Bug

Output of python -c "import pydantic.utils; print(pydantic.utils.version_info())":

             pydantic version: 1.5
            pydantic compiled: True
                 install path: /home/neuro/.local/share/virtualenvs/tmp--yQIjN1Q/lib/python3.7/site-packages/pydantic
               python version: 3.7.5 (default, Dec  3 2019, 16:12:26)  [GCC 9.2.0]
                     platform: Linux-5.6.2-arch1-2-x86_64-with-arch
     optional deps. installed: []

With pydantic 1.5 I've run into some issues with shared fields in different models. Unfortunately I can only recreate the issue when using the models in fastAPI endpoints. It's definitely a change in pydantic that introduced the issue (if I downgrade back to 1.4, the issue disappears), but it might be due to some quirky way that fastAPI is using pydantic models. In that case I'm happy to reraise the issue in the fastAPI project.

The problem I'm seeing is that if I reuse a required Field(...) in two separate models, only the first model correctly tells me that a required input is missing. The second model gets (incorrectly) instantiated with the required value set to None.

from fastapi import FastAPI
from pydantic import BaseModel, Field

app = FastAPI()
required_field = Field(...)


class Model1(BaseModel):
    required: str = required_field


class Model2(BaseModel):
    required: str = required_field


@app.post("/expected")
async def expected(i: Model1):
    return i


@app.post("/unexpected")
async def unexpected(i: Model2):
    return i

Now if you fire some requests, you can see the weird difference in behavior between the /expected and /unexpected endpoint.

$ curl localhost:8000/expected -H "Content-Type: application/json" -d '{}'
{"detail":[{"loc":["body","i","required"],"msg":"field required","type":"value_error.missing"}]}

$ curl localhost:8000/unexpected -H "Content-Type: application/json" -d '{}'
{"required":null}

Maybe you can think of some change to Field between 1.4 and 1.5 that could be behind this weird behavior? If not I'm happy to try and isolate the problem some more and look into exactly how fastAPI is using the pydantic models and what might go wrong there.

Thanks in advance for any help.