Pydantic Model / Ninja Schema Config Not Respected In Serializing?

[sghng]

I'm trying to build endpoints that uses snake_case on Python server side, and all responses sent to JavaScript client sides use camelCase. After some research, I found that this can be done with Pydantic aliases:

class BaseModel(PydanticBaseModel):
    model_config = ConfigDict(
        alias_generator=to_camel,
        populate_by_name=True,  # read from both snake_case and camelCase
        serialize_by_alias=True,  # serialize to camelCase
    )

So I built my Item on top of the customized BaseModel.

@router.get("/{item_id}", response=Item | list[Item])
async def get_item(request, item_id: int, similar: int | None = None):
    if similar is None:
        res = await fetch_item(item_id)
        print(res.model_dump_json())
        return res
    else:
        return await fetch_similar_items(item_id, similar)

In testing, the printed json is in deed using camelCase, but the response sent to client side is still in snake_case. This is very confusing It seems like the in serializing models/schemas, the idiomatic methods like model_dump_json are not used.

EDIT: I also tried extend from ninja.Schema, but the problem stays the same.

[sghng]

Further testing proved that the builtin JSONRenderer does render the object with desired camelCase, so the config is respected at least by the renderer. So that question is why the endpoint doesn't send this result to the client?

@router.get("/{item_id}", response=Item | list[Item])
async def get_item(request, item_id: int, similar: int | None = None):
    if similar is None:
        res = await fetch_item(item_id)
        from ninja import NinjaAPI

        api = NinjaAPI()
        # this prints the correctly formatted camelCase JSON!
        print(api.renderer.render(request, res, response_status=200))
        return res
    else:
        return await fetch_similar_items(item_id, similar)

[sghng]

Found this in operations.py:

        result = validated_object.model_dump(
            by_alias=self.by_alias,
            exclude_unset=self.exclude_unset,
            exclude_defaults=self.exclude_defaults,
            exclude_none=self.exclude_none,
            **model_dump_kwargs,
        )["response"]
        return self.api.create_response(
            request, result, temporal_response=temporal_response
        )

So we do have a by_alias option to control the alias behavior, and the return value from the endpoint is first dumped then serialized.

I think it might be more elegant if we simply use methods on Pydantic model itself to dump data into json. It might save some lines of code, plus users will be able to customize serializing behavior inside the schema and use it everywhere, without having to pass arguments to each decorator.

@vitalik Is this something worse implementing?