Deserialization of nested SQLModel(s) not working when 'table=True'

[emilhe]

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 typing import Literal, Union

from sqlalchemy import JSON, Column
from sqlmodel import Field, SQLModel


class Cat(SQLModel):
    type: Literal["cat"] = "cat"
    calico: bool


class Dog(SQLModel):
    type: Literal["dog"] = "dog"
    service_animal: bool


Animal = Union[Cat, Dog]


class Pet(SQLModel, table=True):
    name: str = Field(primary_key=True)
    animal: Animal = Field(discriminator="type", sa_column=Column(JSON))


jiji = Pet(name="Jiji", animal=Cat(calico=True))
print(jiji.model_dump_json())  #  {"name":"Jiji","animal":{"type":"cat","calico":true}}
jiji_copy = Pet.model_validate_json(jiji.model_dump_json())
print(jiji)  #  name='Jiji' animal=Cat(type='cat', calico=True)
print(jiji_copy)  # name='Jiji' animal={'type': 'cat', 'calico': True}

Description

I am trying to bind nested models to a JSON field. The serialization/deserialization process works as expected when table=False, but when i set table=True, the deserialization is no longer working as intended. In both cases, the serialized form is,

{"name":"Jiji","animal":{"type":"cat","calico":true}}

When table=False the deserialization yields,

name='Jiji' animal=Cat(type='cat', calico=True)

as exptected. However, when i set table=True, the nested model (here, the animal field) ends up as a dict instead of an object of the intended type (here, Cat),

name='Jiji' animal={'type': 'cat', 'calico': True}

I found some related discussions, #63 and #52, that indicates that this behavior may be related to issues with pydantic model validation.

Now, my question is - is there any (proper) way to mitigate this issue?

Operating System

Linux

Operating System Details

No response

SQLModel Version

0.0.18

Python Version

Python 3.12.0

Additional Context

No response

[dadodimauro]

Hi @emilhe, I'am encountering the same problem as you described. Where you able to solve this?

[emilhe]

I didn't find any proper solution.

[20-buck-spin]

I am facing the same issue. For now I just model_validate the field when I intend to use it. But this is a very error prone hard to maintain approach. Any help would be highly appreciated.

My hunch is that I need to pass a special function to the json_deserializer param of create_engine. I am looking into it. I hope this helps someone find the necessary solution

[martinlatrille]

I have the same issue as well, thank you @20-buck-spin for the temporary workaround