
from polyfactory.pytest_plugin import register_fixture
@register_fixture
class CustomerFactory(DataclassFactory[Customer]):
    __set_as_default_factory_for_type__ = True
@register_fixture
class OrderFactory(DataclassFactory[Order]):
    customer = Use(CustomerFactory.build)  # explicit local override

__set_as_default_factory_for_type__ = True lets Polyfactory populate nested Customer fields with CustomerFactory. Use Use(CustomerFactory.build) when one parent factory needs an explicit local override.

<workflow>

Workflow

Step 1: Pick the factory base

Match the base to your model backend (table above). A mismatched base fails during factory class creation. If your project uses multiple backends (e.g., Pydantic for HTTP DTOs + msgspec for internal events), import each base separately and don't try to share a factory across backends.

Step 2: Define one factory per model

Subclass the appropriate base with one concrete generic argument and let Polyfactory infer __model__. Set __model__ explicitly only when the generic declaration cannot identify exactly one model. Keep factories in tests/factories.py or tests/<feature>/factories.py, outside production code paths.

Step 3: Customize only what the test cares about

If a field can take any valid value, leave it for the factory to randomize. Override (literal value or Use(...)) only fields the assertion depends on. Tests that pin every field defeat the purpose of using a factory.

Step 4: Register as a pytest fixture (if used widely)

For factories used in many tests, decorate with @register_fixture in a pytest-discovered module and consume the snake-cased fixture name. For one-off use, call Factory.build() directly inline.

Step 5: Wire cross-model relationships

Set __set_as_default_factory_for_type__ = True on a base factory and let nested fields be populated automatically, or use Use(OtherFactory.build) for a local relationship override.

Step 6: Pin determinism only when needed

Tests that assert on specific generated values need __random_seed__ (or Factory.seed_random(seed) per test). Tests that assert on shape or invariants (e.g., total >= 0) should not — leaving randomization on widens coverage across runs.

</workflow>

<guardrails>

Guardrails