26 Nov 2014 benad   » (Apprentice)

Modern MVC in the Browser

A few months ago, I started learning the AngularJS framework for web development. It is a quite clever framework that allows you to "augment" your HTML with markup to map placeholders with elements in your object model. It reminded me of JSP and markup-based SWING frameworks I used a decade ago. Since then, though, I realized that this template-based approach has quite a few limitations that bother me. First, as it is based on templates, it doesn't handle well highly dynamic elements that generate markup based on context. Second, complex, many-to-many mappings between the view and the model are difficult to express. For example, it is difficult to express a value placeholder that is the sum of multiple elements in the models without having to resort calling a custom function, or having the change in the view trigger the corresponding change in multiple locations in the model without, again, resorting to a custom function. Third, it is a framework that needs to keep track of the model/view mapping, so it is all-encompassing and heavy in configuration.

So, let's break down the problem and look individually at the view, model and controller as separate modules, and see if there could be a better, more "modern" approach.

The biggest problem with the "model" in JavaScript is that it lacks safely hiding properties as functions, as done natively in C# for example, or manually through "getter" and "setter" functions in JavaBeans. Because of that, it becomes difficult to take any normal object and add a layer on top of it that would implement an observer pattern automatically. Instead, you can use something like the model implementation of Backbone.js that fully hides the model behind a get/set interface that can automatically trigger change events to listener objects. It may not be elegant, but it works well.

For the "view", HTML development doesn't support well the concept of a "custom control" or "custom widget" in other GUIs. The view is mostly static, and JavaScript can change the DOM to some extent, and at high cost. In contrast, other GUI systems are based around rendering controls ("GUI controls", not to be confused with the controller in MVC) on a canvas, and the compositing engine takes care of rendering on screen only the visible and changed elements. A major advantage of that approach is that each control can render in any way it pleases without having to be aware of the lower-level rendering engine, be it pixels, vectors or HTML. You could simulate a full refresh of the DOM each time something (model or controller) changes the GUI in HTML, but the performance would be abysmal. Hence, the React library from Facebook and Instagram, which supports render-based custom controls but using a "virtual DOM", akin to "bitmasks" in traditional GUIs, so that only effective DOM changes are applied.

Finally, for the "controller", the biggest issue is how to have both the model and the view update each other automatically, on top of the usual business logic in the controller, without creating cyclic loops. React's approach, named Flux, is a design pattern where you always let events flow from the model to the view, and never (directly) in the other direction. My gripe with that is that it is merely a design pattern that cannot be enforced. This reminds me of the dangers of using the WPF threading model as a means to avoid concurrency issues: Forget to use the event dispatcher a single time, and you will create highly difficult to debug crashes. A new approach, called functional reactive programming is kind of like what dependency injection did for module integration, but for events. Essentially it is a functional way of manipulating channels of events between producers and consumers outside of the code of each producer and consumer. This may sound quite an overhead compared to the inline and prescriptive approach of Flux, but as soon as you build a web page heavy on asynchronous callbacks and events coming from outside the page, having all of that "glue" in a single location is a great benefit. An implementation of FRP for JavaScript, Bacon.js, has a great example of how FRP greatly reduces those countless nested callbacks that are commonplace in web and Node development.

Combined, Backbone.js, React and Bacon.js offer a compelling alternative to control- and template-heavy browser MVC frameworks, and at minimum prevent you from being "locked-in" a complex and difficult to replace framework.

Syndicated 2014-11-26 01:44:49 from Benad's Blog

Latest blog entries     Older blog entries

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!