Skip to content

Best Practices for Developing Conduit Applications

Keep Dart Projects Separate

Because Dart is cross-platform, developers should avoid combining client application projects with Conduit projects. Instead, use a single repository with an independent project for each facet of the system. When there are opportunities for code sharing between platforms (typically between Flutter and AngularDart), shared code can live in a dependency project in the same repository.

A typical directory structure for an multi-faceted application looks like this:


{% hint style="info" %} "Project Definition" A project is a directory that contain a pubspec.yamlfile and lib directory. {% endhint %}

It is tempting to share your data model types between your server and client applications, but this falls apart for anything but the most simple of applications. There are enough behavioral differences between the four representations of your data model - in the database, on the server, on the wire (JSON), and on the client - that a single type will have a hard time encompassing. Instead, generate an OpenAPI specification with conduit document and use one of the many open-source tools for generating client data model types.

Split discrete components into separate packages

It can often be useful to split discrete code pieces into separate packages with their own sub-repo.

Non-trivial Flutter widgets are a good example. Having the code in a separate package facilitate allocating a single owner to the package. Packages with a single owner have much less management overhead and cleaner commit logs. It also stops developers treading on each others toes.

Remember a component isn't just a flutter widget. It can be a collection of classes/functions that provide a specific service, this could be something like permission management or formating/parsing monetary amounts. Any collection of code that together delivers a singular 'concern' is a good candidate.

Use Test Driven Development (or something close to it)

In Conduit, testing is a first-class citizen. The conduit_test package has classes and methods for initializing and running an application for testing, making requests to that application, and verifying the responses. There is value to using tools like Postman or CURL to test proof of concept code, but the conduit_test package is geared specifically for replacing these tools while retaining automated tests as the project grows.

An example test suite looks like this:

void main() {
  final harness = new Harness()..install();

  test("GET /endpoint returns 200 and a simple object", () async {
    final response = await harness.agent.get("/endpoint");
    expectResponse(response, 200, body: {"key": "value"});

Use a bin Script to Verify Assumptions

Keep a simple Dart script file in the bin/ directory that imports your project. Use this file as a scratchpad to test exploratory code before committing to a test suite. Don't check this file into source control.

import 'package:myapp/myapp.dart';

Future main() async {
  var whatIsThis = await someYetToBeNamedUsefullyMethod();

Create New Projects from a Template

Use conduit create to create applications with the appropriate structure and boilerplate. There are templates for different kinds of applications; view these templates with conduit create list-templates.

Use a Debugger

A debugger allows you to stop execution of a running application at a particular line of code to verify variable values, and then continue to step through that code line by line. It can be used when running test suites or when running the application through the bin/main.dart script.

In IntelliJ IDEA, right-click on any file with a main function (which includes test suites) and select Debug option. Use breakpoints (by clicking on the gutter area to the left of the text editing area) to stop execution at a particular line before it is executed.

Use the Suggested Project Directory Structure

See Conduit Project Structure.

Pass Services to Controllers in entryPoint

Pass service objects to controllers in entryPoint and only pass the services the controller will use.

class AppChannel extends ApplicationChannel {
  GitHub githubService;
  PostgreSQLConnection databaseConnection;

  Future prepare() async {
    databaseConnection = new PostgreSQLConnection();
    githubService = new GitHub();

  Controller get entryPoint {
    final router = new Router();

      .link(() => new DBController(databaseConnection));

      .link(() => new GitHubController(githubService));

    return router;

Passing references like this allows for injecting dependencies that depend on the environment; e.g. in production, development or during tests. It also avoids tight coupling between the objects in your application.

Minimize the access a controller has to its dependencies; e.g. don't pass it a StreamController when it only needs Sink or a Stream.

Use a Test Harness

A test harness initializes your application in a test suite. It has built in behavior that you can add to for things that are specific to your application. Documentation for using a test harness in your application is located here.

Use config.src.yaml

Use the convention of config.src.yaml file to prevent configuration errors and inject test dependencies.

Understand how Conduit Uses Isolates

See more in Application Structure.

Use ResourceController Subclasses

Subclassing ResourceController provides significant conveniences, safeties and behaviors used by the majority of an application's request handling logic. Prefer to use this class for non-middleware controllers.

Keep ApplicationChannel Tidy

A ApplicationChannel should handle initialization, routing and nothing more. Consider moving non-initialization behavior into a service object in a separate file.

Avoid Raw SQL Queries

Prefer to use the Conduit ORM. It sends appropriate HTTP responses for different kinds of errors, validates input data and is ensures the queries match up with your data model.

Use API Reference

Conduit is an object oriented framework - behaviors are implemented by instances of some type. The types of objects, their properties and their behaviors all follow similar naming conventions to make the API more discoverable.

Many types in Conduit have a prefix in common with related types. For example, types like AuthServer, AuthServerDelegate and AuthCode are all related because they deal with authentication and authorization. Methods are named consistently across classes (e.g, asMap is a common method name).

When looking for a solution, look at the API reference for the objects you have access to. These objects may already have the behavior you wish to implement or have a reference to an object with that behavior.

Use try-catch Sparingly

All request handling code is wrapped in a try-catch block that will interpret exceptions and errors and return meaningful HTTP responses. Unknown exceptions will yield a 500 Server Error response. In general, you do not need to use try-catch unless you want a different HTTP response than the one being returned for the exception.

Code that throws an exception during initialization should not be caught if the error is fatal to the application launching successfully.