I cannot say that I have extensive experience with ORMs. After a recent bout or two with them, I found that there was something about them that nagged at me. Tonight, I realized what that something was. ORMs exist as a middle layer between an object oriented programming language and an entity-relational database. ORM advocates refer to what they do as solving the issue of "impedance mismatch"[1].
Herein lies the rub. ORM advocates are not solving the problem wrong, but they are solving the wrong problem altogether. They have, correctly, noted that an ER database is not an object store. However, they write the ORM as a large hack so that one can treat an ER database as an object store. They would do far better to write or use an object store--something actually designed to have a 1-1 correspondence between a "record" and an instance of a class.
There has been quite a stir lately with the rising of "NoSQL" solutions. These are, I think, little more than the old object-oriented databases coming back from the dead. While most are key-value stores and there fore not strictly speaking object-oriented, it is not hard to realize that most adapt themselves more readily to use as object stores. I think that this is where the future is heading for ORMs. In due time, the developers who today write and use ORMs will move their work over to document oriented NoSQL like databases, as these are faster (I would argue inherently, since you are not wasting time copying excess data back and forth and all about) and easier still than the ORM layer itself.
This leads to the obvious question: will a convergence of the NoSQL camp and ORMs result in the death of relational databases? I do not think so. ER databases are not bad because they are not object stores. They have their own advantages and I think they will be with us far into the future. The model has not survived forty years on account of having a poor foundation. ORMs, however, I think we can do without. Rather than solving an impedance mismatch, they are a crutch to use the wrong tool for the wrong job.
References