[avolcano]

One of the most interesting 1.4/2.0 changes is first-class asyncio support, not just for core (the query builder) but for the ORM layer as well: https://docs.sqlalchemy.org/en/14/changelog/migration_14.htm...

As this notes, there's several changes you have to make to your assumptions around the ORM interface. SQLAlchemy, for better or worse, supports "lazy loading" of relationships on attribute access - that is, simply accessing `user.friends` would trigger a query to select a user's friends. This kind of magic is at odds with async/await execution models, where you would instead need to run something like `await user.get_friends()` for non-blocking i/o.

It looks like they've done some good work in making the ORM layer work reasonably well with these limitations (https://docs.sqlalchemy.org/en/14/orm/extensions/asyncio.htm...), but I wonder if removing "helpful magic" like this will push more people to stick with the query-builder, rather than the ORM.

[aidos]

It was really interesting. I saw the discussions about it where zzzeek was like, “oh, what, I could kinda just make this wrapper thing. Wait, what am I missing here? Nope, it works”

It was a late entry into the transition rather than a planned thing from what I could tell.

Edit: post is here https://gist.github.com/zzzeek/2a8d94b03e46b8676a063a32f7814...

[orf]

So the secret sauce was just greenlet?