A question about the architectural design of the library (predefined configuration)

[espdev]

Hello,

I was considering the possibility of using ormar in my projects and I have a question about the architectural design of the library. Namely to create models with predefined configuration in OrmarConfig.

Here is an example from the docs:

import databases
import ormar
import sqlalchemy

DATABASE_URL = "sqlite:///db.sqlite"

base_ormar_config = ormar.OrmarConfig(
    database=databases.Database(DATABASE_URL),
    metadata=sqlalchemy.MetaData(),
    engine=sqlalchemy.create_engine(DATABASE_URL),
)


class User(ormar.Model):
    ormar_config = base_ormar_config.copy(tablename="users")

This design forces me to have a module-level (import-time) database configuration. That is, the database configuration should already exist at the time I define the model classes. This works for toy examples, but can be a problem in real large applications when the application configuration is not loaded at import-time. I'm not in favor of having any import-time application logic at all.

I'd say this kind of design isn't exactly declarative. It is rather imperative. SQLAlchemy/SQLModel/TortoiseORM don't obligate me to have some import-time database configuration. In Pydantic/dataclasses I can use default_factory to define a callable to get default values at run-time. So, it's just a declarative definition of models and that's all. I can load the config in run-time at application startup, create an engine on-demand, etc. I can flexibly manage this depending on the environment, also it's easy to test.

So I'm a bit confused and still don't understand how I can use ormar in our projects where all the configuration loading logic is done in run-time rather than import-time.

Thanks.