What I don't get is why do I need a "stable" data store to iterator over? On a project I once didn't get a database until 2 weeks before the actual project was

What I don't get is why do I need a "stable" data store to iterator over? On a project I once didn't get a database until 2 weeks before the actual project was due. Fortunately I was using Spring. So I just mocked the DAL until such time as I actually got a real one. This whole time I changed the contract on the DAL, changed how I used it, etc. Then when the DBA finally got around to having time to make my db I presented him with a decent, thought-out ERD. Sure the mocks had to do some work, but a simple cache allowed me to perform all of the CRUD operations in memory. I can see doing something similar with Mongo/Couch, but having done the DAL with a pure mock set injected via Spring, I don't really see the point. The same goes for HQL or another lightweight in memory DB + Hibernate/JPA. I assume the model of interaction would work with Python or similar languages too.