Why can’t I call my API RESTful?
Semantic debates are fun, aren’t they? I can build an API and call it a REST API, then someone can say it isn’t truly REST and now we’re back to arguing about what is and isn’t REST. The most common point of contention is whether or not to follow the HATEOAS constraint.
For those not familiar, HATEOAS is one part of the “Uniform Interface” constraint. Let’s look at those constraints in full for context:
- Identification of resources — This basically amounts to using URLs for each resource as described above (assuming we’re using HTTP as the protocol, of course…)
- Manipulation of resources through representations — Send a resource represented in e.g. an XML or JSON format of some kind and that should be enough for a client to read or update the resource.
- Self-descriptive messages — Use headers like
Content-Typeto make it clear how to parse responses.
- Hypermedia as the engine of application state (HATEOAS) — A client should only be able to move between resources by following hypermedia links and controls on a particular representation.
The definition of HATEOAS above is an attempt to summarise it in a single sentence, but if you’re not familiar with it at all, it’s not likely to be any clearer that that actually means but also why you might use it.
In fact, Roy Fielding — who coined REST in his famous PhD thesis — has himself in the past attacked HTTP APIs that claim to be “REST” APIs when they do not. There have also been examples of disagreements on whether HATEOAS is needed for a REST API. That there is a significant level of confusion or even disagreements and arguments suggests a lack of clarity around what HATEOAS truly is and why you might use it.
The argument usually boils down to people pointing that out failure to follow the REST constraint does not a REST API make. The common retort is that it is zealous or purist to ask for this constraint to be followed.
A whole camp of API designers will furthermore claim the REST-minus-HATEOAS architectural style to be valid in its own right. Some even use terms such as Pragmatic REST to describe/justify/differentiate their “non-purist” API design.
What’s missing in these arguments? Something that’s missing in a lot of discussion on the web about this: any concrete discussion of the properties of the thing originally built.
There’s two distinct fallacies that appear when a debate around a definition or semantics emerges:
- we talk too much about whether something truly fits a definition without actually discussing the merits of that thing; or
- we dismiss opinions of those who disagree with us by incorrectly regarding a debate as being “just semantics”.
If you already have an opinion on the REST vs. “pragmatic” debate, you might well recognise one of those fallacies in those with whom you disagree, but be careful of falling into the other fallacy in your own views.
Both of these fallacies can be seen as extreme ends of a spectrum where at one end we get sucked into pointless debates over definitions and at the other end we see no value in discussing definitions at all, which is also irrational.
It is indeed the case that an argument about whether my API is RESTful or not is not going to make much progress if we only discuss back and forth as to what REST is or whether it’s useful to use the word only in a purist sense or whether we can allow for “pragmatic” takes on the term. What’s lacking here is discussing it in the context of the API I am building.
Conversely, we should not be so quick to shoot down those that point out where my API does not conform to REST. It is tempting to call such a person zealous or a purist and frame any of their arguments as irrational and therefore not worth hearing by definition.
The more rational, balanced position is to discuss the definition in terms of why all the REST constraints exist and also in the context of a particular problem domain being solved by a “RESTful” API. Here we can rationally see if the constraints not being followed would help or hinder us; we can revisit the advantages brought by lesser-used constraints such as HATEOAS. In this context, the definition is certainly useful and we are not longer arguing “just semantics”.
Does it matter?
So, surely it’s still my business whether I call my API RESTful or not? Well, sort of.
Semantics — as in the meaning of words — relies at least to some extent to people loosely agreeing what words mean. If you present a software system or a technology and use certain words to describe it, then expect confusion when you use a word differently to how others use it. I’m not saying we need to be overly pedantic about words, but if you build an API and call it RESTful, expect some confusion by some who will expect to call it in line with all of the REST constraints.
In some ways, I would plead with people to avoid using the term REST if they are consciously choose not to follow the architectural style fully, but I fear we may have lost control of the term by now to the point where it’s hard to trust any meaning has been followed. In practical terms, it means that if someone presents an API to me and tells me it’s a REST API, I still don’t quite know what to expect until I see it. It is less than ideal for any word to get to a point where it doesn’t actually convey information any more. This is especially true in a technical industry like software engineering.
So, technically you can call your API what you want and really the debates should centre around what is and isn’t useful for the problem you are trying to solve. However, it is worth being mindful of what you are intending to communicate when you use a technical term. Expect people to be confused if you follow a mutated definition of a term.
I personally like to follow they hypermedia/HATEOAS constraints for RESTful APIs and in later posts I’ll discuss why that is and the decoupling people are missing out on. It is also my contention at times that if you do not want the full decoupling of REST in our distributed systems, then it’s questionable why you needed to use any of REST in the first place. That certainly warrants a whole other post.