When using SQL Alchemy for abstracting your data access layer and using controllers as the way to access objects from that abstraction layer, how should joins be handled?

So for example, say you have an Orders controller class that manages Order objects such that it provides getOrder, saveOrder, etc methods and likewise a similar controller for User objects.

First of all do you even need these controllers? Should you instead just treat SQL Alchemy as "the" thing for handling data access. Why bother with object oriented controller stuff there when you instead have a clean declarative way to obtain and persist objects without having to write SQL directly either.

Well one reason could be that perhaps you may want to replace SQL Alchemy with direct SQL or Storm or whatever else. So having controller classes there to act as an intermediate layer helps limit what would need to change then.

Anyway - back to the main question - so assuming you have these two controllers, now lets say you want the list of orders for a certain set of users meeting some criteria. How do you go about doing this? Generally you don't want the controllers crossing domains - the Orders controllers knows only about Orders and the User controller just about Users - they don't mess with each other. You also don't want to go fetch all the Users that match and then feed a big list of user ids to the Orders controller to go find the matching Orders.

What's needed is a join. Here's where I'm stuck - that seems to mean either the controllers must cross domains or perhaps they should be done away with altogether and you simply do the join via SQL Alchemy directly and get the resulting User and / or Order objects as needed. Thoughts?