Accessibility always on in GNOME 3.6.

[Luke Yelavich]

On Fri, Jun 22, 2012 at 06:25:22PM EST, Sebastien Bacher wrote:
> So I've read the bug upstream and I'm not sure how they are
> addressing the performances concerns, is the bottom line there "gtk3
> has no performance issues, a11y will not be turned on for gtk2 but
> only for gtk3"? If that's the case, what does it mean for gtk2
> applications? Do we get part of the desktop accessible and some
> other parts not? Do firefox and libreoffice use enough gtk to rely
> on the a11y to be turned on in gtk2 to be accessible in an usable
> way?

With regards to performance, upstream knows that there are still issues. If one is running an assistive technology like Orca, and opens a directory with a large number of folders and files in it using nautilus, there is currently a performance hit when the directory is opened, because nautilus/gtk/atk has to iterate through all of those entries, and create atk objects to associate with the GTK widgets in the icon view. How that will be solved long term I don't know, but the client tracking by atk and at-spi has been implemented to work around issues like the above.

The only change to the GTK2 module is to be linked against the new libatk-bridge library, which will do all the heavy lifting. The module still has the same symbols that it has always had. The only slight difference is that the GTK2 atk module will get loaded every time, assuming the GTK_MODULES environment variable is set or gnome-settings-daemon tells GTK the list of modules it should load. I think the libatk-bridge library has the smarts to go no further if it can't find the at-spi bus but even then, the same assistive technology client tracking comes into play as above, so no events will be emitted if there is nothing listening for events.

> Out of those questions I'm fine with follow upstream's lead there,
> we should just make sure it lands early so we have time to test and
> have a plan B to disable it again by default if that turns out
> creating too much issues.

It will land this week, and I should have the new libatk-bridge library packaged and ready to go in the archive by the time you and Jeremy get to updating GTK and gnome-shell, at least in source form. It will have to go through binary new due to new binary packages being added. Plan B will essentially involve reverting patches in GTK, GNOME shell, and at-spi2-atk to return to the previous behavior. I am pretty sure the gsettings key will stay around at least for now, and may or may not go away in future cycles depending on how well this a11y always on test works out this cycle.

Luke