Support for (de-)serializing SQLAlchemy models

Question | Feature Request

Maybe I’m missing something obvious but I’m wondering whether there already is or if you can see in future support for creating schemata directly from SQLAlchemy models. Maybe in a similar way to the from pydantic.dataclasses import dataclass decorator?

There exists a marshallow extension that does this quite nicely.

---

I’m especially interested in this in combination with FastAPI so maybe a plugin there is also an option @tiangolo?

[samuelcolvin]

I think this should be a separate package, happy to give my support and even run tests against it here so compatibility is maintained, but I don’t think it should be part of pydantic core.

[zzzeek]

I find when using SQLAlchemy you have to “think in sql” then reverse back to python.

If I can just clarify your observation in terms of SQLAlchemy’s approach, SQLAlchemy’s expression language is structurally as identical to real SQL as possible. The cognitive paradigm one must follow when writing SQLAlchemy queries is to think relationally, which is the same way that we think when we write SQL. IMO one should not be using a relational database if they don’t want to be thinking in terms of columns, rows, and joins. But there’s really no leap between “thinking in SQL” vs “thinking in Python” any more than there is if you were writing a SQL query for PostgreSQL but you were very used to writing them for Oracle, which has syntactical differences.