feat(dto): more iteration.

[peterschutt]

• Adds dto.Attrib.pydantic_type and dto.Attrib.validators where pydantic_type allows the dto field type to be explicitly set, and validators allows arbitrary single argument callables to be specified as dto validators for any model attribute.
• factory() should now handle MappedAsDataclass sqla models.
• decorator() is for decorating pydantic models, and uses the decorated class as the base class for the model.
• adds some tests.

I spent a bit of time experimenting. I think best for the decorator to wrap subtypes of BaseModel and we use the decorated class as the base class for the produced dto. Seems to be the simplest approach IMO, but LMK what you think.

Cheers!

[gazorby]

If base is explictly passed the produced does not inherit MapperBind. We should make sure that MapperBind is part of bases in all cases.

[peterschutt]

Yah, this does need more thought.

The decorated class could inherit from MapperBind instead of BaseModel.

If the decorated class inherits from BaseModel then we can't simply mix in MapperBind as it inherits from BaseModel also.

If the decorated class inherits from MapperBind there is the issue that model needs to be passed as a class kwarg, so the model then gets declared in the decorator signature, and in the class signature on the very next line.. and I was unsure about the ergonomic of that.

I'll take another look with fresh eyes tomorrow.. I experimented with making MapperBind a mixin class, rather than inherit from BaseModel - maybe that's worth revisiting.

[gazorby]

Yes, I think MapperBind should be a mixin rather than a full pydantic model. Both the decorated class and MapperBind should be in bases, so we only want one of them to inherit BaseModel, or none.

[peterschutt]

Got this where I'm pretty happy implementation-wise, does it feel close enough to what you were heading toward with the original PR? This test is a good demo:

def test_dto_decorator() -> None:

    @dto.decorator(Author, dto.Purpose.WRITE)
    class AuthorDTO(BaseModel):
        name: constr(to_upper=True)  # pyright:ignore

        @validator("name")
        def validate_name(cls, val: str) -> str:
            """We're shouting!"""
            return f"{val}!"

        @validator("dob", check_fields=False)
        def validate_dob(cls, val: date) -> date:
            """Off by one."""
            val += timedelta(days=1)
            return val

    assert AuthorDTO.parse_obj({"name": "Bill Bryson", "dob": "1951-12-08"}).dict() == {
        "name": "BILL BRYSON!",
        "dob": date(1951, 12, 9),
    }

Still need to do docs, more tests, I think there'll be an issue with self-referencing relationships to sort out, and will probably break up the module into a sub-package as a bit going on in there now..

Main issue is that mypy isn't recognising the to_mapped() method, either via the factory() or decorator() method, just get these:

tests/utils/controllers.py:33: error: "CreateDTO" has no attribute "to_mapped"  [attr-defined]
tests/utils/controllers.py:45: error: UpdateDTO? has no attribute "to_mapped"  [attr-defined]
tests/unit/test_dto.py:253: error: "CreateDTO" has no attribute "to_mapped"  [attr-defined]

Seems like there is a very longstanding issue with mypy and class decorators, python/mypy#3135 (comment) but I wouldn't have thought that would affect the factory() generated DTOs.

[gazorby]

Hi @peterschutt!

Seems good enough regarding decorator impementation plus validator support. I got a working iteration concerning the references cycles. It's quite a bit of work, maybe we want a new PR on your repo for this. I'll merge this so I can rebase on what your last changes here.

[gazorby]

@peterschutt Could you resolve the conflicts?

[peterschutt]

Hey @gazorby - using https://github.com/topsport-com-au/starlite-saqlalchemy/tree/dto-improvements as feature branch for this, and have opened topsport-com-au#139. All of your original work still in there. Thanks again, dto definitely in a better place now!

I'll start working on restructuring/tests/documentation and if you'd like to chip in for breaking the reference cycles, that'd be great!