Sunday, October 12, 2014

WAI-ARIA support for AutoComplete widget

In this post I would like to discuss the accessibility for an AutoComplete widget. A typically AutoComplete widget provides suggestions while you type into the field. On my current work I implemented an JSF component on basis of Twitter's Typeahead - a flexible JavaScript library that provides a strong foundation for building robust typeaheads. The Typeahead widget has a solid specification in form of pseudocode that details how the UI reacts to events. The Typeahed can show a hint in the corresponsing input field, like the google's search field shows it, highlight matches, deal with custom datasets and precompiled template. Furthermore, the Bloodhound suggestion engine offers prefetching, intelligent caching, fast lookups, and backfilling with remote data.


Despite many features, one big shortcoming of the Typeahead is an insufficient WAI-ARIA support (I would say it was completely missing until now). An AutoComplete widget should be designed to be accessible out of the box to users of screen readers and other assistive tools. I have decided to add a fully WAI-ARIA support, done this taks and sent my pull request to the GitHub. Below is the new "WAI-ARIA aware" markup with an explanaition (not relevant HTML attributes are omitted).
<input class="typeahead tt-hint" aria-hidden="true">

<input class="typeahead tt-input" role="combobox"
    aria-autocomplete="list/both"
    aria-owns="someUniqueID"
    aria-activedescendant="set dynamically to someUniqueID-1, etc."
    aria-expanded="false/true">
  
<span id="someUniqueID" class="tt-dropdown-menu" role="listbox">
    <div class="tt-dataset-somename" role="presentation">
        ...
        <span class="tt-suggestions" role="presentation">
            <div id="someUniqueID-1" class="tt-suggestion" role="option">
                ... single suggestion ...
            </div>
            ...
        </span>
        ...
    </div>
</span>

<span class="tt-status" role="status" aria-live="polite" style="border:0 none; clip:rect(0, 0, 0, 0); height:1px;
      width:1px; margin:-1px; overflow:hidden; padding:0; position:absolute;">
      ... HTML string or a precompiled template ...
</span>
The first input field with the class tt-hint simulates a visual hint (s. the picture above). The hint completes the input query to the matched suggestion visually. The query can be completed to the suggestion (hint) by pressing either right arrow or tab key. The hint is not relevant for the screen readers, hence we can apply the aria-hidden="true" to that field. The hint is ignored by screen readers then. Why is it not important? Because we will force reading the matched suggestion more intelligent by the "status" area with aria-live="polite" (will be explained below).

The next input field is the main element where the user inputs query. It should have a role="combobox". This is a recommended role for an AutoComplete. See the official WAI-ARIA docu for more details. In fact, the docu also shows a rough markup structure of an AutoComplete!

The main input field should have various ARIA states and properties. aria-autocomplete="list" indicates that the input provides autocomplete suggestions in the form of a list as the user types. aria-autocomplete="both" indicates that suggestions are also provided by a hint (additional to a list). The property aria-owns indicates that there is a parent / child relationship between the input field and the list with suggestions. This property should be always set when the DOM hierarchy cannot be used to represent the relationship. Otherwise, screen readers will get a problem to find a list with suggestions. In our case, it points to the ID of the list. The most interesting property is aria-activedescendant. A sightless user navigates through the list via arrow keys. The property aria-activedescendant propagates changes in focus to assistive technology - it is adjusted to reflect the ID attribute of the current child element which has been navigated to. In the picture above, the item "Lawrence of Arabia" is selected (highlighted). aria-activedescendant is set to the ID of this item and screen readers read to blind users "Lawrence of Arabia". Note: the focus stays on the input field, so that you can still edit the input value. I suggest to read more about this property in the Google's Introduction to Web Accessibility.


The property aria-expanded indicates whether the list with suggestions is expanded (true) or collapsed (false). This property will be updated automatically when the list's state changes.

The list with suggestions itself should have a role "listbox". That means, the widget allows the user to select one or more items from a list of choices. role="option" should be applied to individual result item nodes within the list. There is an interesting article "Use "listbox" and "option" roles when constructing AutoComplete lists", which I suggest to read. Not important for the screen readers parts should be marked with role="presentation". This role says "My markup is only for non sightless users". You probably ask, what is about the role="application"? Is it important for us? Not really. I skipped it after reading "Not All ARIA Widgets Deserve role="application"".

The last element in the markup is a span with the role="status" and the property aria-live="polite". What it is good for? You can spice up your widget by letting the user know that autocomplete results are available via a text that gets automatically spoken. The text to be spoken should be added by the widget to an element that is moved outside the viewport. This is the mentioned span element with applied styles. The styles are exactly the same as the jQuery CSS class ui-helper-hidden-accessible, which hides content visually, but leaves it available to assistive technologies. The property aria-live="polite" on the span element means – updates within this element should be announced at the next graceful interval, such as when the user stops typing. Generally, The aria-live property indicates a section within the content that is live and the verbosity in which changes shall be announced. I defined the spoken text for the AutoComplete in my project as an JavaScript template compiled by Handlebars (any other templating engine such as Hogan can be used too).
Handlebars.compile(
    '{{#unless isEmpty}}{{count}} suggestions available.' +
    '{{#if withHint}}Top suggestion {{hint}} can be chosen by right arrow or tab key.' +
    '{{/if}}{{/unless}}')
When user stops to type and suggestions are shown, a screen reader reads the count of available suggestions and the top suggestion. Really nice.

Last but not least is the testing. If you do not already have a screen reader installed, install the Google Chrome extensions ChromeVox and Accessibility Developer Tools. These are good tools for the development. Please watch a short ChromeVox demo and a demo for Accessibility Developer Tools too. Alternative, you can also try a free standalone screen reader NVDA. Simple give the tools a try.

7 comments:

  1. thanks for sharing informative post..Keep updating your blog.

    ReplyDelete
  2. Hi,

    I find this feature very interesting, and I would like to implement it in my JSF-project. Right now I am using primefaces autocomplete, but it has some bugs/flaws wich almost makes it unusable.

    Do you have a little demo-project for this example? I tried to implement different autocomplete-widgets, unfortunately unsuccessful.

    I would be very thankfull if you could share some code snippets or a little demo project.

    --
    Julian

    ReplyDelete
    Replies
    1. I have a composite component for that, but the code is private (belongs to the company, I'm currently working for).

      We plan to add this compoment to PrimeFaces Extensions. There is a task there, in the bug tracker. But my time is limited currently.

      Delete
    2. I understand, it would have been surplus anyway to share the code. One question though, how do you pass the data from the bean into the javascript?

      Delete
    3. See http://ovaraksin.blogspot.de/2014/07/how-to-get-json-response-from-jsf.html

      Delete
  3. Hm, I prepared everything just like in your post. Still I do not understand how to bind the json to the typehead dataset.

    var numbers = [...] this is where i need the json object and I dont understand how to bind my data.

    ReplyDelete

Note: Only a member of this blog may post a comment.