[Webkit-unassigned] [Bug 230257] [GTK][a11y] Add implementation of component interface when building with ATSPI

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=230257

--- Comment #11 from Carlos Garcia Campos <cgarcia at igalia.com> ---
(In reply to Andres Gonzalez from comment #8)
> (In reply to Andres Gonzalez from comment #7)
> > (In reply to Carlos Garcia Campos from comment #6)
> > > (In reply to Andres Gonzalez from comment #5)
> > > > (In reply to Carlos Garcia Campos from comment #4)
> > > > > (In reply to Andres Gonzalez from comment #3)
> > > > > > (In reply to Carlos Garcia Campos from comment #1)
> > > > > > > Created attachment 440592 [details]
> > > > > > > Patch
> > > > > > 
> > > > > > --- a/Source/WebCore/accessibility/isolatedtree/AXIsolatedObject.h
> > > > > > +++ a/Source/WebCore/accessibility/isolatedtree/AXIsolatedObject.h
> > > > > > 
> > > > > > +    RefPtr<AXIsolatedObject> nodeForID(AXID) const;
> > > > > > 
> > > > > > nodeForID belongs to the tree not to the isolated object. I think it is a
> > > > > > better design to keep it in AXIsolatedTree. Do you have a reason to move it
> > > > > > to AXIsolatedObject?
> > > > > 
> > > > > Yes, I need it in AccessibilityObjectAtspi::hitTest() only because of WTR
> > > > > actually. Since WTR calls accessibility api from the web process main
> > > > > thread, I can't use AXIsolatedObject::accessibilityHitTest() because it's
> > > > > always expected to be called from the secondary thread (nodeForID only works
> > > > > from the a11y thread). I can't use callOnAXthreadAndWait, because first
> > > > > thing AXIsolatedObject::accessibilityHitTest() is
> > > > > retrieveValueFromMainThread() but main thread is blocked. So, I split it
> > > > > adding AccessibilityObjectAtspi::objectAtPoint() that is always called by
> > > > > the main thread and does the coordinate conversion + accessibilityHitTest().
> > > > > When called from the main thread (WTR only) it's called directly and the
> > > > > wrapper object returned. When called from the AX thread I need the
> > > > > nodeForID() call to get the wrapper from the isolated object. 
> > > > 
> > > > Can't you get the wrapper from the isolated object? AXCoreObject::wrapper()
> > > > should work in all objects, isolated or live.
> > > 
> > > hmm, I'm not sure it's ok to get the wrapper from the main thread, I assumed
> > > that AXIsolatedObject::accessibilityHitTest() returned the AXID because it's
> > > safe to send the id from main to AX thread, but not the wrapper?
> > 
> > Let's go back for a moment to the origin of the problem. As you said, you
> > couldn't use AccessibilityController::executeOnAXThreadAndWait for HitTest
> > because HitTest  will dispatch back to the main thread that it would be
> > waiting, resulting in deadlock. The way we worked around this was by:
> > 
> > AccessibilityController::executeOnAXThreadAndWait calls
> > AXThread::dispatch that in turn calls
> > axThread.wakeUpRunLoop
> > 
> > On Mac wakeUpRunLoop does
> > 
> > void AXThread::wakeUpRunLoop()
> > {
> >     CFRunLoopSourceSignal(m_threadRunLoopSource.get());
> >     CFRunLoopWakeUp(m_threadRunLoop.get());
> > }
> > 
> > But it is a no-op in GTK. Could we try something equivalent in GTK?
> 
> The issue is not unique to HitTest, but you would have the same problem for
> any call into the API that needs to dispatch to the main thread. So we need
> to find a general solution.

I managed to handle all other cases by just splitting the methods, a common one that always runs in the main thread (or simply calls retrieveValueFromMainThread) and the one that is always expected to be called from the AX thread. For requests coming from the ATSPI service I use the method that is expected to be called from the AX thread and calls the common one. From WTR I call the common one directly from the main thread, since retrieveValueFromMainThread just calls the lambda if already in the main thread. That's the advantage of using the WebCore wrappers directly from WTR, without any a11y toolkit in the middle. There are a few cases like children() that I always run from the AX thread in WTR because those don't schedule a main thread task.

I have the WTR changes in a local commit that I plan to land after landing the text interface support, that's why you still see stubs for everything when building with atspi.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20211010/e9b7e727/attachment.htm>