FastAPI

The didactic-fastapi distribution adapts dx.Model types for use as FastAPI request and response bodies. The current implementation goes through to_pydantic and caches the result; no new class-per-request work happens at runtime.

Installation

pip install didactic-fastapi

The package contributes didactic.fastapi to the namespace and depends on fastapi>=0.115.

Using a Model in a route

import didactic.api as dx
from fastapi import FastAPI
from didactic.fastapi import as_response


class User(dx.Model):
    id: str
    email: str


app = FastAPI()


@app.get("/users/{uid}", response_model=as_response(User))
def get_user(uid: str) -> User:
    return User(id=uid, email="ada@example.org")

as_response(User) returns a pydantic.BaseModel subclass that mirrors User. FastAPI uses it for response validation and OpenAPI generation. Each dx.Model is converted exactly once; subsequent calls return the cached Pydantic class.

Request bodies

as_request is a synonym of as_response. Use whichever name reads naturally in your route signatures:

from didactic.fastapi import as_request


@app.post("/users")
def create_user(payload: as_request(User)) -> User:
    return User(**payload.model_dump())

Validation errors as 422 responses

Install the validation handler once:

from didactic.fastapi import register_validation_handler

register_validation_handler(app)

After this, any dx.ValidationError raised inside a route surfaces as a 422 response. The body shape mirrors FastAPI's own error format:

{
  "detail": [
    {"loc": ["email"], "msg": "...", "type": "validates_failed"}
  ]
}

Client code that already handles FastAPI's 422s sees the same shape.

Caveats

• The adapter goes through to_pydantic. Models with computed or derived fields lose those fields on the Pydantic side; if you need them in the response, declare them on the Pydantic class instead (or move to plain Pydantic for that endpoint).
• The cache key is the dx.Model class identity. If you build Models dynamically per request, the cache will grow without bound; do the conversion once at startup and reuse.