A pattern I'm seeing increasingly, is to compose OpenAPI specification (OAS, formerly known as Swagger) documents from multiple files. This allows re-use, easier collaboration and makes larger documents much easier to follow.
The resultant document can be parsed by tools which fully understand and implement the JSON Reference specification (i.e. they include a resolver), or it can be 'bundled' by de-referencing external references using such a JSON Reference-implementing tool.
But what should those fragments of OAS documents look like?
The OAS specification makes no rulings on this, other than the object being referenced must be of the expected type.
Thus a parameter definition in a self-contained OAS document may look like this:
...
paths:
/:
get:
parameters:
- $ref: '#/components/parameters/sort'
...
components:
parameters:
sort:
name: sort
in: query
schema:
type: string
description: 'The direction of the sort'
enum:
- asc
- desc
A parameter referencing an external fragment may look like this:
...
paths:
/:
get:
parameters:
- $ref: './includes/parameters.yaml#/sort'
and parameters.yaml could simply look like this, with no additional structure:
sort:
name: sort
in: query
schema:
type: string
description: 'The direction of the sort'
enum:
- asc
- desc
If we wished to compose our reusable components across subject-matter areas rather than structurally relating to the OAS, we might structure our documents like this:
paths:
/:
get:
parameters:
- $ref: './includes/parameters.yaml#/parameters/sort'
parameters:
sort:
name: sort
in: query
schema:
$ref: '#/definitions/sortType'
definitions:
sortType:
type: string
description: 'The direction of the sort'
enum:
- asc
- desc
The problem with both styles of included document, is that neither is defined by any standard other than being valid JSON or YAML. Neither can be validated against a specification, even though their contents should be composed of valid OAS objects.
It raises questions such as: is it allowable to share the definition of the sortType schema across OAS v2 and OAS v3 documents, even though the definition of the Schema Object differs between the two specifications?
I would argue there are significant benefits in always structuring your included documents as fully valid OAS documents.
At the cost of the following six lines of metadata overhead (a little more in JSON), and slightly longer $ref values, our parameters.yaml document becomes:
openapi: 3.0.1
info:
title: An include file to define sortable attributes
version: 1.0.0
paths: {}
components:
parameters:
sort:
name: sort
in: query
schema:
$ref: '#/components/schemas/sortType'
schemas:
sortType:
type: string
description: 'The direction of the sort'
enum:
- asc
- desc
This document now has the benefits that it can be validated, linted, converted, transformed and edited using OAS compliant tools. In the case of conversion, it means OAS v2 document fragments can be upgraded in place to OAS v3 without having to bundle the document into a monolith.
The minimal info object gives us properties to describe the expected usage of the fragment, and an ability to version it separately from the 'master' OAS documents which reference it.
Simply by the presence of the empty paths object, we can tell we are dealing with a fragment, not a fully-defined OAS document.
These reusable sub-documents can also be shared between projects, allowing for industry-specific standard components to emerge.
Top comments (11)
You say:
But spec maintainer thinks differently:
Source: github.com/OAI/OpenAPI-Specificati...
That's what Ron Ratovsky said some time ago. He's allowed to change his mind. :)
Ron is also one of the maintainers, but then so am I...
Personally, I've recently made a study to compare different options to define the structure of the data exchanged between the systems using JSON encoding. The final choice I had to make was not OpenAPI because:
I really want to be able to talk just about the data. It might be RPC. It might be messaging. But the data will be (mostly) the same. Of course, you also want to define transport thingies like paths and response codes as well once you get to it. But the data and its structure exchanged by the systems are largely independent of particular transport.
I fear that once unapproved workaround like empty paths is added, the tooling support could degrade rapidly. Again, talking about the data, I would like to be able to validate JSON file for compliance with defined structure. I want to be able to do this in runtime, during build time and when editing the schema or file with data. Most tools for OpenAPI are around REST and it is not quite trivial to validate chunk of data.
So what was your choice ?
Schema language for BLA-BLA-BLA unified format
Changes
Executive summary
It is recommended to use JSON Schema to define JSON encoding of unified format of BLA-BLA-BLA.
Introduction
Unified format for BLA-BLA-BLA needs a formal protocol definition. For this, a schema definition language should be chosen. A universal schema definition language that can be used to define data structures that can later be expressed using several encoding formats like XML and JSON are preferred. However, this is not a mandatory requirement since industry moves away from XML while JSON adoption grows larger and larger. Thus, support for JSON is mandatory requirement while support for other technologies is optional.
Schema definition language should have good tooling support:
Schema definition language should have wide adoption in the industry to avoid usage of a dead or rare technology.
JSON Schema
Home
Official learning
Specification
Books
Tooling
Validate sample by schema – build
Validate sample by schema – runtime
Validate sample by schema – support
Schema authoring
Data authoring
Summary
3GPP TS 28.541 uses JSON Schema to define data model.
Very good candidate to describe JSON data format independent of API/RPC/etc. Tooling support is good enough. However, while it is widely adopted, the specification is not yet finalized. This creates a risk that specification might change in incompatible way and schema will have to be adjusted.
Yang
Home
Official learning
Specification
Books
Tooling
Validate sample by schema – build
Validate sample by schema – runtime
Validate sample by schema – support
Schema authoring
Data authoring
Summary
3GPP TS 28.541 uses Yang to define data model.
While Yang model is independent of encoding and Yang supports both XML and JSON, it seems to be a bad choice. The reasons are:
OpenAPI
Home
Official learning
Specification
Books
Tooling
Validate sample by schema – build
Validate sample by schema – runtime
Validate sample by schema – support
Schema authoring
Data authoring
Summary
Is not a data format definition language but API definition language. Requests to allow it to define the data are rejected. Quote: “It's meant to describe APIs, not catalogs of objects that can be used in APIs.” Source: github.com/OAI/OpenAPI-Specificati... While there are technical ways to circumvent this, this is discouraged by language designers. It is risky since the tools are unlikely to support validation once these workarounds are in place.
OpenAPI is based on JSON Schema but has additional restrictions to define APIs. One can be converted to another manually and using specialized tools.
OpenAPI is used to define 3GPP TS 28.532 and TMF 642 APIs.
Due to complexity and restrictions for usage as data format schema rather than API schema, it is recommended to use a better alternative when available.
Why overload the the semantics of
pathand not simply add a"fragment: true"to theinfoobject? ("x-fragment"until it is adopted in the standard)Sure, it does not make much sense to have an API with no paths. But it will make it explicit that the specification is a fragment.
The presence of
paths: {}isn't intended to overload the meaning, it's becausepathsis a mandatory object in the OAS (2.0 and 3.0.x), although it can be empty.x-fragmentis a good idea to make that explicit.Thank you for this post.
Currently I am working on a software project to develop a catalog of software tooling. As part of this project, I have developed a JSON Schema for describing tools. I want to avoid repeating myself, so I would like to incorporate this schema into my API contract which I am currently writing in OpenAPI 3.0. Do you think it would be problematic if I just incorporated the schema directly, even though I would be embedding a JSON Schema document into an OpenAPI document? What do you generally recommend for interoperability between the two?
If you can limit yourself to the common subset of JSON Schema draft 5 and OpenAPI 3.0 schema objects, you'll be fine. If that's a problem, take a look at speccy (github.com/wework/speccy) which implements a filter to convert JSON Schema to OAS schema objects, using github.com/wework/json-schema-to-o...
Another possible approach is to use the
x-oas-draft-alternativeSchemaextension (github.com/OAI/OpenAPI-Specificati...) although as this is brand new, we don't yet know of any implementations - so would probably only suit your own tooling.Hii
You can also use onlinejsontools.org/ for json validator,beautify,minify,xml,yaml,CSV,bson,plain text,base64,tsv.
Do checkout this site!