api_agent

Technology-agnostic api bindings for your fullstack Dart application.

  1. api_agent bridges the service gap between your Dart frontend and backend by generating type-safe client and server bindings for your api from a single source definition. It automatically keeps your api definitions in sync between client and server.

  2. This package is only concerned about the structure of your api, not the technology or protocols used. Therefore it is completely technology- as well as protocol-agnostic and you have full control over the data transportation layer. It provides some ready to use implementations for common protocols like HTTP, or you can implement your own to support any protocol (e.g. Websockets, MQTT, GRPC, GraphQL, or any proprietary protocol).

Usage Overview

In a shared location, defining your api is as easy as defining an abstract class.

@ApiDefinition()
abstract class MyApi {
  Future<MyResult> myApiEndpoint();
}

On the client, accessing your api is as easy as calling a method.

void main() async {
  var client = MyApiClient( // generated from your api definition
    HttpApiClient( // ready-to-use http client
      domain: "http://my.domain.com", // http-client specific options
    ),
  );
  
  var result = await client.myApiEndpoint(); // generated method from your api definition
}

On the server, implementing your api is as easy as implementing an abstract method.

void main() async {
  var server = await serve( // from package:shelf
    ShelfApiRouter([ // ready-to-use shelf router
      MyApiEndpointImpl(),
    ]),
    InternetAddress.anyIPv4, port
  );
}

class MyApiEndpointImpl extends MyApiEndpoint { // generated from your api definition
  @override
  Future<MyResult> myApiEndpoint(ApiRequest request) {
    // your implementation here
  }
}

For a complete implementation have a look at the example.

Setup & Getting Started

For a fullstack Dart application using api_agent you need the following setup:

Start by creating a new workspace folder:

mkdir my_app && cd my_app

First create a shared dart project, which is going to contain all code used by both the frontend and backend. In our case, we want to put the api definitions in this project.

dart create my_app_shared
cd my_app_shared

This will create a dart project with a simple dart app. Since we want to use this as a shared library, we don't need the generated bin directory, but rather a lib with our definitions.api.dart inside.

rm -r bin
mkdir lib && touch lib/definitions.api.dart

We also need to add api_agent as a dependency together with build_runneras a dev dependency.

dart pub add api_agent
dart pub add build_runner --dev

Next we add our api definitions to the created file:

import 'package:api_agent/api_agent.dart';

@ApiDefinition()
abstract class GreetApi {
  Future<String> greet(String name);
}

Finally in the shared project we need to run build_runner to generate all the necessary api bindings for the frontend and backend. This will generate definitions.client.dart and definitions.server.dart to be used by the respective platform.

dart run build_runner build

Secondly we create our frontend project. This can of course be anything from a command-line app to a flutter app or website written in Dart. For simplicity we will create a simple command-line app, but the usage is the same for all client apps.

cd .. # make sure we are in our workspace directory
dart create my_app_frontend

We again need to add api_agent as a dependency. Additionally we need to add our my_app_shared project as a local path dependency.

cd my_app_frontend
dart pub add api_agent
dart pub add my_app_shared --path=../my_app_shared

Now open the generated bin/my_app_frontend.dart and change the content to the following. This will import our generated api bindings from the shared package and setup a basic http client with it.

import 'package:api_agent/clients/http_client.dart';
import 'package:my_app_shared/definitions.client.dart';

void main(List<String> args) async {
   var client = GreetApiClient( 
      HttpApiClient(domain: "http://localhost:8080"),
   );

   var result = await client.greet(args[0]);
   print(result);
}

Lastly we setup our backend project. This can again have any structure or use any package, but for simplicity we will go with the standard package:shelf app.

cd .. # make sure we are in our workspace directory
dart create -t server-shelf my_app_backend

As usual, we add the needed dependencies on api_agent and our my_app_shared project.

cd my_app_backend
dart pub add api_agent
dart pub add my_app_shared --path=../my_app_shared

Next we modify the generated bin/server.dart. We can keep most of the setup and just plug-in our custom router with our api implementation. Therefore replace only lines 7 to 19 with the following:

import 'package:api_agent/servers/shelf_router.dart';
import 'package:my_app_shared/definitions.server.dart';

final _router = ShelfApiRouter([GreetApiImpl()]);

class GreetApiImpl extends GreetApiEndpoint {
  @override
  String greet(String name, ApiRequest _) {
    return 'Hello $name.';
  }
}

This is all to get a working fullstack Dart app with api_agent.

To test you app, start both the server and client in two separate terminal windows:

# Terminal 1: run server
cd my_app_backend && dart run bin/server.dart
# Terminal 2: run client
cd my_app_frontend && dart run my_app_frontend James

While it is not very scalable, you could also do everything in a single dart project. Have a look at examples/single_project for an example of this.

Usage

As demonstrated in the Setup there are three main parts to a fullstack app using api_agent.

  • The Api Definitions in a shared project
  • The Api Client in a frontend project
  • The Api Endpoints in a backend project

Api Definitions

You define an api definition by using the @ApiDefinition() annotation on an abstract class. The abstract methods in this class define the endpoints the api supports. A method can have any amount of required or optional parameters and must return a Future of any type.

@ApiDefinition()
abstract class GreetApi {
  Future<String> greet(String name);
  
  Future<MyResult> compute(MyData data, {bool? someOptionalParam});
}

An ApiDefinition can also have any amount of nested child ApiDefinitions by defining an abstract getter with any name. The return type must be another ApiDefinition.

@ApiDefinition()
abstract class MainApi {
  UserApi get users;
}


@ApiDefinition()
abstract class UserApi {
  Future<List<User>> list();
}

How nested ApiDefinitions are handled is up to the protocol implementation. For example the http implementation constructs the request url from the nested api definitions and target method. In the above case, calling 'MainApiClient().users.list()' would result in a url path of /users/list;

Api Codec

All protocols will at sometime need to serialize and deserialize your parameters and result objects. Protocol implementations should define some default codec for this (e.g. JsonCodec), but when you want to use any custom types or non-primitive types for your endpoints, you need to define your own codec.

You can do this by implementing the ApiCodec interface and providing your custom codec to the @ApiDefinition() annotation.

@ApiDefinition(codec: GreetCodec())
abstract class GreetApi {
  // GreetResult will be en-/decoded by the provided GreetCodec
  Future<GreetResult> greet(String name);
}

class GreetCodec implements ApiCodec {
  @override 
  dynamic encode(dynamic value) {
    // encode value
  }
  
  @override
  T decode<T>(dynamic value) {
    // decode value
  }
}

Generic Endpoints

Through the magic of type_plus, this package allows endpoints to be generic functions. The type parameter can be used in any way, for both the return type and parameter types.

Simply specify an endpoint with up to 5 generic type parameters like this:

@ApiDefinition()
abstract class UserApi {
  Future<T> someGenericEndpoint<T, U>(U myParam);
}

By default, it is only supported to call a generic endpoint with standard Dart types, like bool int, String, List, Map and others. If you want to use custom types or classes as generic type arguments, you need to specify these types in the annotation using @ApiDefinition(useTypes: [MyCustomClass, SomeOtherType]).

Api Clients

api_agent generates api clients for each of your api definitions. To use a client, instantiate it with a protocol client as its only parameter:

var client = GreetApiClient( // generated from your api definition
   HttpApiClient(), // specific protocol client
);

After that you can call the endpoints that you defined in your api definition as normal methods.

var result = await client.greet();

api_agent already comes with the following clients:

HttpApiClient

This will use the http package to make requests to your api.

  • Will use GET when a method is prefixed with get (e.g. getUsers), otherwise POST
  • With GET requests, parameters are encoded as query parameters
  • With POST requests, parameters are sent as json payload
  • The url is constructed from the class name(s) and method name in the following way:
    • Api suffixes on class names are ignored in the url (GreetApi -> /greet).
    • get prefixes on method names are ignored in the url (getUsers() -> /users).
    • Method and class names are transformed into snake case (myMethodName() -> /my_method_name)

The HttpApiClient constructor accepts

  • a domain, where your api is hosted,
  • an optional path, where your api is mounted, defaults to /,
  • an optional list of ApiProviders,
  • an optional ApiCodec to use as a fallback if none is defined by the api definition

Api Providers

ApiProviders can be used to intersect and modify a request before it is sent. A common use-case would be to add an authentication header to the request:

class AuthProvider extends ApiProvider<HttpApiRequest> {
  @override
  FutureOr<HttpApiRequest> apply(HttpApiRequest request) {
    return request.change(headers: {'Authorization': 'Bearer my-bearer-token'});
  }
}

Writing custom Api Clients

To define a custom client that uses your chosen protocol, simply implement the ApiClient interface:


class MyProtocolApiClient implements ApiClient {
   @override
   ApiClient mount(String prefix, [ApiCodec? codec]) {
     
   }
   
   @override
   Future<T> request<T>(String endpoint, Map<String, dynamic> params) {

   }
}

The mount() method should just return a new instance of your api client respecting the given prefix and codec.

The request() method should do the actual work of calling the api though the chosen protocol and technology. It gets the name of the requested endpoint (the name of the called client method) and a map of the provided parameters. A standard implementation would:

  1. Construct an instance of ApiRequest, possibly a protocol-specific subclass like HttpApiRequest
  2. Apply any given ApiProviders through request.apply(providers)
  3. Execute the request (protocol specific implementation)
  4. Return the result, possibly decoded using codec.decode<T>(result)

It is generally a great starting point to look at existing implementation of api clients, like the included HttpApiClient.

Api Endpoints

api_agent generates ApiEndpoints for each of your ApiDefinitions. ApiEndpoints need to be implemented on the server and the passed to an ApiBuilder that e.g. spins up a http server and passes incoming requests to your endpoints.

To implement your endpoints you will need to do the following:

First our api definition from above will generate a GreetEndpoint that expects an implementation in form of FutureOr<String> Function(String name, ApiRequest request).

There are multiple ways to implement an endpoint.

  1. Passing a handler function
var greetEndpoint = GreetEndpoint.from((name, r) {
  return 'Hello $name.';
});
  1. Extending the abstract class
var greetEndpoint = GreetEndpointImpl();

class GreetEndpointImpl extends GreetEndpoint {
  @override
  FutureOr<String> greet(String name, ApiRequest request) {
    return 'Hello $name.';
  }
}

You can freely choose between these two options on a case-by-case basis.

Next api_agent will also generate an GreetApiEndpoint that expects a GreetEndpoint as its child. You can again choose to provide the endpoint directly, or implement the abstract class:

  1. Passing a child endpoint
var greetApi = GreetApiEndpoint.from(
  greet: greetEndpoint,
);
  1. Extending the abstract class
var greetApi = GreetApiEndpointImpl();

class GreetApiEndpointImpl extends GreetApiEndpoint {
  @override
  FutureOr<String> greet(String name, ApiRequest request) {
    return 'Hello $name.';
  }
}

This might seem ambiguous, but it gives you the freedom to choose the right level of cohesion. With a small simple api you might not want to create a custom class for every endpoint. But with a larger and more complex api you might find this more appropriate to separate your logic.

Api Middleware

In more complex apis with nested inner apis, your endpoints will form a tree-like structure:

var myApi = MyApiEndpoint.from(
  users: UsersEndpoint.from(
    list: ...,
    getById: ...,
  ),
  publications: PublicationsEndpoint.from(
    articles: ArticlesEndpoint.from(
      list: ...,
      ...
    ),
  ),
)

You can add an ApiMiddleware anywhere in this tree to monitor or guard requests to the endpoints in the chosen subtree. A typical use-case is to authenticate the requesting user.

Implement a custom ApiMiddleware by implementing the ApiMiddleware interface:

class AuthMiddleware implements ApiMiddleware {
  @override
  FutureOr<dynamic> apply(covariant ShelfApiRequest request, EndpointHandler next) {
    String? token = request.headers['Authorization'];
    if (validateToken(token)) {
      return next(request);
    } else {
      throw ApiException(401, 'Authentication token is invalid.');
    }
  }
}

Then, insert your middleware using the ApplyMiddleware endpoint:

var myApi = MyApiEndpoint.from(
  users: ApplyMiddleware(
    middleware: AuthMiddleware(),
    child: UserEndpoint.from( // middleware applied to all child endpoints
      ...
    ),
  ),
  publications: PublicationsEndpoint.from( // middleware not applied to endpoints
    ... 
  ),
);

Api Builders

An ApiBuilder consumes your api endpoints and constructs the communication channel to which the client can connect to. api_agent already comes with the following builders:

ShelfApiRouter

This will use the shelf_router package to construct a router that can be use with the shelf package and e.g. passed to serve.

It is intended to be used with the HttpApiClient and has the same request rules.

To use it simple pass your api endpoint to ShelfApiRouters constructor and use it as a shelf handler:

void main() {
  var router = ShelfApiRouter([greetApi]);
  
  serve(router, InternetAddress.anyIPv4, port: 8080);
}

Writing custom Api Builders

To define a custom api builder that uses your chosen protocol, simply implement the ApiBuilder interface:

class MyProtocolApiBuilder implements ApiBuilder {
  @override
  void mount(String prefix, List<ApiEndpoint> children) {
    // mounts [children] under [prefix]
    // should call [ApiEndpoint.build()] for each child with
    // a new instance of your api builder
  }
  
  @override
  void handle(String endpoint, EndpointHandler handler) {
    // registers the [handler] for this [endpoint] 
  }
}

After that you should call the build(ApiBuilder builder) method of your root ApiEndpoint(s) and provide an instance of your custom ApiBuilder. These will then subsequently call mount() or handle() to register themselves with your api.

I would recommend looking at existing implementation of api builders, like the included ShelfApiRouter.

Libraries

api_agent
builder
client
clients/http_client
server
servers/shelf_router