Node client for generating crude GraphQL specs from Swagger OpenAPI.
💅 Prettifies output via Prettier.
GraphQL Features | |
---|---|
Enum | ✅ |
ID | ✅ |
Union (oneOf ) |
✅ |
Non-nullable | ✅ |
Primitives (string, boolean, number) | ✅ |
Query | 🚫 |
Mutation | 🚫 |
To compare actual generated output, see the example folder.
npx @manifoldco/graphql-gen schema.yaml --output schema.graphql
# 🚀 schema.yaml -> schema.graphql [2ms]
This will save a schema.graphql
file in the current folder. The CLI can
accept YAML or JSON for the input file.
Say you have multiple schemas you need to parse. I’ve found the simplest way
to do that is to use npm scripts. In your package.json
, you can do
something like the following:
"scripts": {
"generate:specs": "npm run generate:specs:one && npm run generate:specs:two",
"generate:specs:one": "npx @manifoldco/graphql-gen one.yaml -o one.graphql",
"generate:specs:two": "npx @manifoldco/graphql-gen two.yaml -o two.graphql"
}
Rinse and repeat for more specs.
For anything more complicated, or for generating specs dynamically, you can also use the Node API (below).
Option | Alias | Default | Description |
---|---|---|---|
--output [location] |
-o |
(stdout) | Where should the output file be saved? |
--swagger [version] |
-s |
2 |
Which Swagger version to use. Currently only supports 2 . |
npm i --save-dev @manifoldco/graphql-gen
const { readFileSync } = require('fs');
const graphqlGen = require('@manifoldco/graphql-gen');
const input = JSON.parse(readFileSync('spec.json', 'utf8')); // Input be any JS object (OpenAPI format)
const output = graphqlGen(input); // Outputs GraphQL schema as a string (to be parsed, or written to a file)
(OpenAPI format), and return a GraphQL schema in string format. This lets you pull from any source (a Swagger server, local files, etc.), and similarly lets put The Node API is a bit more flexible: it will only take a JS object as input the output anywhere. It even allows for some post-processing in-between if desired.
If you are working with local files, you’ll have to read/write files yourself. Also, if your specs are in YAML, you’ll have to convert them to JS objects. A library such as js-yaml does make this trivial, though! Lastly, if you’re batching large folders of specs, glob may also come in handy.
Name | Type | Default | Description |
---|---|---|---|
namespace |
string |
OpenAPI2 |
How should the output be namespaced? (namespacing is enforced as there’s a high chance of collision) |
swagger |
number |
2 |
Which Swagger version to use. Currently only supports 2 . |
Why does this exist apart from yarax/swagger-to-graphql ?
That didn’t work for our Swagger 2.0 specs 🤷. While normally a PR is the best course of action, this repo exists because it was less effort to build something that works for us than rewrite a popular library (it wasn’t a quick fix).
While it’s possible to generate something from OpenAPI, it’s intentionally omitted from autogeneration here. Queries and Mutations are best left up to humans, so you can determine what developers should access, and how. Ultimately generating these will always fall short of how real humans could—and should—use your GraphQL endpoint.
GraphQL is a spec, just like OpenAPI. For this reason, automatic generation isn’t ideal long-term. This library should probably be used as a first-pass to migrate an OpenAPI endpoint to GraphQL. This can generate types, but can’t intelligently generate the best queries and mutations for your specific endpoint.
A common example of this: Swagger has a concept of format: datetime
.
GraphQL cares about this, but doesn’t assume the formatting. Is this UNIX
time? ISO? Are there timezones? Types can be so much more descriptive than
mere string
or int
, and GraphQL gives you the tools to declare this
yourself.