Avoiding double-encoded JSON?

[liquidcarbon]

Hello!

Starting from a "hello-world" example on FastAPI+Mangum from here,
I got a simple endpoint returning a dict up and running:

@app.post("/")
async def process_data(request: Request) -> dict:
    body = await request.json()
    data = body["data"]
    # do stuff
    res = dict(...)
    return res

It's my understanding that there are two ways to use the Lambda from python:

1. Invoking through boto3 client, I have to parse the response using

json.loads(json.loads(response.read())["body"])

Is this expected? Is it possible to avoid dumping to a string twice?

2. Invoking through requests to Lambda URL gives the once-encoded bytestring, and on response.json() I recover the plain dict I was asking for.

Are there any pros of using 1), or cons of using 2) ? Are there better options - in the context of shuffling around significant MB-sized payloads?