A theoretical fix, it should:
- Catch any NSException (as it was initially expected);
- Probably fix possible memory leaks (by moving the layoutId var inside autoreleasepool).
(cherry picked from commit ae9520ce4f)
* Execute AccessibleAnnouncer.nativeAnnounce on a background thread on Windows to fix UI freezes. IntelliJ calls this method from EDT, but it doesn't need to run on EDT because on Windows it simply calls screen readers API without interacting with UI components. Additionally, when using a background thread, the JAWS SayString method, which previously could have been running for multiple seconds, is now executed immediately as expected, but the root cause of previous delays is unclear.
* In JawsAnnouncer, initialize COM library with the multithreaded model to allow executing it from different threads. Now COM is initialized and uninitialized on every call of the method as required by the [documentation](https://learn.microsoft.com/en-us/windows/win32/learnwin32/initializing-the-com-library): "Each thread that uses a COM interface must make a separate call to this function. For every successful call to CoInitializeEx, you must call CoUninitialize before the thread exits". IJawsApi COM object is still static and reused by different threads, which is allowed with a multithreaded concurrency model. It shouldn't cause issues because it has no state and only forwards calls to JAWS.
(cherry picked from commit 8cc4cd5cfd)
(cherry picked from commit d7d8d9b8e4)
(cherry picked from commit 8bfd24f89f)
Fixes different platform-specific issues with disabling/enabling input methods support via java.awt.Component#enableInputMethods(boolean):
* Windows: disabling used to require to deactivate and then activate back the app window to be performed completely ;
* macOS: disabling used to leave the input method window visible (however, it wasn't affecting the input) ;
* Linux: with fcitx5 IMF (may not be reproduced with iBus) disabling and then enabling back the input method support used to reset the used input source (keyboard layout) to a default one. It's done via disabling the current XIM when the input method support is disabled instead of disposing it .
(cherry picked from commit 2933ea89f0)
The patch also contains the fix for "JBR-6711: java/awt/LightweightDispatcher/LWDispatcherMemoryLeakTest.java: JButton JPanel not collected."
It unsets its references to tracked components thus not preventing GC from collecting them.
(cherry picked from commit e9aab98a6c)
(cherry picked from commit 13284aa508)
* Call UAZoomChangeFocus function when the keyboard focus is changed or when an accessible selection event is fired. Zoom viewport fill follow the accessible frame of the newly focused component, and if it has a selected child, its frame will be passed as part that needs to be highlighted.
* Fix an exception in ComboBoxAccessibility.accessibilitySelectedChildren when its selection is nil. This happens when a combo box doesn't have a selected item, and it this case Zoom wouldn't be able to follow the combo box location.
* Move the native handler of CAccessibility.focusChanged method from the legacy JavaComponentAccessibility to the new CommonComponentAccessibility class. It calls a class method (like a static method in Java), so there is no difference in which specific class it's located, but it allows to write the new code in the appropriate class.
(cherry picked from commit 75b06421ff)
(cherry picked from commit d0f47ced65)
