add model_dump options to sqlmodel_update

[A00474880]

As per discussion #1838, this PR allows users to use model_dump parameters directly in sqlmodel_update function.

New Usage Pattern

Previous recommend use of sqlmodel_update is as follows:

def update_hero(
    *, session: Session = Depends(get_session), hero_id: int, hero: HeroUpdate
):
    db_hero = session.get(Hero, hero_id)
    ...
    hero_data = hero.model_dump(exclude_unset=True)
    db_hero.sqlmodel_update(hero_data)
    ...

This PR allows the following while keeping backward compatibility with the above:

def update_hero(
    *, session: Session = Depends(get_session), hero_id: int, hero: HeroUpdate
):
    db_hero = session.get(Hero, hero_id)
    ...
    db_hero.sqlmodel_update(hero, exclude_unset=True)
    ...

Tests

This PR also adds assertions in the test_update function that increases test coverage by 2 lines:
before:

Name                                                                                                        Stmts   Miss  Cover   Missing
-----------------------------------------------------------------------------------------------------------------------------------------
sqlmodel/main.py                                                                                              375     16    96%   593, 645, 666, 673, 675, 690, 712, 715, 721, 723, 725, 727, 729, 908, 996, 1004

after:

Name                                                                                                        Stmts   Miss  Cover   Missing
-----------------------------------------------------------------------------------------------------------------------------------------
sqlmodel/main.py                                                                                              368     14    96%   593, 645, 666, 673, 675, 690, 712, 715, 721, 723, 725, 727, 729, 908

[A00474880]

I can't seem to be able to add labels to this PR. So it'd be great if I can get this reviewed

[YuriiMotov]

@A00474880, thanks for your interest and efforts!

I pointed to a couple of issues with this implementation

[YuriiMotov]

kwargs will not give autocompletion, and it's error-prone

[YuriiMotov]

This will break use cases with models that have fields with exclude=True:

from sqlmodel import Field, SQLModel


class Item(SQLModel):
    id: str
    param: str = Field(exclude=True)


a = Item.model_validate({"id": "1", "param": "1"})
b = Item.model_validate({"id": "1", "param": "2"})


a.sqlmodel_update(b, exclude={"id"})
# a.sqlmodel_update(b)


assert a.param == "2"

.. and probably some other cases when model has settings that change the default serialization schema.

[A00474880]

Hi @YuriiMotov , thank you for the review. My recent commits tries to address your feedback. Please let me know if you need anything else.

Change summary:

1. Added type support for the model_dump_kwargs with a TypedDict definition
2. To address the issue with Model fields with serialization settings: Added steps to create temporary UpdateModel schema which removes the extra model settings before using model_dump on an instance of the new UpdateModel obj

[A00474880]

Hi all I've had some time to carefully consider the changes and how it affects use cases. With the current implementation of creating a temporary "clean" UpdateModel, some model dump kwargs are effectively ignored when using them directly in the new sqlmodel_update, while the remaining provide slight convenience. Here's a summary:

Effective (8 — work correctly on clean UpdateModel)

These don't depend on field-level metadata:

Kwarg	Mechanism
exclude_unset	Checks model_fields_set (preserved via model_construct)
exclude_none	Compares value against None
exclude	Filters by field name
include	Filters by field name
mode	Controls serialization format
round_trip	Round-trip compatible output
warnings	Controls warning behavior
context / fallback / serialize_as_any	Passed through to serializer

Ignored (3 — clean UpdateModel lacks the corresponding metadata)

Kwarg	Why	Impact
by_alias	Clean model has no serialization_alias on any field	Arguably an improvement. This prevents aliased fields from being ignored in sqlmodel_update(Hero.model_dump(by_alias=True)), with current sqlmodel_update(Hero, by_alias=True), aliased fields are also updated
exclude_defaults	Clean model has all fields as required (no defaults)	Unexpected but harmless. User expects to filter default-valued fields; they all pass through. Original branch had no exclude_defaults at all, so this is a new-but-broken feature, not a regression.
exclude_computed_fields	Clean model has no @computed_field	OK. model_fields is the basis; computed fields aren't in it to begin with.

Comparison to main branch (getattr approach)

The original sqlmodel_update used getattr(obj, field_name) — no serialization at all:

Aspect	Original (getattr)	New (clean model + model_dump)
Field(exclude=True)	Ignored ✓	Ignored ✓
Field(serialization_alias)	Ignored ✓	Ignored ✓
exclude_unset	Not supported	Added ✓
exclude_none	Not supported	Added ✓
exclude / include	Not supported	Added ✓
by_alias	Not supported	no-op (same as original)
exclude_defaults	Not supported	no-op (same as original)
Custom serializer on type	Not applied (raw value)	Applied (serialized value)

Note

We can actually add support for exclude_defaults by adding defaults to the fields_def definition:

            fields_def = {
                fname: (finfo.annotation, finfo.default)
                for fname, finfo in ObjClass.model_fields.items()
            }

Please let me know if you have any suggestions!