Swagger KeyError when two classes contain inner classes with the same name

[tanyaofei]

Describe the bug

KeyError when two classes contain inner classes with the same name. If i rename one of the inner classes, app runs good.

To Reproduce

Steps to reproduce the behavior with a minimum self-contained file.

Replace each part with your own scenario:

1. Create a file with:

from __future__ import annotations

from typing import Union

import uvicorn
from fastapi import FastAPI
from pydantic import BaseModel


class X(BaseModel):
    class A(BaseModel):
        pass

    a: X.A


X.update_forward_refs()


class Y(BaseModel):
    class A(BaseModel):
        pass

    y: Y.A


Y.update_forward_refs()

app = FastAPI()


@app.get("/")
def route(request: Union[X, Y]):
    return str(request)

3. Open the browser and call the endpoint /docs.
4. It show Fetch error Internal Server Error /openapi.json and the app raise an error

  File "/usr/local/lib/python3.7/site-packages/fastapi/utils.py", line 64, in get_model_definitions
    model_name = model_name_map[model]
KeyError: <class '__main__.X.A'>

Environment

• OS: MacOS
• FastAPI Version : 0.55.1
• Python: 3.8.2

[tiangolo]

Why would you create a Pydantic class inside of other Pydantic class?

I think that's not really supported by Pydantic...

But yeah, it wouldn't be supported by FastAPI, as there wouldn't be a clean way to get unique IDs for the models.

[github-actions]

Assuming the original issue was solved, it will be automatically closed now. But feel free to add more comments or create new issues.

[vbyrd]

I suppose this would matter when using modules that return customized pydantic models. I am playing with the pydantic_jsonapi module with FastAPI and encountering the same error. It uses GenericModel from Pydantic to accomplish a jsonapi model. Interestingly, this doesn't seem to happen with the same models used as "response_model" in the decorator and it works fine outside of Fast API.

File "/usr/local/lib/python3.8/site-packages/fastapi/utils.py", line 64, in get_model_definitions
     model_name = model_name_map[model]
KeyError: <class 'pydantic_jsonapi.request.RequestDataModel[typing.Literal['vendor'], VendorCreateModel]'>

https://pydantic-docs.helpmanual.io/usage/models/#generic-models
https://github.com/DeanWay/pydantic-jsonapi

[sureshjoshi]

I just ran into this today. In terms of "Why would you create a Pydantic class inside of other Pydantic class" - in my case, the problem happened with two Schemas that had the same inner enum class (not Pydantic inner classes). The reason for it is that the purpose of the type is semantically the same, while happening to have different values.

I can't seem to get string literals to show up in OpenAPI documentation, otherwise I wouldn't use the enums.

Of note: Not a big deal, as there are many workarounds - I just wish the call stack was a bit more descriptive in terms of which object threw the exception, rather than just the enum key :)

[guissart]

@sureshjoshi @tanyaofei reading this issue saved my day.

I had a duplicated Enum class with the same name across two Pydantic model.
And the Key error that it produced doesn't give many clue on what happened.

Naming differently the two Enum class solve the problem.

[mike0sv]

I generate endpoints dynamically based on some config, and generate models for those endpoints with pydantic.create_model. Code in get_model_name_map silently drops models with identical names, which leads to uninformative KeyError later.