The IAccessible2 native UAA bridge is misbehaving on Windows XP when "experimental features" are enabled and an AT (NVDA 2013.2) is active and tracking with the Accprobe RCP. Windows XP sp3, 4GB RAM (VM Workstation 9.03 on Windows 7 sp1 64-bit OS). Also on a native Intel P4 PC with Windows XP sp3. With current daily of LODev4.3.0 master, briefly will see ia2 events reported and focus shifting, before getting an error dialog pop-up: LibreOfficeDev 4.3 - Fatal Error Caught Unkown Exception: Aborting! Version: 4.3.0.0.alpha0+ Build ID: 56ad826194d0b9b42ed8bde35836b0e59c46baa6 TinderBox: Win-x86@39, Branch:master, Time: 2013-11-30_11:04:29 And with a current daily build of LODev 4.2.0.0 beta1+, it is not crashing, but I was seeing some very strange focus event tracking--and resulting AT response. Version: 4.2.0.0.beta1+ Build ID: 829621680b028909d58b0480a78a6aee7154f31a TinderBox: Win-x86@42, Branch:libreoffice-4-2, Time: 2013-11-30_12:20:53 Both builds are stable on XP without the ia2 bridge activated using "experimental features".
Created attachment 90046 [details] simple test .odt document (In reply to comment #0) > > And with a current daily build of LODev 4.2.0.0 beta1+, it is not crashing, > but I was seeing some very strange focus event tracking--and resulting AT > response. > > Version: 4.2.0.0.beta1+ > Build ID: 829621680b028909d58b0480a78a6aee7154f31a > TinderBox: Win-x86@42, Branch:libreoffice-4-2, Time: 2013-11-30_12:20:53 I was using an HTML markup document for testing--just too complex. With a less complex Writer document, the 4.2.0beta1+ build was actually not that bad with its rendering in AT. And the build behaves about the same on Windows 7 in that on both OS the AT is tending to over announce the Headings List details when moving from paragraph to heading content. For the same document, the current master on Windows 7 has a cleaner rendering of each paragraph but it just won't run on Windows XP and throws the error noted.
I am seeing this crash: [1], when trying to test this change [2] on top of the current master. Interestingly, i see the crash only when NVDA is activated. LO crahes immediately after writer gained focus. Without running NVDA and with setting SAL_FORCE_IACCESSIBLE2=1, IA2 bridge get activated (seen in debuger) and not crashed (didn't verify if AT events are reported in AccProbe). [1] http://paste.openstack.org/show/54248 [2] https://gerrit.libreoffice.org/6819
Forgot to menion: Winodws 7 64 bit, compiled with MSVS 2010 in debug mode.
(In reply to comment #3) > Forgot to mention: Windows 7 64 bit, compiled with MSVS 2010 in debug mode. @David With TB-39 master on Windows 7 sp1 64-bit, AT with NVDA (2013.2) and Accprobe 1.2.1.1 active. I get clean IAccessible2 AT support, NVDA is stable and no issue with master. Version: 4.3.0.0.alpha0+ Build ID: 4b0d5b7cb08148a53ff7f64c48f52a3291b7a0c3 TinderBox: Win-x86@39, Branch:master, Time: 2013-12-01_01:21:52 This is an /A administrative install, so WRITEREGISTRY=1 not set. Are you installing to windows registry with your build? But I guess it could also be an issue with NVDA. Have you checked if it will run cleanly against one of the current TB-39 builds? And, this same build on Windows XP sp3, 32-bit (on VMWare Workstation 9.03) continues to crash with the same unknown exception error. AT with NVDA and Accprobe is active and showing AT focus events prior to crash, simple cursor or mouse movement seems to be enough to trigger the crash. The build is otherwise stable on XP sp3 when the AT is NOT active. Kendy is attempting to put up a symbols debug library for the build when I have that I should have a useable stack trace.
Today, on Windows XP sp3, 32-bit (on VMWare Workstation 9.03) Version: 4.3.0.0.alpha0+ Build ID: 90b2128711c26121205918f36385dd6c6e4df249 TinderBox: Win-x86@39, Branch:master, Time: 2013-12-03_05:44:11 No longer crashing with AT activated. Possibly this commit of Michael S's resolved... http://cgit.freedesktop.org/libreoffice/core/commit/?id=b54d441ad728ffce8e543de10bd897f0900422b5