• By Martijn Faassen
  •  • 
  • 2011-11-02
  •  • 
  • Tags: 
  • obviel
  • rest


This document is an introduction to Obviel from the perspective of how it was developed. For those impatient with history, you can just go to http://www.obviel.org and read the documentation there, however.


A few years ago I worked on a dynamic web user interface on a project for Gocept, using JavaScript of course. There I invented a somewhat unusual approach to do:

  • use JSON Template to do client-side templating
  • the server would return a JSON object and the client side template would then be used to render it.

Callback Complexity

In the fall of 2010, I working on a few related customer projects to create complex web applications that needed a good UI. As they were applications, not web sites, I decided to create those UIs using a single-page approach. The server would just present a single web page with JavaScript to build and update the UI.

We decided to use jQuery for those projects. jQuery is very popular, with a vibrant community with a lot of components choose from. It would mean we would have to do more integration than we would need if we chose a more integrated framework such as YUI, which I also have experience with, but I was a bit tired of big frameworks. And jQuery's terse and powerful way to express things is attractive. And you just have to make a choice to see what happens at some point, right?

The JavaScript code in those applications was quickly growing out of control. We were trying to do proper REST: the client would just start with a single URL, which would resolve to a JSON object that would have links to other JSON objects on the server, which would in turn have links to other JSON objects. Proper REST is nice: you can decouple your client from the server more, because your client code doesn't depend on the specific URL structure of your server -- instead it just follows links.

But doing proper REST with JavaScript without taking special measures tends to result in endlessly nested callbacks in callbacks in callbacks. The code was too tightly coupled and hard to follow.

This made me think about the stuff I did at Gocept again. I felt there was quite a bit of potential in those patterns, potential I hadn't explored fully yet. So I added it to one of our projects and started to rewrite the UI to use it, to try to clean things up.


At this stage more people started to be involved. Ideas tend to get better with good input, and with usage by a larger group of people. I will forget some of the people involved in this project, for which I apologize, but I'm going to drop some names now.

The original idea was born at Gocept. I was doing the work for Christian Zagrodnick there, and I had some good conversations with him about it then. He also used the early code in a few of his own projects later on. Just as I had checked in the view stuff into the new project I was doing I found myself in Germany at Gocept again, and had further productive conversations with him about it. Thanks Zagy!

Going to Germany, where Gocept is based, from where I live, the Netherlands, means a train ride, and in the train I had the opportunity to bounce ideas off Jan-Wijbrand Kolman (and him ideas off me), something I always enjoy tremendously. The biggest thing that came out of that was Fanstatic, but that's the topic for another discussion. It also helped me think through what later became Obviel. Thanks JW!

The idea developed around this time was that the JSON returned by the server to JavaScript should have a type marker to describe what kind of JSON we are dealing with then. In Obviel we call this type marker an iface, short for interface. On the client, we could then hook up views to ifaces. This way if the client is confronted with some unknown JSON, it would know what to do with it by looking up the appropriate view for that iface. This helps with REST and therefore with loose coupling: the client code not only doesn't need to know what resources the server is presenting, but also would be able to know what resources are on what URLs; the resources would disclose this themselves. In this, "JavaScript views" (as we were calling it at the time) resemble dynamic view lookup like practiced in the Zope Component Architecture (don't you all run away now, though!).


New approaches get better when you try to actually use them in practice. You learn a lot. The customer projects I was doing were too big to do on my own, I was working with Izhar Firdaus on them. Izhar gave me a lot of invaluable feedback and improvements and came up with new patterns of using these "JavaScript views" (as we were calling them) throughout the projects that use Obviel; I had great pleasure working with him for over a year.

I think I already rewrote the system halfway back then, and I remember Izhar doing something similar. Eventually after some experience I sat down and started to write a document describing how I wanted this JavaScript views thing to work. So I got Guido Wesdorp involved. This meant Izhar and I were freed up to build more application specific code and gain experience with it.

Working with Guido in this way was a new experience: I had designed a framework but Guido was implementing it. Guido focused on making the code solid. I, Guido and Izhar together came up with various new approaches.


The projects we were working on needed a number of complex web forms. I have a lot of experience with server-side form libraries. Some time before I had been in a rich-client project with Jasper Op de Coul and he suggested creating a JavaScript form library instead. This would offer some benefits, such as inline validation for everything by default, and web-framework neutrality. While we didn't implement the idea at the time, it did stick in my mind.

Here we needed lots of complex forms again, and server-side validation with errors didn't really fit well in a rich client UI. So I sketched up a way to do JavaScript forms using a new technology jQuery datalink. The concept of Obviel forms is to maintain a JavaScript object with the form contents in parallel to the HTML form (an earlier conversion with Antonin Amand helped me develop this concept). When the user modifies the form, the JavaScript object is automatically updated. A form submission then simply submits that JavaScript data object as JSON to the server.

I had Guido go ahead and start building the form library for us.


"JavaScript views" had already gone through a few incarnations. I'd rewritten it at least once, Izhar had given it another shot, and Guido had rewritten it again.

Meanwhile Sylvain Viollon at Infrae had started using Obviel as well. He had decided to do its own rewrite, to my chagrin at the time, as now we had two parallel versions. But it turned out for the good: he had introduced some new ideas into his version that were very useful: when a view was rendered, it would be bound to the object it had just rendered, and the element it was rendered on. These bound views would allow a better structuring of view code than passing this information along as parameters all the time.

We kept learning however, and after a while, when Guido had already moved on from the project, several things in Obviel forms in particular started to be frustrating.

So I decided to rewrite Obviel forms cleaning it up. Among new things introduced were composite sub-forms and a generic repeat widget to allow the user to enter repeating sub-forms as well, and a way to maintain the inline error messages as a datalinked object as well. Govert Buijs, who had started to work on the projects with us, also helped work out advanced form validation scenarios.

On a roll, I decided to rewrite the Obviel core as well, introducing Sylvain's ideas about bound views. Along the way I also started using the new jQuery deferreds in its implementation (Sylvain then turned out to have done so in his version as well, in parallel).


Now it is time to attract more people to Obviel. It's been battle tested in a number of big projects.

To attract people to a project you not only need good code. We do have good code I think: it's unit tested and coverage tested. But you also need great documentation. Along the way I and Guido had written bits of documentation about Obviel. It took me months of stops and start to get this documentation in the state it is now, so that newcomers to Obviel aren't lost.

So, if you are curious about Obviel and want to find out more, go to http://www.obviel.org

And give us feedback, please!