`Field(alias="")` doesn't work with Pydantic V2
Privileged issue
- [x] I'm @tiangolo or he asked me directly to create an issue here.
Issue Content
Currently alias parameter doesn't work as expected with Pydantic V2 installed.
Let's use this issue to track this.
👇 Here is a list of tests that showcase the problem (in the details):
from typing import Type, Union
import pytest
from pydantic import VERSION, BaseModel, ValidationError
from pydantic import Field as PField
from sqlmodel import Field, SQLModel
# -----------------------------------------------------------------------------------
# Models
class PydanticUser(BaseModel):
full_name: str = PField(alias="fullName")
class SQLModelUser(SQLModel):
full_name: str = Field(alias="fullName")
# Models with config (validate_by_name=True)
if VERSION.startswith("2."):
class PydanticUserWithConfig(PydanticUser):
model_config = {"validate_by_name": True}
class SQLModelUserWithConfig(SQLModelUser):
model_config = {"validate_by_name": True}
else:
class PydanticUserWithConfig(PydanticUser):
class Config:
allow_population_by_field_name = True
class SQLModelUserWithConfig(SQLModelUser):
class Config:
allow_population_by_field_name = True
# -----------------------------------------------------------------------------------
# Tests
# Test validate by name
@pytest.mark.parametrize("model", [PydanticUser, SQLModelUser])
def test_create_with_field_name(model: Union[Type[PydanticUser], Type[SQLModelUser]]):
with pytest.raises(ValidationError):
model(full_name="Alice")
@pytest.mark.parametrize("model", [PydanticUserWithConfig, SQLModelUserWithConfig])
def test_create_with_field_name_with_config(
model: Union[Type[PydanticUserWithConfig], Type[SQLModelUserWithConfig]],
):
user = model(full_name="Alice")
assert user.full_name == "Alice"
# Test validate by alias
@pytest.mark.parametrize(
"model",
[PydanticUser, SQLModelUser, PydanticUserWithConfig, SQLModelUserWithConfig],
)
def test_create_with_alias(
model: Union[
Type[PydanticUser],
Type[SQLModelUser],
Type[PydanticUserWithConfig],
Type[SQLModelUserWithConfig],
],
):
user = model(fullName="Bob") # using alias
assert user.full_name == "Bob"
# Test validate by name and alias
@pytest.mark.parametrize("model", [PydanticUserWithConfig, SQLModelUserWithConfig])
def test_create_with_both_prefers_alias(
model: Union[Type[PydanticUserWithConfig], Type[SQLModelUserWithConfig]],
):
user = model(full_name="IGNORED", fullName="Charlie")
assert user.full_name == "Charlie" # alias should take precedence
# Test serialize
@pytest.mark.parametrize("model", [PydanticUser, SQLModelUser])
def test_dict_default_uses_field_names(
model: Union[Type[PydanticUser], Type[SQLModelUser]],
):
user = model(fullName="Dana")
data = user.dict()
assert "full_name" in data
assert "fullName" not in data
assert data["full_name"] == "Dana"
# Test serialize by alias
@pytest.mark.parametrize("model", [PydanticUser, SQLModelUser])
def test_dict_default_uses_aliases(
model: Union[Type[PydanticUser], Type[SQLModelUser]],
):
user = model(fullName="Dana")
data = user.dict(by_alias=True)
assert "fullName" in data
assert "full_name" not in data
assert data["fullName"] == "Dana"
# Test json by alias
@pytest.mark.parametrize("model", [PydanticUser, SQLModelUser])
def test_json_by_alias(
model: Union[Type[PydanticUser], Type[SQLModelUser]],
):
user = model(fullName="Frank")
json_data = user.json(by_alias=True)
assert ('"fullName":"Frank"' in json_data) or ('"fullName": "Frank"' in json_data)
assert "full_name" not in json_data
When run with Pydantic V2:
t.py::test_create_with_field_name[PydanticUser] PASSED
t.py::test_create_with_field_name[SQLModelUser] FAILED
t.py::test_create_with_field_name_with_config[PydanticUserWithConfig] PASSED
t.py::test_create_with_field_name_with_config[SQLModelUserWithConfig] PASSED
t.py::test_create_with_alias[PydanticUser] PASSED
t.py::test_create_with_alias[SQLModelUser] FAILED
t.py::test_create_with_alias[PydanticUserWithConfig] PASSED
t.py::test_create_with_alias[SQLModelUserWithConfig] FAILED
t.py::test_create_with_both_prefers_alias[PydanticUserWithConfig] PASSED
t.py::test_create_with_both_prefers_alias[SQLModelUserWithConfig] FAILED
t.py::test_dict_default_uses_field_names[PydanticUser] PASSED
t.py::test_dict_default_uses_field_names[SQLModelUser] FAILED
t.py::test_dict_default_uses_aliases[PydanticUser] PASSED
t.py::test_dict_default_uses_aliases[SQLModelUser] FAILED
t.py::test_json_by_alias[PydanticUser] PASSED
t.py::test_json_by_alias[SQLModelUser] FAILED
All tests parameterized with both, Pydantic model and SQLModel model, to show the difference in behavior.
With Pydantic V1 all tests pass. With Pydantic V2 most of tests fail with SQLModel model.
I would also expect that alias is used as a default value of column name. But it should be possible to override it by providing sa_column=Column("column_name") or sa_column_kwargs={"name": "column_name"}.
Any feedback is welcome. Please, review the set of tests, suggest use-cases that are currently not covered.
Would be nice to add set of tests for validation_alias and serialization_alias as well.
Facing to the same issue to be able to use aliases in my project.
In the Fields(), the alias=xx is not enough, the schema_extra={"validation_alias": "xx", "serialization_alias": "xx"} must be added.
Any updates on this? Facing the same problem.
Hey @svlandeg is the issue fixed now?
@YuriiMotov @svlandeg
Hi! I've attempted to create a fix for this issue in PR #1577: https://github.com/fastapi/sqlmodel/pull/1577
I hope this approach helps address the problem where Field(alias="...") wasn't working as expected in Pydantic v2.
Proposed Solution
The fix tries to address the core issue by:
- Automatic alias propagation: When using Pydantic v2, SQLModel automatically sets
validation_aliasandserialization_aliasfrom thealiasparameter if they're not explicitly provided - Backward compatibility: Maintains compatibility with Pydantic v1 by filtering out v2-specific parameters
- Additional v2 support: Enables the new
validation_aliasandserialization_aliasparameters for users who want more granular control - Test coverage: Added tests covering the scenarios mentioned in this issue
Thanks for the PR @ravishan16! We'll have a look when we can and leave feedback on the PR directly 🙏