My project over the last two weeks was to create a web app with Go, or at least a framework for one that I could use to bootstrap further web apps in Go.

I had two primary goals with this project: - Create a framework for a Go web app that could scale easily and be organized. - Learn Backbone.js so I could framework my client-side code.

I started with what I knew, basic Go server-side code (like the stuff in goauth2) and basic client-side javascript with jQuery. I decided to use a basic Todo-list application as a simple example. Getting this set up was pretty simple, but I wanted to make myself use an MVC framework, or something like it. I’ve never done a project with an MVC-like framework before, so this was kind of new ground. Using web.go, I was free to organize the code however I liked. The core model (a todo-list) went in the model folder. Views, which were just templates got stuck in the view folder. And controllers, things like ajax request handlers would go in the controller folder, updating the model as necessary. This seemed like a workable framework. You can see what I had at this point with this commit.

But this only accomplished one of my goals. I wanted to learn a client-side framework like Backbone. I know it had built-in server-side persistence and syncing, and I yearned to easily scale this framework to a large application, so I needed a client-side and server-side framework that worked hand-in-hand. After a few Backbone.js tutorials (I recommend this one), I kind-of understood what was going on and jumped in.

I soon realized the Backbone.sync function makes 4 different CRUD calls to the server: POST, GET, PUT, and DELETE. I decided to map the client-side CRUD to a server-side CRUD and made the following Go interface (see controller.go for full code):

// A Controller for a corresponding Backbone Model object.
type Controller interface {
    // Create model object based on parameters
    // attr is a json-formatted string of attributes
    // Return a json-formattable object of all model attributes
    Create(id, attr string) (interface{}, error)
    // Read a model object back
    // ID may be empty string
    // Return a json-formattable object of all model attributes
    Read(id string) (interface{}, error)
    // Update a model object based on parameters.
    // ID is required and will be non-empty
    // attr is a json-formatted string of attributes
    // Return a json-formattable object of updated model attributes
    // If no attributes other than the updated ones changed, it is acceptable to return nil
    Update(id, attr string) (interface{}, error)
    // Delete a model object.
    // ID is required and will be non-empty
    Delete(id string) error

Using this interface, I defined func BindController(svr *web.Server, path string, ctrl Controller) which bound the 4 CRUD methods to their HTTP verb equivalents, taking care of ID handling (for backbone), and JSON serialization. I left the incoming JSON un-deserialized because of Go’s interesting (and sometimes crazy) JSON handling.

In the end (this commit), I was able to rid myself of the controller folder all-together. The total amount of Go-Code that isn’t a test or in the helpers folder is less than 200 lines and most of that is the model itself. This is a framework that can scale. It will never be as fast to write as rails or node, but the performance will beat rails and node at scale.

Upcoming, I’ll go into a post about the speed of Go web servers (I’ll back up the claim I just made), and one about a few of the tricks I used in this todo app. For now, check the source!