AccessibleObjectFromEvent() works fine out-of-process in 64 bit builds. However, in-process, it seems to return an oleacc client accessible instead of a Gecko accessible. I have no idea why this is the case. As far as Gecko is concerned, WM_GETOBJECT should be received and handled the same way, regardless of whether it's in-process or out-of-process.
Marking as unconfirmed because this needs to be tested/verified outside of NVDA code. I'd use Accprobe, but I'm pretty sure it can't do in-context events for 64 bit processes.
Related NVDA ticket: http://www.nvda-project.org/ticket/1454
It seems to be AccessibleObjectFromWindow with the actual problem, though AccessibleObjectFromEvent calls it internally.
A work around has been implemented in NVDA changeset:main,4474 which replaces a call to AccessibleObjectFromEvent with custom code that sends WM_GETOBJECT, calls ObjectFromLresult, and tries accChild. All the things that I'd think AccessibleObjectFromEvent/AccessibleObjectFromWindow would do normally.
this allows NVDA to render buffers for 64 bit Mozilla Gecko documents.
Although further testing needs to be done to prove this, I don't think this bug can be Gecko specific.
Actually, after further investigation, the bug is Gecko's. when gecko handles WM_GETOBJECT it checks lParam to see if its OBJID_CLIENT: http://mxr.mozilla.org/mozilla-central/source/widget/src/windows/nsWindow.cpp#5236
However, MSDN states in http://msdn.microsoft.com/en-us/library/dd373892%28v=vs.85%29.aspx that the lParam must be cast to DWORD before comparison with one of the OBJID constants as on 64 bit Windows not doing this can cause issues with the sign extension.
My own testing with a custom 64 bit window proc handling WM_GETOBJECT confirms that the comparison does fail when AccessibleObjectFromWindow is called from 64 bit to 64 bit, but not called from 32 bit to 32 bit or 32 bit to 64 bit.
I do not understand yet why my own custom code I implemented in NVDA which sends the WM_GETOBJECT message manually gets around the bug.
Marking as new given Mick's thorough testing and findings.
(In reply to comment #3)
> Marking as new given Mick's thorough testing and findings.
guys, you've passed more than half of road :) after verbal patch description one step is to upload patch and ask for review ;) Thank you for doing this!
Created attachment 542408 [details] [diff] [review]
Created attachment 542409 [details] [diff] [review]
the previous one is wrong
try server build http://firstname.lastname@example.org
Comment on attachment 542409 [details] [diff] [review]
Review of attachment 542409 [details] [diff] [review]:
Mick verified fixed in Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:8.0a1) Gecko/20110720 Firefox/8.0a1