Boston 2007 Accessibility Summit Summary
QUICK SUMMARY: AWESOME SUMMIT!!! We focused on very concrete things and came away with people ready to work (or already working) on them. Yeah! What a great crowd.
Attendees
- Willie Walker - Chair
- Aaron Leventhal
- Mike Pedersen
- Steve Lee
- Eitan Isaacson
- Marco Zehe
- Ariel Rios
- Jorge Silva
- Charles Chen
- Mick Curran
- Scott Haeger
- Patrick Gu
- Evan Yan
- Janina Sajka
- Joanmarie Diggs
- Tim Keenan
- Mike Gorse
- Kristian Lyngstøl
- JP Rosevear
- Boyd Timothy
- Gijs Kruitbosch
- Attila Domokos
Testing
Eitan Isaacson, Patrick Gu, Evan Yan, Attila Domokos and Steve Lee met to discuss testing. There are a number of testing projects and efforts. These not only include testing platforms themselves (e.g., Dogtail and LDTP), but also includes companies and projects doing their own testing (e.g., Sun, RedHat, Nokia). We believe the following:
- There is not enough testing
- Better coverage would help prevent accessibility regressions
- There is considerable overlap in testing efforts
As a potential solution, we discussed creating an initiative on the gnome.org level, similar to build.gnome.org that would test UIs of the various GNOME modules. It would be easy to tack on accessibility tests to those. These tests might also be useful for the LSB a11y conformance initiative. Some ideas for testing include watching for specific AT-SPI events as well as looking at static snapshots of the AT-SPI hierarchy.
Eitan and Willie also talked about ideas for adding assertion capability to the Orca tests. The idea would be to assert that Orca is speaking and brailling expected output. The Python logging API coupled with the Orca HTTP service seems like it would be a good way to go. Willie will look into this further.
Adding MouseTweaks to Gnome
As a group, we discussed MouseTweaks, which provides features to improve mouse access for those that find it difficult to control a standard mouse (dwell click and left button access to context menu, see Demo Video).
The group agreed these features are useful and we'd recommend it is included after HIG review of the applets. While the features could possibly better fit as an X input device, we decided that it would get to users faster in its current form due to the long timescales and process required to get extension acceptance into X. It was also noted that the related mouse gesture project chickenscratch offers interesting features.
Yelp
Yelp accessibility via Orca is not very good, and we're waiting for yelp to migrate to the Gecko 1.9 rendering engine before investing more time in it (see bug 356041). Ariel Rios has contacted the maintainer and the discussion has been forwarded to gnome-doc-devel-list.
Firefox
See also the Mozilla Accessibility Summit page for more detailed information on the work accomplished at the Friday meeting.
Aaron Leventhal described detailed examples of how ATs such as Orca would use the live region support of WAI ARIA Live Regions to improve the user experience by providing features not otherwise available for dynamically updated pages such as stock or score results. As a group, we walked through several sample use cases (e.g., chat, timer, baseball scores) and Charles Chen's work on FireVox support of WAI ARIA. The existing priority queue described in the ARIA report seems to work well, and Scott Haeger has a patch ready for Orca to support it.
Scott Haeger, Mike Pedersen, and Willie Walker discussed a user interaction model for being able to override the politeness level settings of live regions. Mike will write it up and Scott will implement it.
Aaron Leventhal, Mick Curran, and Willie Walker discussed "role extensions" work being proposed by Aaron. The idea is to allow for subclassing of existing accessibility roles as well as providing a way to extend the object attributes. After some discussion, the idea seems like it will work and Aaron will take the ball forward for a future release of Firefox.
Scott Haeger and Ariel Rios discussed the AT-SPI collections API. Scott will be doing some performance evaluation of it by using it for Orca's page summary feature.
ORBit replacement with DBUS
Ariel Rios explained how Bonobo and ORBit are planned for deprecation. The big issue we're facing is that there is no clear plan for how to handle this deprecation from the AT-SPI point of view. As a group, we discussed two main ideas:
- Removing the Bonobo dependency in AT-SPI, keeping ORBit, and recommending ORBit not be deprecated. This is fairly straightforward, but will break binary compatibility and it requires working with the GNOME community to undeprecate ORBit. Finally, it has the potential of making the AT-SPI world one of the last in GNOME to use ORBit, raising concerns about support and future work. For example, would the small a11y community be the one left supporting ORBit?
- Migrating the AT-SPI to DBUS, working to keep API compatibility for AT's at the pyatspi level. The main issue with this space is that the performance concerns are unknown. In addition, the scope of the work is large and various bridges (e.g., atk-bridge, Java access bridge) would need to be rewritten.
On the idea of keeping ORBit, we noted that Bonobo is the component of AT-SPI that seems to have the most resistance from the community, and representatives from the embedded space indicated they would support staying with ORBit. As such, removing the Bonobo dependency seems to be the most expedient path for the short term.
From the DBUS perspective, there are concerns about how (or if) ATSPI could work on DBUS and existing performance results are not particularly realistic or reliable. We discussed several ways to investigate this space, including working on a prototype to help us understand the performance issues involved in accessing the object model and working with the bursty nature of AT-SPI events. Late on Sunday, we also engaged Rob Taylor, who is the maintainer of the D-Bus GLib bindings. Between Ariel and Rob, we think we have a good plan for investigating the viability of an AT-SPI-over-DBUS solution and will follow up after the summit. The Mozilla Foundation is working on providing funding for this and will also act as an impartial party.
Magnification
Kristian Lyngstøl, Joanmarie Diggs, Mike Pedersen, and Willie Walker discussed the eZoom work that Kristian is doing and ideas for how to move forward with magnification from both a technical and an end-user standpoint. One major tension that exists is the following:
- Some users want to use magnification only and do not want to have to run Orca to do so
- Users of Orca and magnification do not want to hunt around on the system for different configuration UI's. Instead, they want "one stop shopping".
We also discussed the Bonobo/ORBit issue, because the current magnification solution used by Orca (gnome-mag) is based on Bonobo/ORBit. We reached the following conclusions:
- The magnifier can provide its own user interface for configuring it and interacting with it. This allows users to use and configure the magnifier without requiring Orca.
- We can work on developing a simple DBUS-based mechanism for allowing Orca to talk with the magnifier. To start, we can work on a simple API for having Orca suggest the area of the screen the magnifier should present.
- To address the "one stop shopping" problem, the DBUS API can include a command to tell eZoom to display its configuration dialog.
- We will work with the existing DBUS interface to eZoom for now as a means to experiment/prototype what the DBUS interface should provide, and will also work with Carlos Eduardo Rodrigues Diógenes, the gnome-mag maintainer, to move towards a DBUS API that can be supported by magnifiers in general. The intent is that we will gradually evolve the eZOOM DBUS interface to allow for support by other magnification solutions (e.g., gnome-mag migrated to metacity).
- In parallel to the DBUS work, we will engage the magnifier-user community to make sure the work being done is in line with user requirements and expectations.
Other
Willie Walker and JP Rosevear worked on debugging why accessible login wasn't working on SUSE. After a fair amount of debugging, we were able to narrow the problem down to the AT-SPI registry not being started by gdm. JP will investigate further, but it appears as though the gdm "--with-atspi-dir" build option might be able to help.
Willie promoted accerciser as a powerful tool to the main Gnome group and Eitan demonstrated it to the Tom Tom developers, showing how it can be used to test and improve the accessibility of applications that use custom widgets.