Hi Dmitry,

If you make a detailed look on the Component class methods like setEnables(), setFocusable() and setVisible() the most recent focus owner is cleared or transferred only after the component is made non-focusable. Since the fix is aimed for concurrent scenarios the recheck actually makes sens.

Also it would be more correct always trying to determine the first focusable component in traversal order before before the attempt to set focus owner as it is done in restoreFocus(Window aWindow, Component boolean clearOnFailure):

                if (!toFocus.isShowing() || !toFocus.canBeFocusOwner()) {
                    toFocus = toFocus.getNextFocusCandidate();
                }
otherwise when focus is restored to actually non-focusable component the final focus owner may become null while the next component in container traversal order is expected to have the input focus.

--Semyon


On 10/04/2017 03:25 AM, Dmitry Markov wrote:
Hi Sergey,

Good catch!
Actually we check restoreFocusTo before its usage, (i.e. we compare restoreFocusTo with most resent focus owner for the window; most recent focus owner is visible and focusable, otherwise it is removed from recent focus owners map). So the checks before assignment are not really necessary. I updated the fix (reverted it to the previous version). Please find webrev here: http://cr.openjdk.java.net/~dmarkov/8155197/webrev.03/ <http://cr.openjdk.java.net/%7Edmarkov/8155197/webrev.03/>

Thanks,
Dmitry
On 3 Oct 2017, at 18:30, Sergey Bylokhov <sergey.bylok...@oracle.com <mailto:sergey.bylok...@oracle.com>> wrote:

Hi, Dmitry.
If this check is valid then probably the same check should be added to restoreFocus()? before:
#173 restoreFocusTo = toFocus;

But as it was mentioned before, the component can become non-visible/non-focusable at any point, for example after this check but before the assignment. And instead of data validation before assignment we should check them before use. As far as I understand we already do this, all usages of restoreFocusTo are a checks == or != to some other component which is visible and focusable, isn't it?

On 10/3/17 08:23, Dmitry Markov wrote:
Hi Semyon,
I have updated the fix based on your suggestion. The new version is located athttp://cr.openjdk.java.net/~dmarkov/8155197/webrev.02/ <http://cr.openjdk.java.net/%7Edmarkov/8155197/webrev.02/>
Also I slightly modified the test to simplify it.
Thanks,
Dmitry
On 2 Oct 2017, at 18:32, Semyon Sadetsky <semyon.sadet...@oracle.com <mailto:semyon.sadet...@oracle.com><mailto:semyon.sadet...@oracle.com>> wrote:

Hi Dmitry,

Actually the parent frame doesn't own the input focus when the issue happens. The focus is on the dialog already and when FOCUS_GAINED event comes for the dialog the KFM discovers that the dialog should not have the focus and rejects it in line 588 of the DefaultKeyboardFocusManager:

restoreFocus(fe, newFocusedWindow);

This happens when the dialog became non-focusable (non-visible) after the focus request is sent, but before the corresponding FOCUS_GAINED event is handled on the EDT. In this case the focus is directly restored to the previously focused window which doesn't have the focus at this moment and input focus cannot be requested to one of its components synchronously.
Please confirm whether you agree on the root cause of the bug.

You are right. I agree with your evaluation.
Thanks.
Before setting the restoreFocusTo to toFocus in line 190 I would recheck for toFocus.isShowing() && toFocus.canBeFocusOwner() once again because the component can be made non-focusable concurrently.

--Semyon




--
Best regards, Sergey.


Reply via email to