Re: GNOME Accessibility on by default, and Firefox

[Willie Walker <William Walker Sun COM>]

Hi All:

There are a lot of things to think about here. From the internals point 
of view, we can think about the following:

1) The accessibility gconf setting: this says whether or not 
accessibility has been requested for the session.  Right now, this 
tends to be used by GTK+ just to say whether or not to load the 
accessibility modules and start the AT-SPI registry.  We've been trying 
for a long time to not require this setting and just always have the 
support turned "on".  But, there are performance and stability issues 
that prevent us from doing this, some of which might be addressed by 
making things more dynamic.  BTW, I'm not sure about the details of 
what the Gecko implementation does, but it would surprise me if it 
*always* loaded the accessibility modules regardless of the gconf 
setting.

2) ATK and the ATK bridge: the ATK is a utility toolkit for use by 
applications and graphical toolkits to create an accessible 
representation of widgets: ATK peers are created for 
application/toolkit widgets and ATK events are issued for 
application/toolkit events.  The bridge provides the communication with 
the "outside world." That is, the bridge is the thing that currently 
speaks CORBA and which we are moving to D-Bus.  Note that ATK and the 
ATK bridge are not required for an application to participate in the 
AT-SPI infrastructure; they merely make it easier to do so.  So, GTK+, 
OOo, and Gecko all use the ATK and ATK bridge.  Java, on the other 
hand, currently has its own bridge and speaks CORBA directly (I've 
argued that Java should use ATK via JNI to help normalize things 
somewhat and make it less dependent upon the AT-SPI transport).

3) The AT-SPI registry.  This provides the rendezvous mechanism between 
applications and assistive technologies.  When an application starts, 
it lets the registry know it exists so that assistive technologies can 
discover it.  An application also issues events to the registry, which 
then delivers them to assistive technologies.  When an assistive 
technology starts, it lets the registry know the event types it is 
interested in.  Like the ATK bridge, the communication with the 
registry is dependent upon the IPC mechanism being used (i.e., CORBA or 
D-Bus).

4) GAIL: this provides the accessibility implementation for GTK+, 
creating ATK peers for GTK+ widgets and causing ATK events to be 
issued.  It is designed to loaded as a GTK+ module at GTK+ 
initialization time and is currently not designed to be 
unloaded/reloaded over the course of an application's lifetime.  
Modifying it to be more dynamic might be rather difficult, and this is 
independent of the IPC mechanism in use.  That is, I don't believe its 
behavior matters whether we use CORBA or D-Bus.

From the assistive technology point of view, we can think of the 
following:

1) Assistive technologies that mostly examine object hierarchies.  GOK 
tends to be this type of assistive technology.  Accerciser fits into 
this category as well.  These kinds of ATs might work with the "just 
wakes up" model, especially if one queries from the top of the object 
hierarchy down.  However, *something* needs to already be awake so that 
an assistive technology can discover the top level application object 
in the first place.

2) Assistive technologies that mostly listen for events.  Orca tends to 
fit into this category.  I believe GOK also needs to listen for events, 
however, so it knows where to enter text and which UI needs to be 
grabbed.  These kinds of assistive technologies might be more difficult 
to support the "just wakes up" model since they tend to depend upon 
receiving events from the application in order to learn that an 
application exists.

Given these, I'm not sure it's really feasible to completely shut off 
accessibility in an application and dynamically turn it on when an 
assistive technology appears.  That is, I think something probably 
needs to live in the application to at least support a rendezvous with 
an assistive technology.  I also believe the main performance issues we 
face are the ATK peering of objects and event notification: we don't 
want unnecessary ATK peers to be created and we don't want unnecessary 
events to be issued.  As such, I think we at least need the ability for 
assistive technologies to discover applications that are already 
running and to be notified when new applications start.  From there, we 
could investigate generating ATK peers and issuing ATK events on an "on 
demand" basis.

So...to make a long story short, I'd guess most of the work needed is 
independent of CORBA/D-Bus and would live in the AT-SPI implementation 
for the toolkit (e.g., GAIL).  Li Yuan would be a good person to help 
us understand the scope of this problem.

Will

On Oct 23, 2008, at 9:54 PM, David Bolter wrote:

> Hi Aaron,
>
> Replying to a message from Aaron which didn't have the lists cc'ed.
>
> As Steve has brought up separately, this might indeed fit into the 
> D-BUS
> AT-SPI migration work. It doesn't seem terribly complex to add this
> small thing that will have nice benefit (like what T.V. points out).
>
> @Will Walker, we could maybe touch on this Monday night at 10 ;)
>
> Are there any objections to all this?
>
> cheers,
> David
>
> Aaron Leventhal wrote:
> > In Firefox, there's a lot of code that doesn't need to run, and a lot
> > of objects don't need to be created, when a11y isn't needed.
> >
> > Please, add lazy instantiation for accessibility. Would it fit to do
> > this in the D-BUS AT-SPI migration project?
> >
> > - Aaron
> >
> >
> > On 10/22/2008 8:35 PM, David Bolter wrote:
> > > This is a common pattern. I hear it is also true for other things in 
> > > FF
> > > like dom mutation events.... only fire if there is a listener... 
> > > makes a
> > > lot of sense.
> > >
> > > A tree should only fall in the forest if someone is listening.<grin>
> > >
> > > D
> > > Aaron Leventhal wrote:
> > >
> > > > Mario, on Windows we use lazy instantiation. There is no need for an
> > > > "enable a11y" flag in the OS. Just the fact that something asks us 
> > > > for
> > > > an accessible object wakes us up. We get a WM_GETOBJECT event -- and
> > > > before that a11y is not loaded.
> > > >
> > > > We really need lazy instantiation like that under Gnome. If the AT 
> > > > is
> > > > loaded and starts asking us for objects, then we can wake up.
> > > >
> > > > - Aaron
>
>
>
> _______________________________________________
> gnome-accessibility-list mailing list
> gnome-accessibility-list gnome org
> http://mail.gnome.org/mailman/listinfo/gnome-accessibility-list