Version 0.0.26 introduced a type-checking error for sqlmodel.Field

[jakekasan]

First Check

• I added a very descriptive title here.
• I used the GitHub search to find a similar question and didn't find it.
• I searched the SQLModel documentation, with the integrated search.
• I already searched in Google "How to X in SQLModel" and didn't find any information.
• I already read and followed all the tutorial in the docs and didn't find an answer.
• I already checked if it is not related to SQLModel but to Pydantic.
• I already checked if it is not related to SQLModel but to SQLAlchemy.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from sqlmodel import Field

Description

From 0.0.26 onwards, importing/using sqlmodel.Field and having pyright/basedpyright in "strict" mode causes the type checker to report an error. This means that I either have to set a type: ignore or dial back the overall type "strict-ness" in a project to stop the LSP from complaining.

Full example:

(the code above saved as "main.py")

pyproject.toml - minimal just to set "strict" mode for type-checking

[project]
name = "foo"
version = "0.1.0"
requires-python = ">=3.13"

[tool.pyright]
typeCheckingMode = "strict"

run with version 0.0.25 - no errors or warnings reported (as expected)

pip install --quiet sqlmodel==0.0.25 && pyright

version 0.0.26 fails with reportUnknownVariableType

pip install --quiet sqlmodel==0.0.26 && pyright

Operating System

macOS

Operating System Details

No response

SQLModel Version

0.0.26

Python Version

3.13.7

Additional Context

No response

[YuriiMotov]

This is caused by from __future__ import annotations in main.py added by #1560

[markacianfrani]

I think this might also be related and/or to pyright 1.1.407 specifically

[evans23]

Downgrading did not resolve this for me.

[mpyuglgwkxcmodyrpo]

I tracked down why this is happening but I don't know the best solution.
sqlmodel main.py imports Undefined and UndefinedType from _.compat and then again from pydantic_core in the if TYPE_CHECKING case. For some reason the type checker (when using future annotations) considers both types in the type check, so any argument that uses UndefinedType becomes Unknown | UndefinedType, which leads to the error.

I don't know how it would affect older versions and tests, but adding an else clause to the if TYPE_CHECKING statement and importing UndefinedType and Undefined from ._compat in that block, while removing them from the ._compat import above fixes the issue pyright is having. It seems like pyright is evaluating the if-else block in ._compat to be a union of the possible types in the if IS_PYDANTIC_V2: ... else: ... clause, not the ones that are actually present based on the version installed.

The relevant part of the import block now looks like this:

from ._compat import (  # type: ignore[attr-defined]
    IS_PYDANTIC_V2,
    PYDANTIC_MINOR_VERSION,
    BaseConfig,
    ModelField,
    ModelMetaclass,
    Representation,
    SQLModelConfig,
    _calculate_keys,
    finish_init,
    get_annotations,
    get_config_value,
    get_field_metadata,
    get_model_fields,
    get_relationship_to,
    get_sa_type_from_field,
    init_pydantic_private_attrs,
    is_field_noneable,
    is_table_model_class,
    post_init_field_info,
    set_config_value,
    sqlmodel_init,
    sqlmodel_validate,
)
from .sql.sqltypes import AutoString

if TYPE_CHECKING:
    from pydantic._internal._model_construction import ModelMetaclass as ModelMetaclass
    from pydantic._internal._repr import Representation as Representation
    from pydantic_core import PydanticUndefined as Undefined
    from pydantic_core import PydanticUndefinedType as UndefinedType
else:
    from ._compat import Undefined, UndefinedType

Thoughts? I can submit a PR to see if it still passes tests, if you want; though I feel this is just a band-aid to a larger issue... As far as I understand pydantic_core didn't exist until pydantic v2, so the type checker will get mad at the if TYPE_CHECKING block in main.py if pydantic v1 is in use anyway. Is pydantic v1 support a priority moving forward with this project?