Support custom generic container types in v2

[dmontagu]

In v2, currently the following code:

from typing import TypeVar, List

from pydantic import BaseModel

T = TypeVar('T')


class GenericList(List[T]):
    pass


class Model(BaseModel):
    field: GenericList[int]

results in:

pydantic_core._pydantic_core.SchemaError: Invalid Schema:
typed-dict -> fields -> field -> schema
  Input tag 'genericlist' found using self-schema does not match any of the expected tags: 'any', 'none', 'bool', 'int', 'float', 'str', 'bytes', 'date', 'time', 'datetime', 'timedelta', 'literal', 'is-instance', 'is-subclass', 'callable', 'list', 'tuple-positional', 'tuple-variable', 'set', 'frozenset', 'generator', 'dict', 'function', 'function-wrap', 'function-plain', 'default', 'nullable', 'union', 'tagged-union', 'chain', 'lax-or-strict', 'typed-dict', 'model', 'arguments', 'call', 'recursive-ref', 'custom-error', 'json', 'url', 'multi-host-url' [type=union_tag_invalid, input_value={'type': 'genericlist', '...chema': {'type': 'int'}}, input_type=dict]

I believe support for custom generic containers was added to v1.X by PR #2554; that PR referenced this comment: #2465 (comment) as providing its motivation.

[samuelcolvin]

Should be able to cover this with the same logic we use for deque.

[dmontagu]

I think actually we maybe shouldn't support this (though we should make sure we provide a better error) out of the box, but should instead raise an error message that links to documentation for how to implement __get_pydantic_core_schema__ for the custom generic class.

The gist I link in this comment: #2274 (comment) shows how to create a non-contrived pydantic core schema for a custom generic container, I think we should put something like this in the docs and ideally link to it in the error message.

(In this PR: #5629 I implement similar "error for custom generic containers" logic specifically for subtypes of Sequence)

[dmontagu]

Added @adriangb since I think he's looking into changes to how we do subclasses/generic containers more generally. @adriangb I think you can close this issue if you implement something that overrules any discussion above / etc.

[adriangb]

It's a bit unclear to me what the original issue is asking for. I see two different things going on:

• Generic containers: we fully support these, at least for the common standard library types dict, set, Counter, etc.
• Subclasses of known types: these are supported, you just need to implement __get_pydantic_core_schema__ and you'll get an error telling you to do so if you don't

So this issue is resolved unless I'm missing something.