Implementation Report

Table of Contents

• Introduction
• Discovery and access to descriptions
• Exposure through Accessibility API
• Reflecting the attribute
• Reported implementation Experience
• Multi-function tests - discovery, availability to the user
• Multi-function tests - API exposure
• DOM reflection

Discovery and access to descriptions

A set of tests are used to test the following requirements for User Agents (in section 3.03):

1. If the longdesc value is valid, User agents must make the link available to all users through the regular user interface(s).
3. User agents should enable users to discover when images in a page contain a long description.

The tests The results of these tests are shown in table 1. There are a number of user agents that pass these tests.

Exposure through Accessibility API

The same set of tests were used to demonstrate that user agents make the relevant information available through the MSAA accessibility interface, as shown in Table 2, as per the requirement

2. If the longdesc value is valid, User agents must expose the link to relevant APIs, especially accessibility-oriented APIs.
   

In addition, the DOM is used as an accessibility API by many systems (e.g. screenreaders interacting with Internet Explorer, such as NVDA). The test showing that longdesc is reflected in the DOM by multiple independent browsers (table 3) is further evidence that this requirement is met.

In practice, the fact that nearly all screenreaders except VoiceOver provide native access to longdesc demonstrates that this requirement is effectively met in the real world. For example in order to get data from Firefox, NVDA uses the IAccessible2 (IA2) accessibility API. Although there are no formal results testing the API directly, the fact that NVDA on Windows and Orca on Linux each successfully implement the specification with Firefox shows that in practice Firefox exposes correct data to both IA2 on Windows and AT-SPI/ATK on Linux.

Reflecting the attribute

The test that longdesc is reflected in the DOM primarily shows that the requirement

4. The longdesc IDL attribute must reflect the HTML content attribute longdesc.

(section 3.1.1) is met by multiple browsers, as per Table 3.

Reported implementation Experience

Extensions

There are a number of independently developed extensions for different browsers that make longdesc more readily available to users. These demonstrate a variety of approaches to enabling discoverability, whether by adding elements to the rendered webpage itself (as is also typical of screenreaders, which do not have a visible browser UI) or leaving the rendering intact and providing relevant indicators through the visible browser UI.

Dirk Ginader ports his extension from Chrome to Firefox as part of a one-hour presentation. Jim Thatcher's "favelet" is designed as a cross-browser 'extension'. An experimental (but not release-quality) port of the TellMeMore extension from Opera/Presto to Blink took a few hours. Whle the major bug (support for longdesc in frame and iframe) was not fixed in this experimental effort, the difficulties are not relevant to longdesc, but the complications of interacting with frame and iframe.

Opera

The Opera browser implemented support for longdesc in version 10.1. The implementation took about a week to develop, including time to debug and fix.

NVDA

During the Last Call for the specification, it was implemented in the NVDA screenreader. The developers had stated that they thought it made more sense to implement directly in the browser, but eventually decided that it was more important to provide the functionality to their users than wait for Firefox to implement longdesc natively. As the test results show, NVDA managed to successfully implement for at least two browsers. The developers clarified in mail that they are using different accessibility APIs on the same platform.

Orca

During the CR period, Orca (the most popular Linux screenreader) did an implementation from the specification without assistance from members of the Task Force. It reportedly took a few hours, and as Tables 1 and 2 shows, is a highly interoperable implementation.

Table 1: Multi-function tests - discovery, availability to the user

Notes

• ^[1] Discovery is achieved by user-config­urable high­lighting.
• ^[2] Access to the description is poor. You can get the URL of the longdesc, then copy it and paste it into a tab.
• ^[3] Discovery is weak - there is a longdesc option in the context menu when there is a longdesc available.
• ^[4] Data:uri images fail to render properly. (This is not a longdesc bug).
• ^[5] N/A: For tests where the image is empty, Gecko fails to expose the presence of the longdesc via ATK or "to all users" and hence Orca doesn't see it; this is a Gecko bug, not an Orca or longdesc bug.
  ✕: For tests where the description is found in a fragment, Orca does not present the page starting with that fragment. It presents the page starting from the top. Note, this is a Gecko bug, not an Orca bug.

Table 2: Multi-function tests - API exposure

Notes

• ^[1] accDescription has value of "data:text/html"
• ^[2] accDescription has value of "pass.html"
• ^[3] accDescription has value of "#pass"
• ^[4] accDescription has value of "pass-fragment.html#fragment"
• ^[5] accDescription has value of "fail.html"
• ^[6] ✕: accDefaultAction: [null]
• ^[7] ✔: accDefaultAction: showlongdesc
• ^[8] ✕: a bug prevents Gecko from making longdesc available when there is no image; this is neither an Orca bug nor a longdesc bug. See Mozilla bug 1056334.

Table 3: DOM reflection