Enhancement: Support custom provider on types created through 3.14 type statements

[bullfest]

Summary

Python 3.14 added type aliases through type statements: https://docs.python.org/3/library/typing.html#type-aliases
These aliases are preserved in the annotations:

type ConstrainedString = Annotated[str, pydantic.AfterValidator(validator)]
ConstrainedStringOld = Annotated[str, pydantic.AfterValidator(validator)]
print(repr(ConstrainedString | None))  # ConstrainedString | None
print(repr(ConstrainedStringOld | None))  # typing.Optional[typing.Annotated[str, AfterValidator(func=<function validator at ...)]]

Thus this is something that polyfactory could leverage to allow for easier ways to add custom providers for types that are just constrained versions of built-in types.

Basic Example

Here's an example of some usage of a type alias.
Note that coverage() (accidentally) works when the type alias is part of a Union.

from typing import Annotated

import pydantic
from polyfactory.factories.pydantic_factory import ModelFactory


def validator(value: str) -> str:
    if len(value) % 2 == 0:
        raise ValueError("Must be an odd length string")
    return value


type ConstrainedString = Annotated[str, pydantic.AfterValidator(validator)]

ModelFactory.add_provider(ConstrainedString, lambda: "f" + "oo" * ModelFactory.__random__.randint(0,100))


class MyModel(pydantic.BaseModel):
    s: ConstrainedString

class MyModelFactory(ModelFactory[MyModel]):
    __check_model__ = True

# Neither of these work
print(list(MyModelFactory.coverage()))
print(list(MyModelFactory.batch(10)))

class MyModelWithNone(pydantic.BaseModel):
    s: ConstrainedString | None

class MyModelWithNoneFactory(ModelFactory[MyModelWithNone]):
    __check_model__ = True

print(list(MyModelWithNoneFactory.coverage())) # Surprisingly works
print(list(MyModelWithNoneFactory.batch(10))) # But this doesn't work

Drawbacks and Impact

No response

Unresolved questions

No response

[bullfest]

I'm not sure what the wider implications of it is, but the fix could be as simple as just not normalizing TypeAliasType types when building FieldMetas.

That would make generation fail for types defined through type statements which don't have providers defined, but that could possibly be resolved by extracting the types further in, as values are being generated.

An alternative could be to store the original annotation somewhere and using that explicitly when checking for providers.