apigentools

Purpose

The purpose of the apigentools package is to provide an easy way to generate API clients from an OpenAPI v3 specification for multiple languages using openapi-generator.

Note on "downstream/upstream" nomenclature: Throughout the code and documentation of this project, the words "downstream" and "upstream" are used when talking about certain pieces of functionality and/or templates. In this context, "upstream" is used to reference functionality/templates coming from openapi-generator or other external tools used; "downstream" is used to reference configuration and modifications that you will be providing as inputs to apigentools.

Quickstart

Note: This quickstart opts for the easiest approach possible. To access advanced features of apigentools, refer to the rest of this documentation.

Installation

Install apigentools with

pip install apigentools

Containerized Version

For all operations (except the simple apigentools init which doesn't need any complex dependencies), we highly recommend everyone to use the containerized version of apigentools available via apigentools/apigentools image on Dockerhub. The apigentools Python package includes a container-apigentool script which utilizes the image in the background. It can be used as easily as the non-containerized apigentools script, while also ensuring presence of all tools and dependencies necessary for apigentools to work properly.

Setup Spec Repo

A spec repo is a repository that holds the OpenAPI spec of your API, configuration for both apigentools and openapi-generator, and some additional files, such as downstream templates and template patches. Apigentools work with this repo to perform various tasks, such as validating your spec or generating client code. To get set up, run:

apigentools init myspecrepo
cd myspecrepo

This creates an apigentools supported directory, myspecrepo, where you can start defining your API spec.

Writing the OpenAPI Specification

One of the apigentools core ideas is being able to generate clients that can access multiple major versions of your API (e.g. v1, v2, ...). This means it also expects you to have a separate spec for each of the major versions. The init command, by default, generates a directory structure holding spec/v1 directories for v1 of your API.

Since OpenAPI specification files can grow very large, apigentools allow you to split them into multiple "sections". Initially, only two files are created. These are also files that are expected to exist for each major version of your API:

• spec/v1/header.yaml - A header file that should only contain the OpenAPI keys openapi, info, externalDocs and servers (note that apigentools can fill in servers dynamically based on your config/config.yaml).
• spec/v1/shared.yaml - A file containing any OpenAPI objects that are referenced from more than one "section"—components (which includes schemas and security_schemes), security, and tags.

Add a file spec/v1/users.yaml with the following content:

paths:
  /api/v1/users:
    get:
      description: Get all registered users
      summary: Get all users
      operationId: GetAllUsers
      responses:
        '200':
          description: OK
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Users'
        '400':
          description: Bad Request
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Error400'
components:
  schemas:
    Users:
      type: object
      properties:
        users:
          type: array
          items:
            $ref: '#/components/schemas/User'
    User:
      type: object
      properties:
        name:
          type: string
        email:
          type: string
          format: email

Note that you didn't add the Error400 entity in the file. Assume that this would be a generic error response returned by all API endpoints and put it in the spec/v1/shared.yaml file:

components:
  securitySchemes: {}
  schemas:
    Error400:
      type: object
      properties:
        errors:
          type: array
          items:
            type: string
tags: []
security: []

Now you just need to add the users.yaml spec section to config/config.json. Add it to spec_sections.v1 like this:

{
    "codegen_exec": "openapi-generator",
    "languages": {},
    "spec_sections": {
        "v1": ["users.yaml"]
    },
    "spec_versions": [
        "v1"
    ]
}

Now that you have a complete spec with a single API endpoint and all schemas defined for it, you can start using apigentools to process it.

Validating the OpenAPI Spec

Run:

container-apigentools validate

If everything else is in place, this command exits successfully. If not, it prints a meaningful error message telling you where the problem is. It will also create a spec/v1/full_spec.yaml, which is a single file containing the whole OpenAPI spec definition of your API.

Add a Language Configuration

Configure apigentools to actually generate code for one language. Try doing this for Go. First, add it to config/config.json:

{
    "codegen_exec": "openapi-generator",
    "languages": {
        "go": {
            "github_repo_name": "my-api-client-go",
            "github_org_name": "myorg",
            "spec_versions": ["v1"],
            "version_path_template": "myapi_{{spec_version}}"
        }
    },
    "spec_sections": {
        "v1": ["users.yaml"]
    },
    "spec_versions": [
        "v1"
    ]
}

This will make sure that Go client will be generated with the following aspects:

• It will be generated inside generated/my-api-client-go directory.
• Only the v1 client will be generated. If you ever add another version of this API, you'll need to also explicitly add it to its spec_versions.
• The code for v1 of the API will reside under generated/my-api-client-go/myapi_v1. See config.json format for information on how the version_path_template works.

Next, add an openapi-generator configuration file for Go, called config/languages/go_v1.json:

{
    "gitUserId": "myorg",
    "gitRepoId": "my-api-client-go",
    "isGoSubmodule": true,
    "packageName": "myapi_v1",
    "packageVersion": "0.0.1",
    "withGoCodegenComment": true
}

You're finished configuring apigentools, and you can now start generating code!

Generating Client Code

To generate client code using the configuration you've created, run:

container-apigentools generate --builtin-templates

Note that you're using the --builtin-templates argument here to tell apigentools to use templates that are built in to openapi-generator.

If using template patches, the containerized version of apigentools preprocesses the templates and runs the code generation as part of container-apigentools generate. When using non-containerized version, template preprocessing has to be done prior to code generation using the templates command.

You can now browse the generated code under generated/my-api-client-go/myapi_v1. Note that if you want to make this a proper Go module with per-major-API-version submodules, you need to also add top-level go.mod and go.sum files. Place these in generated/my-api-client-go. If you want to continue working on this example, see the downstream templates documentation for instructions.

Also note that not all language templates included in openapi-generator are compatible with apigentools' per-major-API-version submodules approach. If you find a language for which this is not possible, consider contributing this as a feature to openapi-generator to make the project better for everyone!