support optional annotated field with default value

[proever]

Currently, it is not possible to specify a SQLModel with an Optional complex type, such as a Decimal, as follows:

from decimal import Decimal
from typing import Annotated
from sqlmodel import Field, SQLModel


class PriceModel(SQLModel, table=True):
    id: int | None = Field(primary_key=True, default=None)
    price: Annotated[Decimal, Field(decimal_places=2, max_digits=9)] | None = None

Doing so results in the following error (see #67, #312) in get_sqlalchemy_type:

TypeError: issubclass() arg 1 must be a class

This PR attempts to fix this issue by adding a check in get_sqlalchemy_type for whether the type of the Field is a typing_extensions.AnnotatedAlias. If it is, instead of using the class of the AnnotatedAlias for the following comparisons (which results in the above error), it uses AnnotatedAlias.__origin__. Similarly, it infers the metadata from AnnotatedAlias.__metadata__ instead of the AnnotatedAlias itself.

In my testing, this approach seems to work well. I also added a simple test here that checks whether restrictions on an optional annotated field are enforced on the database side. It could probably be improved or extended.

[proever]

looks like some (many) tests are failing, I'll try to fix them!

Is this feature something that should work with Pydantic v1 too? I'm not too familiar with it.

[bootc]

I've just hit exactly this problem; is there any chance you could update your MR please @proever?

[msftcangoblowm]

Would argue this PR fixes a bug and is not a FR

pydantic validators is a core reason for the existence of both pydantic and SQLModel.

SQLModel currently only supports validators using decorator-pattern

Which is neither DRY nor reusable

SQLModel does not support reusable Validators using annotated-pattern

This PR would fix this.

Example pydantic validator

def validator_is_dead_or_cursed(value: int) -> int:
    if value > 120:
        msg_warn = (
            f"{value!s} indicates this person very likely dead "
            "or bibically cursed"
        )
        raise ValueError(msg_warn)

    return value

NotDead = Annotated[int, pydantic.AfterValidator(validator_is_dead_or_cursed)]

Apply to Field

age: Optional[NotDead] = Field(default=None, gt=0)

or

age: Annotated[Optional[int], pydantic.AfterValidator(validator_is_dead_or_cursed)] = Field(default=None, gt=0)

Equivalent to age: Optional[int] = Field(default=None, gt=0, le=120)

[YuriiMotov]

As I see it, this PR is basically about fixing the following scenario:

MyType: TypeAlias = Annotated[str, Field(max_length=5)]

class PriceModel(SQLModel, table=True):
    id: int | None = Field(primary_key=True, default=None)
    working: MyType  # VARCHAR(5)
    not_working: Optional[MyType] = None  # VARCHAR(255)

You are going to open a Pandora's box :)
I think this PR only handles one specific case, but in general we need to merge all nested Field annotations and handle conflicts. This seems to be a lot of work and quite error-prone.
Although, it would be nice to support this one day.

I suggest we skip it for now and get back to this feature later.

Seems to be related: #1281

---

As for initial code example, you can make it working by moving | None inside Annotated:

    price: Annotated[Decimal | None, Field(decimal_places=2, max_digits=9)] = None

---

@proever, thanks for your interest and efforts!

[YuriiMotov]

Rename to use_type?