Better support for JSON schema for Callable validators

[sydney-runkle]

This isn't a high priority, but I think we could add better default support for Callable type validator JSON schemas, as well as changing the docs to reflect how to change the default schema in these cases.

See more info here: #7957

[Viicos]

Was hit by this, and here is a use case that could be interesting and would avoid some boilerplate:

from typing import Annotated, Any

from pydantic import BaseModel, BeforeValidator

def _from_string(value: Any):
    if isinstance(value, string):
        return value.split(".")
    return value


class Model(BaseModel):
    screen_position: Annotated[tuple[int, int], BeforeValidator(_from_string)]

Ideally for BeforeValidator, the JSON schema should be updated (in validation mode only) if the callable allow different types (in this case string).

Currently, WithJsonSchema can be used, but really verbose:

class Model(BaseModel):
    screen_position: Annotated[tuple[int, int], BeforeValidator(_from_string), WithJsonSchema({'anyOf': [{'maxItems': 2, 'minItems': 2, 'prefixItems': [{'type': 'integer'}, {'type': 'integer'}], 'type': 'array'}, {'type': 'string'}]})]

Two solutions I've been thinking about:

• Infer the JSON schema from the type hint of the provided callable to BeforeValidator:

def _from_string(value: tuple[int, int] | str):  # Infer JSON schema from this type
    ...

Could be confusing, because in theory before validator can take any type. Plus, how to handle untyped callables?

• Add an extra argument to BeforeValidator/field_validator (when mode='before'): BeforeValidator(_from_string, new_type=tuple[int, int] | str)

[sydney-runkle]

Thanks for the example @Viicos! Will keep this in mind when adding this feature 😄!

[sydney-runkle]

Also relevant for PlainSerializer annotations: #8548 (comes as a surprise to some users)