Skip to content

Open API (OAS 3.0) Typescript generator

License

Notifications You must be signed in to change notification settings

Embraser01/typoas

Repository files navigation

Typoas

npm
npm
npm
GitHub
npm bundle size

Typoas is an OpenAPI 3.X generator for Typescript. It's inspired by openapi-generator
but is written in Typescript for Typescript. The generator uses the Typescript AST to generate code instead on relaying
on templates which allows better schemas definitions and other cool stuff.

Main features are:

  • Fully typed and fully customizable
  • References $ref handling (including cyclic refs)
  • Tree Shaking out of the box
  • React Query integration
  • Support for allOf, oneOf and anyOf schemas.
  • Automatically convert format: 'date-time' to JS Date
  • Handle API Key, HTTP Config, OAuth21 and OIDC1 auth security schemes
  • JSDoc for schemas and operations
  • Uses fetch api (can be customized)
  • Non JSON content type support
  • Small bundle size
  • And more...

1: OAuth2 and OpenIDConnect scheme do not handle flows to retrieve an accessToken.
You need to provide your own accessToken through the provider.getConfig() function.

The project is split into 4 packages:


Installation

It will generate a single TS file containing all the code specific to the underlying API.
This file only has a single dependency on @typoas/runtime.
You need to manually add @typoas/runtime to your dependencies.
It handles common things like serialization/authentification

Usage

Use the generator from the CLI

You can generate the TS client from the spec from the command line:

yarn dlx @typoas/cli generate -i my-spec.json -o src/client.ts
npx @typoas/cli generate -i my-spec.json -o src/client.ts

Here is a short list of supported command line options:

    -i, --input [path/url]         Path or URL to the OpenAPI JSON specification (yaml/json format)
    -o, --output [path]            Path where to write the generated TS file
    -e,--generate-enums            Generate enums instead of literal string types where possible
    -p,--prettier                  If set, the generated file will be formatted with Prettier
    -r,--full-response-mode        Enabled by default, generate functions that only throws on network errors, can be disabled using --no-full-response-mode
    --js-doc, --no-js-doc          Whether to add JS Doc to the generated code (default: true)
    --wrap-lines-at                Define a maximum width for JS Doc comments, 0 to disable (default: 120)
    --only-types                   Use it to only generate types in #components/schemas/
    --no-fetcher-options           Use it to disable the additional param added to every operations
    --version                      Output the version number
    -h, --help                     Display help for command

or you can use it in code:

import cli from '@typoas/cli';

await cli.run(
  [
    'generate',
    '-i',
    'https://raw.githubusercontent.com/github/rest-api-description/main/descriptions/api.github.com/api.github.com.yaml',
    '-o',
    `./src/github.ts`,
  ],
  {
    stdin: process.stdin,
    stdout: process.stdout,
    stderr: process.stderr,
  },
);

Use the generator from the API

The API is still at an alpha stage, so it may break between minors.

It uses typescript API to generate usable code:

import { readFileSync, writeFileSync } from 'fs';
import { createPrinter, NewLineKind, SourceFile } from 'typescript';
import { generateClient, getStringFromSourceFile } from '@typoas/generator';

const specs = JSON.parse(readFileSync('path/to/github-openapi.json', 'utf8'));
const src = generateClient(specs, {
  /* options */
});
const data = getStringFromSourceFile(src);

writeFileSync('./src/client.ts', data, 'utf8');

Use the generated code

Once the file is generated you'll be able to use it like this:

import { ServerConfiguration, ok } from '@typoas/runtime';
// Client generated from:
// yarn dlx @typoas/cli generate -i https://raw.githubusercontent.com/github/rest-api-description/main/descriptions/api.github.com/api.github.com.yaml -o client.ts
import { createContext, pullsList } from './client';

const ctx = createContext();

// By default, returns a StatusResponse that contains statusCode, data and headers.
const res = await pullsList(ctx, { repo: 'typoas', owner: 'embraser01' });
if (res.ok) {
  res.data.forEach((pr) => console.log(pr.title));
}

// An helper `ok` function can help throw on non 2xx status code and return the data
// The second argument is an optional expected status code (useful when multiple success status code are possible)
const prs = await ok(
  pullsList(ctx, { repo: 'typoas', owner: 'embraser01' }),
  200,
);

You can directly generate functions that throw on non 2xx status code by using the --no-full-response-mode option. This feature may be removed in the future and isn't the recommended way.

You can customize multiple things in the createContext function.

Customize server configuration

By default, the createContext function will use the first servers entry in the spec. You can override the server configuration by passing a ServerConfiguration object:

import { ServerConfiguration } from '@typoas/runtime';
import { createContext } from './client';

const ctx = createContext({
  server: new ServerConfiguration('https://{env}.api.example.com', {
    env: 'staging',
  }),
});

The way it's done currently is not the best. It will be improved in the future.

Customize fetch implementation

The default fetcher implementation uses the fetch API. You can override the fetcher implementation if you want to customize things (like adding a retry policy, headers, etc.).

For example, you can use axios instead of the fetch API:

import axios, { AxiosError, AxiosResponse, Method } from 'axios';
import { Fetcher, RequestContext, ResponseContext } from '@typoas/runtime';
import { createContext } from './client';

class AxiosHttpLibrary implements Fetcher {
  async send(request: RequestContext): Promise<ResponseContext> {
    let resp: AxiosResponse;
    try {
      resp = await axios.request({
        url: request.getUrl(),
        method: request.getHttpMethod() as Method,
        headers: request.getHeaders(),
        data: request.getBody(),
      });
    } catch (e) {
      const err = e as AxiosError;
      if (!err.response) {
        throw e;
      }
      // Typoas handles errors itself
      resp = err.response;
    }

    return new ResponseContext(resp.status, resp.headers, {
      text: async () => resp.data as string,
      binary: async () => resp.data as Blob,
      json: async () => resp.data,
    });
  }
}

const ctx = createContext({ fetcher: new AxiosHttpLibrary() });

Node.js >= 18.0.0 includes the fetch implementation, so Typoas should work out of the box.

Handling authentification

The createContext function can take an authProviders object to handle authentification.

For example, for the petstore sample:

import { createContext } from './petstore';

const ctx = createContext({
  authProviders: {
    api_key: {
      async getConfig() {
        return 'MyAPIKey';
      },
    },
    petstore_auth: {
      getConfig() {
        const accessToken = globalStore.getAccessToken();

        if (accessToken) {
          return null;
        }
        return { accessToken };
      },
    },
  },
});

It supports 5 types of security schemes:

  • apiKey mode
  • http bearer and basic mode
  • oauth2 mode
  • openIdConnect mode

The getConfig function should return the configuration needed to authenticate the request. Returning null will skip the authentification.

Customizing the serialization

You can switch some serialization options by passing a serializerOptions object to the createContext function.

Not every serialization option is supported. See #11 for more information.

Using with React Query

Documentation is available in the @typoas/react-query package.

Examples

You can find examples in the examples folder.

Notes

Here is some notes on some known issues.

External references

External references are not supported. Every $ref must be in the spec.
An issue is open here.

Parameters serialization

Typoas has partial support for serialization specified here: https://swagger.io/docs/specification/serialization/

  • It does support array serialization for query.
  • It does NOT support serialization for path parameters with style label or matrix.
  • It does NOT support serialization for query parameters with nested objects. It will be JSON.stringify
  • It does NOT support serialization for headers or cookie parameters.

On query serialization, there can only be one style for a full operation. The first query param will set the style for
the whole operation.

Migrating from v1 to v2

In v1, the whole API was generated in a single class. In V2 this was replaced by individual function
which allow Tree Shaking. To get a similar result,
you can use wrapApi helper:

import { ServerConfiguration, wrapApi } from '@typoas/runtime';
import { createContext, pullsList, issuesList } from './client';

const ctx = createContext();
const ghClient = wrapApi(ctx, {
  pullsList,
  issuesList,
  // ...
});

ghClient
  .pullsList(ctx, {
    repo: 'typoas',
    owner: 'embraser01',
  })
  .then((list) => console.log('List of PRs', list))
  .catch((err) => console.error('Error while getting PRs', err));

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT