<AWT Dev> Proposal for consolidation of KeyboardFocusManagerPeer

[Roman Kennke]

Hi there,

One thing that's bugging me for a while is how the ComponentPeer's
requestFocus() method is supposed to work. As far as I could figure out,
it's basically always like this (I use KFMHelper to call the
corresponding KeyboardFocusManager's private methods by reflection):

    public boolean requestFocus(Component lightweightChild, boolean
temporary,
            boolean focusedWindowChangeAllowed, long time, Cause cause)
{
        if (KFMHelper.processSynchronousLightweightTransfer(window,
                lightweightChild, temporary, focusedWindowChangeAllowed,
time)) {
            return true;
        }

        int result = KFMHelper.shouldNativelyFocusHeavyweight(window,
                lightweightChild, temporary, focusedWindowChangeAllowed,
time,
                cause);

        switch (result) {
        case KFMHelper.SNFH_FAILURE:
            return false;
        case KFMHelper.SNFH_SUCCESS_PROCEED:
                    requestFocusImpl(window, lightweightChild,
temporary,
                                 focusedWindowChangeAllowed, time,
cause);
        case KFMHelper.SNFH_SUCCESS_HANDLED:
            // Either lightweight or excessive request - all events are
            // generated.
            return true;
        default:
            return false;
        }
    }

The only thing that really differs between implementations would be the
requestFocusImpl() method call in the SNFH_SUCCESS_PROCEED case. The
rest seems to be the same in all implementations, except that in one
case (Windows I believe) it is done in JNI while in others (X11) it's
done by reflection.

I think this can be consolidated by doing the above directly in the
KeyboardFocusManager, before calling the peer requestFocus(), and have
the peer's requestFocus() only do the requestFocusImpl() handling. This
way we could avoid duplicate code and avoid reflection/JNI altogether.

Maybe I am missing something?

If not, I would work on a patch to move the above KeyboardFocusManager
calls into the KFM and have the peer only bothers with the part that is
requestFocusImpl() in the above example. Does that sound reasonable? It
would certainly make some things simpler in OpenJDK as well as Cacio and
the JavaFX SwingView that I am working on.

Best regards, Roman