Friday, October 21, 2022 · 6 min read

Towards a stable JSON Schema

About this time last year, I hosted a discussion at the API Specification Conference about the future of JSON Schema. The most popular topic of discussion was, when JSON Schema is going to be "done". Of course we've heard that question a lot. The question stems from the "draft" label we put on your releases. The term "draft" has led to quite a bit of confusion for our community over the years, so let's take a moment to understand where it comes from.

Why is it a "draft"?

JSON Schema has been loosely following the IETF standards track RFC process. That means our releases are in the form of an Internet Draft (I-D). That's why we refer to them as drafts. However, because of JSON Schema's wide use in production systems, it's not really possible to treat the spec like a typical I-D. Therefore, our use of the term "draft" is a bit of legacy artifact from when the IETF process made more sense for JSON Schema.

This is has been a problem because when people hear "draft", they hear "unfinished" or "not production ready". That's not the way we treat our releases. Every release is expected and encouraged to be used in production. It's no different than OpenAPI releasing a new version. No one asks when OpenAPI is going to be "done". JSON Schema is just perceived differently because we call our releases "drafts".

The real problem

But this isn't just a branding problem. When people ask when JSON Schema is going to be out of "draft", what they really mean is, when is JSON Schema going to be "stable". They want to be able to write a schema and be sure that it will continue to work the same way no matter how JSON Schema evolves in the future. They want to be able to update their dependencies and not have to update their schemas that were working just fine.

This effects library maintainers as well. Needing to support multiple versions of JSON Schema with no backward or forward compatibility guarantees gets cumbersome and has led many maintainers to drop support for older releases. When this happens, it can leave users with a choice to have to update all of their existing schemas that weren't broken or pin to a no-longer-supported version of the JSON Schema library they are using.

Our solution

Those are the problems we're aiming to solve in the next release. Instead of continuing to release a new immutable and incompatible version of JSON Schema with each release, our next release will be a long-lived version that is stable, but evolving. In this case, "stable" means that there will be strict backward and forward compatibility requirements that must be followed for any change. It will be much like JavaScript in that as it evolves you can always be sure that your existing schemas will continue to work with any JSON Schema library you're using, but you take a risk using newer features because not all libraries will have implemented those features yet.

That vision of a stable yet continuously evolving spec doesn't fit well with the IETF process. There are paths we considered, but nothing was proposed that we thought would allow us to continue to evolve the standard and get out of "draft" any time soon. Therefore, the first step to achieving our vision is to decouple our main spec development from the IETF process. This split allows us to pursue a new model for the main spec development that is more conducive to our vision.

Whether or not you're a fan of the direction the JavaScript language has evolved, it seems clear that they've come up with an effective process for allowing for continuous evolution without sacrificing interoperability and longevity. That's why we've chosen to base our new process on the process that is used to evolve the JavaScript language. In our next release, most of the keywords and features you use today will be declared stable and they will never change in a backward incompatible way again. Features and keywords we aren't comfortable making stable yet will become part of a new staged release process that we are working on defining now. The goal of the staged release process is to ensure that the feature gets sufficient implementation, testing, and real world vetting for us to be confident in declaring it stable. The process should not only make us more confident, but also allow us to achieve that confidence much more quickly.

Standards considerations

Starting with the next release, the JSON Schema specification will be self-published on our website.

One of the concerns about self-publishing is about other standards being able to reference the JSON Schema specification. We have received feedback from people involved in standards development that our approach would be acceptable for them to reference our specification in their standards based on our membership with the OpenJS Foundation. We don't know if all standards bodies will come to the same conclusion, but this feedback gives us confidence that it isn't going to be a significant problem.

Although the main specification will be self published, we are continuing to work through the IETF process where it makes sense. For example, we are in the process of registering our media types such as application/schema+json through the HTTPAPIs WG. We are also looking into standardizing reusable components such as Relative JSON Pointer through IETF.

The bottom line

Details about the new process will be shared in a separate post when they are finalized, but here are some outcomes users can expect.

  • If you use only stable features, you are guaranteed interoperability between JSON Schema libraries and you will never need to update your schemas just for the sake of keeping up with new releases.
  • You can safely use new features before they are stable, as long as the libraries consuming your schemas support the feature.
  • Compatibility/Interoperability guarantees will only apply to the next release and beyond. You will need to update your schemas to the stable version, but you won't have to continue to update them as JSON Schema evolves.
  • Custom dialects and vocabularies will continue to be central concepts for customizing and extending JSON Schema.
  • Implementers won't need to maintain distinct code to support past stable releases. A library that supports the 2025 release will automatically support the 2023 and 2024 releases. The past stable releases will no longer need to be maintained as distinct versions. However, implementations that continue to support the "draft" releases would still need to maintain those as distinct versions from the current stable release.