refactor(typing): use abstract container types where possible

[sisp]

I've replaced a few occurrences of concrete container types (Dict and List) by abstract container types (Mapping and Sequence) which is slightly better practice AFAIK. See, e.g.,

• https://blog.logrocket.com/understanding-type-annotation-python/#adding-type-hints-dictionaries
• https://blog.logrocket.com/understanding-type-annotation-python/#adding-type-hints-lists

for an interesting article on this topic.

[sisp]

Hm, I'll need to check why the tests are failing for Linux and macOS. Surprisingly, they're passing for Windows. 🤔

[yajo]

I agree with the change, once the CI goes ✔️ .

Keep in mind that probably those annotations have a runtime effect with Pydantic. It might be the cause of the test problems.

[sisp]

I think you're right that the problem originates from Pydantic. Pydantic v1.10.2 handles the Mapping type incorrectly the way I see it:

from typing import Any, Dict, Mapping

from pydantic import parse_obj_as
from pydantic.dataclasses import dataclass


@dataclass
class Model:
    m: Mapping[Any, Any]
    d: Dict[Any, Any]


v = [["key", "value"]]
x = parse_obj_as(Model, {"m": v, "d": v})
print(x.m)  # ['key']
print(x.d)  # {'key': 'value'}

This behavior of Mapping vs. Dict is inconsistent and unexpected IMO.

[sisp]

I've reverted one usage of Mapping to Dict where Pydantic's behavior is odd. The tests are passing now; the one test that's failing is because of a Codecov upload issue.

[sisp]

I've created an issue over at Pydantic to indicate the Mapping bug: pydantic/pydantic#4588

[sisp]

I've submitted a PR to Pydantic which fixes the Mapping bug: pydantic/pydantic#4589

I suggest to proceed here without waiting for that PR to get merged because it's uncertain if and when it'll get merged and when a patch release for Pydantic will become available since they're working on Pydantic v2.

[sisp]

It seems this bug won't be fixed in Pydantic v1 and parsing a list of pairs as a dict seems to be a coincidental feature (see pydantic/pydantic#4588 (comment)).

[yajo]

the one test that's failing is because of a Codecov upload issue

Yet again... I can't believe how much often that fails... 🤦🏼‍♂️