Client-Side JavaScript

Perhaps nothing in Tapestry has changed over the years so much as the way client-side JavaScript is supported. From the get go, the goal was to make JavaScript a first-class citizen in the Tapestry world, and make it easy to encapsulate JavaScript within components.

The legacy JavaScript page discusses the earlier approaches; the main feature of Tapestry 5.4 is a total rewrite of all things client-side, with the following goals:

  • Break the hard linkage of Tapestry to Prototype and Scriptaculous, by introducing an abstraction layer
  • Remove the clumsy Tapestry and T5 "namespaces"
  • Reduce the amount of page-specific JavaScript initialization
  • Make it easier to override behavior associated with client elements
  • Support CoffeeScript and (potentially) other languages that target JavaScript
  • Organize client-side JavaScript using modules
  • Make pages load faster
  • Integrate Bootstrap
  • Make it easier for rich client libraries such as Backbone or AngularJS to operate within a page
  • Properly document Tapestry's client support

The Overall Vision

The overall vision for the client-side in Tapestry is encapsulation, at several different levels.

On the server-side, a Tapestry component (or mixin) exposes configurable parameters. The component writes DOM elements or attributes, as well as some amount of JavaScript initialization. The encapsulation here allows developers with little or no knowledge of client-side JavaScript to enjoy the benefits (as consumers of components created by developers who are versed in client-side coding and Tapestry components).

On the client-side, the JavaScript combines with the special markup to produce the behaviors that are desired ... anything from controlling the initial focus field, to performing client-side input field validation, to running complex Ajax workflows.

Where possible, all of this behavior is driven by data- attributes on the elements, combined with top-level event handlers. On the client side, events are used not only to respond directly to user actions (with "click", "mouseOver", "submit", or other event listeners) but also to allow elements to collaborate in various ways.  For example, input validation is based on triggering a specific custom event on each form control element, and top-level event handlers can then manage the validation for any number of fields.

Prototype vs. jQuery

For several years, it has been obvious that Tapestry "backed the wrong horse" with respect to Prototype and jQuery. When the first code was being laid down in 2007 or 2008, it wasn't so clear that jQuery with its odd abstractions and unfamiliar approach, would go on to conquer the world. Meanwhile, Prototype was very strongly integrated into Ruby on Rails and had first class documentation and books.

That being said, jQuery is not the be-all and end-all either. Tapestry 5.4 introduces an abstraction layer, that allows many components to write code that doesn't care whether the foundation framework is Prototype or jQuery or something else. If you like jQuery then there's no problem: write your application using just jQuery and you can ignore a lot of the features in the abstraction layer. Your code will likely be just a bit more efficient.

If you are building a reusable component or library, writing to the abstraction layer may be worth the effort; it is entirely possible that someone may write a replacement for the abstraction layer that targets your favorite foundation framework, such as ExtJS, MooTools, or something not even known of today.

Heavy vs. Light

Earlier Tapestry JavaScript was heavy. Essentially, each component would write some very specific JavaScript initialization that would include the component's DOM id and many other details. This initialization would reference a function on the T5.inits namespace.

The function there would typically locate the specific element by its client DOM id, then attach event handlers to the one element. It might also create some form of client-side controller object. There were issues due to this: for complex pages (or perhaps even typical pages), the "blob" of JavaScript initialization at the bottom of the page could be quite large.

The use of individual event handlers meant that Tapestry applications would use more client-side objects that a bespoke jQuery solution ... because the normal approach in jQuery is to attach a single event handler to the document or body that handles any events that bubble up to the top and match a CSS selector.

In Tapestry 5.4, the goal is to make things light. In most cases, there isn't a specific initialization function; instead a JavaScript module is loaded, and it installs one or more top-level event handlers; the elements has data- attributes that are used by those top level handlers to recognize which elements they are responsible for.

This is more of a full lifecycle approach; adding or removing page content (such as with a Zone component) is both cheaper and less error prone using top-level event handlers than per-element event handlers; there's also less of a chance of memory leaks under Internet Explorer.

Internet Explorer is pretty well known for memory leaks; its DOM and JavaScript run in different kinds of memory, which are garbage collected individually. This means that a reference from JavaScript to a DOM element will keep the DOM element live, even if that's the only reference to the DOM element anywhere. Meanwhile, event handler JavaScript functions are kept live from the DOM element, making a cycle that can't be broken. Libraries like Prototype and jQuery have to expend some effort to break this link by unregistering event handlers from DOM elements when removing them from the DOM.

A specific example of this approach is how client-side validation now works; in the past, there was a complex system of classes and event listeners that were specific to each individual field. Field controllers had to register with Form controllers. Validators had to register with Field controllers.

Under 5.4, there are a number of data- attributes that can be attached to any DOM element. A form searches for elements with a non-blank value for data-validation; each such element has a series of custom events triggered on it. The top-level handlers for those events receive notifications for elements throughout the document.

t5/core/ (partial)
define ["underscore", "./dom", "./events", "./utils", "./messages", "./fields"],
  (_, dom, events, utils, messages) ->

    dom.onDocument events.field.optional, "[data-optionality=required]", (event, memo) ->
      if utils.isBlank memo.value
        memo.error =  (@attr "data-required-message") or "REQUIRED"

    dom.onDocument events.field.validate, "[data-validate-min-length]", (event, memo) ->
      min = parseInt @attr "data-validate-min-length"
      if memo.translated.length < min
        memo.error = (@attr "data-min-length-message") or "TOO SHORT"
        return false

The t5/core/events module defines constants for different custom event name, it's also a handy place to hang documentation about those events.

The t5/core/dom namespace is the abstraction layer.  onDocument is a handy way to attach a top-level event handler.

Fields that are required have the attribute data-optionality=required; the event handler is passed a memo object that includes a value property, the value from the field. This makes it easier to generate an error if the value is blank.  Because the exact error message may be customized or localized, it is provided in the element as well, as the data-required-message attribute. Setting memo.error to a validation error string causes the field to be decorated with the error message and indicates that the form itself is in error and not ready for submission.

A different event is triggered after the optionality check; The memo.translated property is the value translated before validation (for a numeric field, it would be translated from a string to a number, for example). Again, the error property is set, and the return false ensures that the event will stop bubbling to containing elements or event handlers.

What's very useful in this overall approach is that it no longer matters whether the fields were rendered by Tapestry on the server, or rendered locally (perhaps using Backbone or AngularJS) on the client. As long as they have the correct data- attributes, then they can participate in Tapestry's overall form validation and submission cycle, and even leverage the default validation decoration behavior.

The Abstraction Layer

The abstraction layer is defined by the t5/core/dom module. This module currently has two different implementations - one is a wrapper around Prototype, and the other is a wrapper around jQuery.

The resulting abstraction layer is a bit of a hybrid; it mostly looks like jQuery, but events look a bit more like Prototype. It also doesn't have jQuery's concept of operating on a matched set of elements.

The abstraction is both transitional and permanent. It is transitional in that it is about allowing existing sites with a heavy investment in Prototype to continue to operate with Prototype in the mix. It is permanent  in that it is desirable for third party library developers to keep an abstraction layer between Tapestry's client-side code and any underlying framework, so that particular applications can provide their own abstraction layer and operate without breaking built-in components.

Most applications should transition to jQuery and feel free to use jQuery directly.  It is still best to inject module jquery into your own modules (usually as parameter $). 

If you are writing a third-party application and want to maximize re-use, then use the abstraction layer.

It is often easier to use the abstraction layer to respond correctly to custom Tapestry events.