[QUESTION] SQL related tables and corresponding nested pydantic models in async

Really impressed with FastAPI so far… I have search docs github, tickets and googled the issue described below.

Description

How best to work with related tables and corresponding nested pydantic models whilst persisting data in a relational database in an async application?

Additional context

I have been attempting to extend the example in the docs https://fastapi.tiangolo.com/advanced/async-sql-databases/ which relies on https://github.com/encode/databases

Using three test pydantic models as an example:

class UserModel(BaseModel):
    id: int
    title: str = Field(..., min_length=2, max_length=50)
    firstname: str = Field(..., min_length=1, max_length=50)
    firstname: str = Field(..., min_length=1, max_length=50)
    username: str = Field(..., min_length=3, max_length=50)
    email: str = Field(..., min_length=3, max_length=50)
    favourite_book: int = Field(...)

class FavouriteBook(BaseModel):
    id: int
    title: str = Field(...)
    author: str = Field(...)


class ExtendedUser(BaseModel):
    id: int
    title: str = Field(..., min_length=2, max_length=50)
    firstname: str = Field(..., min_length=1, max_length=50)
    firstname: str = Field(..., min_length=1, max_length=50)
    username: str = Field(..., min_length=3, max_length=50)
    email: str = Field(..., min_length=3, max_length=50)
    favourite_book: FavouriteBook

the route would ideally be along the lines of…

@router.get("/extended", response_model=List[ExtendedUser])
async def list():
    query =  **sqlAlchemy/databases call that works**
    return database.fetch_all(query=query)

How can a user create a route that returns the nested ExtendedUser from the database without resorting to performing two queries? An SQL join is a standard way to do this with a single query. However, this does not work with SQLAlchemy core as the two tables contain ‘id’ and ‘title’ columns.
It is possible to work with SQLAlchemy orm - but not in an async way as far as I know. (async is my reason for using FastAPI ). I could rename the columns to something unique ( but to rename ‘id’ column seems like poor database design to me).

[tiangolo]

Yep, I know. We still don’t have a solution for that. I have been thinking about creating a small layer on top of SQLAlchemy mixing SQLAlchemy and Pydantic… but we’ll see 😅

[leonh]

Thanks for the advice and continued work on FastAPI. I like the performance and syntax of PonyORM but it seems to fallen into a period of sporadic maintenance which is a cause for concern. My understanding was that as it was built in a way that would allow it to possibly be async friendly in the future even if it wasn’t there right now. I hope the use of contextvars for asyc’ing PonyOrm works. I think the developers did a really good job on it when it was being built a few years ago, but maybe their priorities have moved in a different direction since then.

Planning to give Tortoise ORM a test run at some point, seeing as its been created to work well with FastAPI.

Side thought: The developer experience of so many similar but different typing systems in my code. Dataclasses, ORM models, Pydantic validation, OpenApi schema, JS Form libraries, or typescript, state management models for React. It gets quite repetitive from end to end if no type system is designated as the canonical form within a web stack.