The vmIDsToWindowHandles cache is base on an incorrect assumption
that there is a mapping between a jvmid to a _single_ Window Handle.
This is incorrect as a Java application can have many top level
windows (each top level frame corresponds to a unique Window Handle).
The cache was introduced in 2007:
This code was incorrect at the time, but was good enough to make things
work with Java applications that have only one top level frame.
A commit in 2010 implemented the correct way of retrieving the window
handle from a given AccessibleContext:
This effectively removed the need of using cache, but the cache was
not removed for some reason.
The symptom of the issue is described in bug #5732.
…in a Java application with multiple windows such as IntelliJ and Android Studio.
Previously, NVDA incorrectly assumed a vmid mapped to a single hwnd. However, a Java application can have many top level frames, each mapping to its own hwnd.
To fix this:
* Use getTopLevelObject and getHWNDFromAccessibleContext to retrieve the hwnd of a given vmid/accessible context.
* Fallback to using a "last known" hwnd cache when the above fails.
* Keep track of the hwnd when firing event_gainFocus and event_caret so that we don't need to retrieve the hwnd using both methods above when processing those events.
* Use jab context only for equality, ignore window handles. Given comparing accessible context for equality does not require valid windows handles, and given that retrieving window handles for accessible context is not 100% reliable, we implement the JAB(Window)._isEqual method to be similar to UIA(Window): Compare the accessible components, and ignore the window handles.
Fixes #5732. Closes #5733 (PR).