This document intends to establish guidelines which build a transparent, open mechanism for deciding how to evolve the OpenAPI Specification. The Open API Technical Contributor Board will initially follow these processes when merging changes from external contributors or from the TCB itself. This guideline document will be adjusted as practicality dictates.
The OpenAPI Specification should be use-case driven. We can specify support for hypothetical use cases as we see fit, but specifications should be backed by realistic scenarios.
The specification will change from the original 2.0 version. We should typically make changes when any of the following criteria are met:
- Clarity. The current "way" something is done doesn't make sense, is complicated, or not clear.
- Consistency. A portion of the specification is not consistent with the rest, or with the industry standard terminology.
- Necessary functionality. We are missing functionality because of a certain design of the specification.
- Forward-looking designs. As usage of APIs evolves to new protocols, formats, and patterns, we should always consider what the next important functionality should be.
- Impact. A change will provide impact on a large number of use cases. We should not be forced to accommodate every use case. We should strive to make the common and important use cases both well supported and common in the definition of the OAI Spec. We cannot be edge-case driven.
- Use GitHub for all spec designs, use cases, and so on.
- As with 2.0, the human readable document is the source of truth. If using a JSON Schema again to document the spec, it is secondary to the human documentation. The documentation should live in a *.md file, in parallel to the 2.0 document (versions/3.0.0.md for example).
- The
master
branch shall remain the current, released OpenAPI Specification (i.e., 2.0). We will work in an OpenAPI.next branch, which shall be described and linked to on the default README.md on master. - Examples of how something is described currently vs. the proposed solution should accompany any change proposal.
- New features should be done in feature branches which, upon approval, are merged into the OpenAPI.next branch.
- Use labels for the workflow of specification changes. Examples of labels are
proposed
,needs migration review
,needs tooling review
,needs documentation
,rejected
, andneeds approval
. These labels must be assigned by project committers. - An issue will be opened for each feature change. Embedded in the issue, or ideally linked in a file via pull-request (PR), a document about use cases should be supplied with the change.
- A PR will be used to describe the proposed solution, and linked to the original issue.
- Not all committers will contribute to every single proposed change. There may be many open proposals at once, and multiple efforts may happen in parallel.
- When the OpenApi.next spec is complete and approved for release, the branch will be merged to master.
For each change in the specification we should always consider the following:
- Migration. Is this a construct that has a path from the existing 2.0 specification? If so, how complicated is it to migrate to the proposed change?
- Tooling. Strive to support code generation, software interfaces, and spec generation techniques. Some features may be impossible to support in different frameworks/languages. These should be documented and considered during the change approval process.
- Visualization. Can the specification change be graphically visualized somehow in a UI or other interface?
Spec changes should be approved by a majority of the committers. Approval can be given by commenting on the issue itself, for example, "Approved by @fehguy". After voting criteria is met, any committer can merge the PR. (TODO: we will want to formalize what voting criteria actually is).
No change should be approved until there is documentation for it, supplied in an accompanying PR.
We should always be as transparent as possible. Sometimes there will be discussions that use customer names, sensitive use cases, and so on. These must be anonymized, discussed in a private repository, or conducted offline.
- Asynchronous discussions should live in the GitHub issues for this project.
- Realtime discussions should be in a public chat such as IRC or Slack.