Everything I know about Python...

Learn to Write Pythonic Code!

Check out the book Writing Idiomatic Python!

Discuss Posts With Other Readers at discourse.jeffknupp.com!

Why I Hate HATEOAS

Most of the population of people who have read Roy Fielding's dissertation tell us that, while we may understand HTTP verbs, we don't know anything about hypermedia. Our REST APIs, they say, are not really RESTful, and certainly do not exhibit Hypertext As The Engine Of Application State, or HATEOAS. The fact that we have APIs you need to read documentation to understand is a clear signal, they tell us, that the client is using out-of-band information to navigate the server. Hypermedia, they say, is the answer. We simply embed enough information in our responses that the client can intelligently choose what action to take next using only the contents of the HTTP response.

Why This Is Tremendously Stupid

When designing a hypermedia API, you're really designing for a client that does not, and will never, exist. Imagine you're asked by your manager to create a "REST API" for your business. Knowing the latest trends in API design, you come up with a hypermedia-driven design that uses HAL or some such nonsense. When your manager asks you if the API is ready, you answer in the affirmative.

When he asks if the documentation is ready as well, you gleefully tell him that such extraneous information is not necessary, and that your API responses include enough information for smart clients to use the API without any explicit documentation, minus a brief spec that describes what your rel values mean and so-forth.

Your manager then asks you again: "Is the documentation also ready?"

At this point, you hurry to write the "standard" REST documentation with explicit endpoints, the verbs they accept, and how to use them.

What you've really provided by writing the "standard" documentation is exactly the thing that makes hypermedia APIs ridiculous: lack of semantic understanding. There is no magical "smart client" that somehow knows that rel=comments means that the link leads to comments about the current resource and can figure out it should POST there to create a new comment. It has no idea what the hell a "comment" is.

Unless you tell it explicitly. But using out-of-band information to give meaning to your content is not acceptable. Instead, you must wait for the "smart client" that will never exist. Why am I sure of that? Because if it did, then it would effectively be a single client that could make use of every single (properly built) REST API in existence without requiring documentation. What would that even mean?

Here's a question:

Why are we collectively trying to build servers for which there can never be acceptable clients.

People are trying to wrangle their APIs into hypermedia formats or (worse) inventing new ones, all because someone decided that everything must be self-documenting, despite the fact that the consumers of that implicit documentation (machines) have no hope of understanding it.

Why are we killing ourselves to build HATEOAS systems? Are today's "don't call me REST" REST APIs so broken that we must write code with our heads in the sand, pretending that if we solve the server side issues, the client side will eventually catch up? Well, news flash. It won't catch up. Ever.

Keep Doing What You've Been Doing

Rather than striving for hypermedia systems, design ones that people can actually use without having to read 12 different specs and no documentation. Today's automated REST API documentation tools are fantastic, use them. Keep building "REST" APIs and calling them that, because there is no sensible alternative.

Of course, if you want to design a system using principles which, by definition, your system can never be said to use, go right ahead. In the meantime, I'll be getting stuff done.

Posted on by

Discuss Posts With Other Readers at discourse.jeffknupp.com!

comments powered by Disqus
Web Analytics