Pydantic 2.0 ignores Optional in schema and requires the field to be available

Question

Pydantic 2.0 seems to have drastically changed.

Previously with FastAPI and Pydantic 1.X I could define the schema like this, where receipt is optional:

class VerifyReceiptIn(BaseModel):
    device_id: str
    device_type: DeviceType
    receipt: Optional[str]

Then in FastAPI endpoint I could do this:

@router_verify_receipt.post(
    "/",
    status_code=201,
    response_model=VerifyReceiptOut,
    responses={201: {"model": VerifyReceiptOut}, 400: {"model": HTTPError}},
)
async def verify_receipt(body: VerifyReceiptIn):
    auth_service = AuthService()
    ...

And the unit test without receipt in body was fine with it, but now with Pydantic 2.0 it's failing. Now it claims that Receipt is required and throws a 422 error.

response = await client.post(
            "/verify-receipt/",
            headers={"api-token": "abc123"},
            json={
                "device_id": "u1",
                "device_type": DeviceType.ANDROID.value,
            },
        )

But this is why we had Optional. Why do I have to pass receipt=None in body? This is not ideal as it will break everything on production. Is there a way around this? Thanks

Answer 1

Indeed Pydantic v2 changed the behavior of Optional to a more strict and correct one. The Pydantic 2.0 Migration Guide has a special section describing the new behavior.

According to the Guide Optional[str] is now treated as required but allowed to have None as its value. See the example:

from pydantic import BaseModel

class Model(BaseModel):
    value: Optional[str]

m = Model(value='abc')  # passes
m = Model(value=None)  # passes
m = Model()  # fails

Think of Optional[str] as of str | None. And in general A | B requires the value to be of type A or B.

For the value to be not required it has to have a default value like in the following example:

from pydantic import BaseModel

class Model(BaseModel):
    value: str | None = 'nothing'

m = Model(value='abc')  # passes
m = Model(value=None)  # passes
m = Model()  # passes and sets `value` to `'nothing'`

If you'd change value: str | None = 'nothing' to value: str | None = None you'll get a value that is not required and has a default value None.