Docs: Add example to use PydanticFactory with SQLAlchemy persistence

[WilliamDEdwards]

Summary

Hi,

I use FastAPI with a 'typical' structure: Pydantic schemas are converted to SQLAlchemy models.

For tests, currently, SQLAlchemy models are created in fixtures. Obviously, as the amount of models and validations grows, this doesn't scale. Therefore, I will use per-test objects created by factories.

polyfactory looks ideal, but one thing is missing: I would like to pass a Pydantic schema (PydanticFactory), then convert it to an SQLAlchemy model and create it, using a custom (CRUD layer) implementation.

Reason: Pydantic schemas validate business logic that should not be violated in tests (which happens by creating SQLAlchemy models with only data validation).

This looks relatively easy to implement by passing a custom handler to __sync_persistence__ on a factory (https://polyfactory.litestar.dev/latest/usage/configuration.html#persistence-handlers). (Still need to figure out how to subclass my factories from a base class so I don't have to repeat that...)

Instead of writing the implementation myself, would the maintainers be interested in adding a first-party supported example to the documentation? I could imagine this use case is quite common, as this is relevant for basically all larger FastAPI/Pydantic/SQLAlchemy projects.

[guacs]

If my understanding is correct, you want a way to convert pydantic models into sqla models? Could you show an example of what it is that you're trying to achieve?

[WilliamDEdwards]

If my understanding is correct, you want a way to convert pydantic models into sqla models? Could you show an example of what it is that you're trying to achieve?

I would like to use PydanticFactory, then save the corresponding SQLAlchemy model to the database. According to the documentation, this can be achieved by adding a persistence handler to PydanticFactory. I would then convert and save in the save method.

[guacs]

Aah gotcha. And how is that translation from pydantic model to SQL model happening?

[WilliamDEdwards]

Aah gotcha. And how is that translation from pydantic model to SQL model happening?

For SQLAlchemy to Pydantic: https://docs.pydantic.dev/latest/concepts/models/#arbitrary-class-instances

The other way around, which is what I'm looking for:

from pydantic import BaseModel, ConfigDict

from sqlalchemy import create_engine
from sqlalchemy.orm import Mapped, mapped_column
from sqlalchemy import Integer
from sqlalchemy.orm import declarative_base, sessionmaker

engine = create_engine("mysql+pymysql://username:password@host/database")
database_session = sessionmaker(bind=engine)()

Base = declarative_base()

class TableOrm(Base):
    __tablename__ = "table"

    id: Mapped[int] = mapped_column(Integer, primary_key=True)

class TablePydantic(BaseModel):
    model_config = ConfigDict(from_attributes=True)

    id: int

pydantic_schema = TablePydantic(id=1)
pydantic_schema_as_dict = pydantic_schema.model_dump(mode="json")  # mode ensures only JSON-serializable types
sqlalchemy_object = TableOrm(**pydantic_schema_as_dict)

database_session.add(sqlalchemy_object)
database_session.commit()
database_session.refresh(sqlalchemy_object)

[guacs]

Hm...I'm not a 100% sure on whether we should add this to the documentation. That being said, I do see the benefit of this and agree that it may be a relatively common use case (especially for those using FastAPI like you said). So I'm kind of 50/50 on this. Anyone have any opinions on this, @litestar-org/members?