How to generate JSON-Schema from Swagger API Declaration

Question

I have Swagger API Declaration for services using Swagger v 1.2

My original feeling about Swagger was, that it is very close to JSON Schema (Draft 3 and lately Draft 4)

Answer 1

I've had some success doing it like this:

swagger.yaml -> proto -> jsonschema

I used openapi2proto to generate proto files from Swagger yaml, then protoc-gen-jsonschema to generate the JSONSchemas from that. It's good enough to get a typed JSONSchema, but proto3 doesn't support "required" types so you miss out on this.

Answer 2

After longer fight with using Swagger for specifying REST API and reusing it in related test suites, I will share my own experience with it (answering my own question).

Swagger supports only subset of JSON Schema Draft 4

Specification of Swagger 1.2 and 2.0 states, it supports only subset of JSON Schema Draft 4 (s. here). This means, that:

• one cannot rely, that each valid JSON Schema can be completely supported by Swagger.
• thinking of XML, Swagger supports only canonical representation of subset of JSON structures provided by JSON Schema Draft 4.

In other words:

• Swagger (1.2 and 2.0) does not support usage of many JSON structures, which are valid in terms of JSON Schema Draft 4 (the same applies to Draft 3).
• Swagger does not support general XML data structures, only very restricted structures are allowed.

In practice, you cannot start with designing your data in JSON or XML, with Swagger you have to start and end with Swagger.

Getting JSON Schema is theoretically possible, but not easy

I have spent some time coding a library, which would take Swagger API Specification and create JSON Schema Draft 4. I gave up for couple of reasons:

• it was not easy at all
• got disappointed finding, that I can use only subset of what JSON Schema provides. We had some JSON payload already proposed and had to start modifying it just to fit what Swagger specification framework allows.

Apart from having really nice looking UI for showing and testing the API (yes, everybody agrees, it is visually very pleasing), I have found it weird, that a specification framework is not allowing us to use what we want, but adds unexpected restrictions to our design.

If you want full JSON or XML Schema support, use RAML

Researching other API specification frameworks, I have found RAML. As it is built from ground up by supporting any JSON Schema Draft 3/4 or W3C XML Schema 1.0 data structures, the experience was excellent - having structure of my payload designed, I was able authoring API specification very quickly and following validation of real requests and responses against defined schemas was very easy, as the schemas are essentials components of the specification without adding any restrictions on them.

RAML was at version 0.8 that time (version 1.0 was not released yet).

Correcting the question leads to real solution

Good question makes half of the solution. My question was wrong as it missed fulfilling my real expectations. Corrected question would be:

What specification framework and technique to use, to specify REST API using payload defined by arbitrary JSON Schema Draft 4 or W3C XML Schema 1.0.

My answer to such a question would be:

1. Design your payload in JSON Schema Draft 4 or W3C XML Schema
2. Describe your REST API by means of RAML (v0.8 at the moment).

There might be other specification frameworks usable, but Swagger (neither v1.2 nor v2.0) is definitely not the case. Apart from providing really a lot of features (code generation, very nice looking documentation of the API and much more), it simply fails in providing solution to the updated question stated above.

Answer 3

There is a python tool to do the same by the name openapi2jsonschema. You can simply install it using pip.

The readme for openapi2 shows the simplest way to use it:

openapi2jsonschema https://raw.githubusercontent.com/kubernetes/kubernetes/master/api/openapi-spec/swagger.json

Hope this helps.

Answer 4

I just wrote a tool pyswagger seems fit your need.

It loads Swagger API declaration, and able to convert python object to/from Swagger primitives. Also provide a set of client implementations (including requests & tornado.httpclient.AsyncHTTPClient) that able to make request to Swagger-enabled service directly.

This tool tends to solve the first problem I encountered when adapting Swagger in python, and is still pretty new now. Welcome for any suggestion.