ORM is Not a Choice

Posted on April 20, 2010 by romanb


NOTE When speaking of "ORM" or "object-relational mapping" in : this post I am referring to the act of mapping an object-oriented domain model to a relational database. There are other, alternative forms of object-relational mapping.

"Should I use an ORM?" is a frequently asked question that somehow misses the point, because ORM is usually not some optional thing you can either use or not. The choices are elsewhere. Furthermore, if there is a dislike for ORM tools it helps to clarify what exactly is the cause. The cause can be a dislike of object-oriented domain models. For example, if you prefer to separate data from behavior/logic, as I've read recently on Twitter , then it is a sign that you don't like domain models, at least not rich ones, maybe anemic ones , and you probably don't like OOP much at all then, because bundling data with behavior is what OO is about, usually.

NOTE It's OK not to like OOP these days. Its not the holy grail anymore, mostly due to the usually messy concurrency characteristics and problems resulting from typical object-oriented design which revolves around imperative programming with direct manipulation of mutable state, but that does not apply to PHP as much as to some other languages due to its thread-confined nature/execution model.

You can not use a relational database in combination with an object-oriented domain model without mapping.

The choices that lead to the need for ORM are the following:

  • You can choose to use a relational database or not.
  • You can choose to create an object-oriented domain model or not.

If you want a relational database and you want an object-oriented domain model, you need ORM, there is no choice. If you choose to use a relational database but not an object-oriented domain model, you might need some other form of object-relational or other mapping, depending on how you want to model your application and your business logic around the data in particular.

People not being aware of the above two choices, especially the second one, is unfortunately what makes some of them "choose ORM" and then sometimes getting frustrated. Using an ORM without even having or wanting a domain model and with the head still exclusively full of tables and rows, which are at the beginning and at the end of every thought about the software being built, is a wrong choice. With these preconditions, ORM quickly becomes a pain, and its wrong. Forcing your relational data into objects even though you don't really know what to do with them, maybe it just seems nice to have them as "data containers", is the wrong motivation. If you don't want to combine your business data with your business behavior, something that can be done nicely in an OO domain model, then there is not much value in having objects wrapped around your data. Some other, wrong motivations for "choosing an ORM", at least when they are the main motivations, are:

  • Database vendor independence
  • "Hiding SQL"

These are just additional benefits you can get but they are not the main purpose of an ORM tool. Then what is the main purpose? State management. Synchronizing the state of your in-memory object model with a relational database for the purpose of persistent storage. Neither does an ORM need to be database vendor independent, nor does it need to hide SQL to do this. Everything else is icing on the cake. If all you want is to centralize your database access paths, which is a good thing for caching and all, you don't need an ORM for that either, any database access layer, existing or self-made, will do.

I hope it is a bit clearer now that ORM is not a choice but a need that results out of other choices.