mypy annotation and constrained lists recommendation

Question

Please complete:

• OS: Debian 10 (python:3.7-slim Docker image)
• Python version: 3.7.5
• Pydantic version 0.32.2

Hello! I’m hoping to find the best practice for the interaction between mypy checks and the ConstrainedList/conlist() pydantic types. For the MWE below, we’re expecting a BaseModel taking a fixed-length list of floats (this is used as a request body in FastAPI):

from pydantic import BaseModel, conlist

class EmbeddingRequest(BaseModel):
    embedding: conlist(float, min_items=100, max_items=100)

which behaves as expected, but complains when run through mypy checks (same error as in the older #239 issue). From #862 it sounds like it’s still on deck to add improvements to documentation for mypy+constrained types, so hopefully this can help. Alternately, I suppose this might be something that is resolved in v1.0/1.1 - for the time being I’m pinned to 0.32.2 for FastAPI compatibility, but I doubt that will last long.

Older issues propose a few options:

option 1

per #239,

from typing import List, Type
from pydantic import BaseModel, conlist

Vector: Type[List[float]] = conlist(float, min_items=100, max_items=100)

class EmbeddingRequest(BaseModel):
    embedding: Vector

runs, but complains that Vector is not valid as a type (and defining it as type ConstrainedList rather than List[float] per #156 fails for type mismatch).

option 2

also in #239 , explicitly subclassing the type

from pydantic import BaseModel, ConstrainedList

class Vector(ConstrainedList):
    item_type = float
    min_items = 100
    max_items = 100
    __args__ = [float]

class EmbeddingRequest(BaseModel):
    embedding: Vector

functions properly and passes mypy, but forcing the explicit __args__ overloading for it to behave properly relative to conlist() feels evil to me.

option 3

just punt and add a type: ignore flag on the original conlist call.

Any advice on this would be much appreciated!

[rustam-python]

If you (as I’m) don’t get what does it means: “With the new Annotated syntax, it should now be possible to make it work properly”

In Python 3.9+:

class _Artist(pydantic.BaseModel):
    image: typing.Annotated[list, pydantic.conlist(_Image, min_items=5, max_items=5)]

[samuelcolvin]

oh, that probably doesn’t work but would be the best solution,

The other thing to do is

from typing import TYPE_CHECKING

if TYPE_CHECKING:
    Vector = List[float]
else:
    Vector = conlist(float, min_items=100, max_items=100)

or something.