Hi, I had the same problem with maintaining two sets of models and took inspiration from encode/orm (not finished) and ormantic (not maintained) and started ormar that uses encode/databases and sqlalchemy core under the hood, and can be used in async mode. @tiangolo feel free to check it out and let me know if you find it useful!

I have some tables that I'd like to exclude some fields that don't have a python type in sqlalchemy. I'd like to keep them in my model and process them into pydantic manually after the other fields get processed with this utility, but the exclude logic runs after the determination of the type, so I get a NotImplementedError() regardless of exclusion in this case.

I'd like to just move the exclusion logic above the type determination, so that this wouldn't occur in my use-case.

Hi there,

thanks for all your efforts in getting this model translation to work! Really appreciated. Upon a quick search (there really seems to be an entire zoo of possible orm's + pydantic + graphene models etc. etc.) I came upon this: https://github.com/kolypto/py-sa2schema From the descriptions and listed examples it seems they are trying to achieve the same thing.

Now the hard question: which one should I use? Any opinions on pro- & cons?

Thanks, Carsten

:bug: Do not include defaults from SQLAlchemy models in Pydantic models.

SQLAlchemy defaults are supposed to be generated at insertion/update time, not generated by the Pydantic model before creating a new object/record.

It's also currently broken, as the default included is a SQLAlchemy object, not an actual default.

I think that the depth is a useful feature when serializing the model has many relationships like the DRF serializer's depth option

Other libraries that it has depended on Pydantic

• https://github.com/vitalik/django-ninja/blob/master/ninja/orm/factory.py#L29

I went ahead and added the new_model_name parameter to close #51 . Will add tests as well.

I also added an only field, that allows the reverse from exclude: only the mentioned fields will be included.

I am also thinking to add a class that does something similar. Basically, you would inherit from it set what model you want to use (similar to Django's Meta class for e.g. views) and which fields and in the constructor it calls the sqlalchemy_to_pydantic function (using the new class name as model name) and creates the schema.

For the create method of SQL model I need to exclude the id of a model, I'd like to call the sqlalchemy_to_pydantic function twice on the same model for example like this:

PyUser = sqlalchemy_to_pydantic(User)
PyUserCreate = sqlalchemy_to_pydantic(User, exclude=['id'])

Now the server and every thing starts just fine but when I try to fetch the openapi.json, I get a keyerror because the first model is not in the model map.

I think line 36 in main.py is at fault

pydantic_model = create_model(
        db_model.__name__, __config__=config, **fields  # type: ignore
    )

The db_model.__name__ is the same two times so it must be overwriting some key (Disclaimer: I have no idea about the inner workings of pydantic but it seems to be creating some kind of global map of all models?)

I suggest to add an optional parameter name to the function and replace that line.

Hi, I like this package, and personally, I used it twice in my little projects. There are some pull requests (including mine) that are not replied to. Is this project no longer maintained? Thanks.

Using sqlalchemy_to_pydantic helped me a lot, but now I need the other way around. Is it supposed to be part of this library?

Using SQLModel is no proper alternative to me, because it makes things harder, not easier right now.