On 1/15/20 9:11 am, Phil Race wrote:
You say this is for 14, but it is not a 14 regression.
Seems like it would be nice to fix but clearly not many people are running FX
on top of the latest JDK.
Can you please join the dots for me on how a change of false to true in
setBounds> is the way to fix it ? It is not obvious.
That flag is responsible of updating the target(usually noop) and posting
events after the size of the peer
is changed.
setBounds() will call handleMove() if the component was moved
https://github.com/openjdk/client/blob/master/src/java.desktop/macosx/classes/sun/lwawt/LWComponentPeer.java#L597
and handleMove() will post event if the flag is "true"
https://github.com/openjdk/client/blob/master/src/java.desktop/macosx/classes/sun/lwawt/LWComponentPeer.java#L1124
What testing have you done to ensure there are no other regressions either on
macos or other OSes ?
The fix in question affects the macOS only, and the file which was changed is
used by the FX only.
We have only one test which uses JLightweightFrame and it passes, I have run
client-tests on all
platforms just to be sure that no unexpected issues were found.
On 1/14/20 8:10 PM, Sergey Bylokhov wrote:
Hello.
Please review the fix for JDK 14.
Bug: https://bugs.openjdk.java.net/browse/JDK-8236953
Fix: http://cr.openjdk.java.net/~serb/8236953/webrev.00
The Swing components depends on the COMPONENT_RESIZED and
COMPONENT_MOVED posted by the AWT. These events are posted
in three situations:
1. If the Component is invisible then events are always posted
by the Component class.
2. If the Component is visible and the size is changed by the
user via setBounds() method then events are posted by the
component only if the component is not Window.
3. If the Component is visible and the size is changed by the
OS then the peer will post these events.
Note that in point 2. we skip any top-level windows because we
know that sometime later we will get a native callback and will
post events at point 3.
In JDK-8144125
I note that fix had only one reviewer. We should have two unless there's a good
reason not to.
I have fixed a bug when we always posted these
evens at point 2., and as a result, we got duplicated events after
point 3. But the fix caused current regression, the JLightweightFrame
which is used by the SwingNode is a kind of top-level window but w/o
the native part, so it will never get a native callback and should post
events ourself, we already do this on windows, and probably should do
this on Linux(it is currently works because on Linux AWT posts too
many events).
Is there a bug filed on this ?
-phil.
--
Best regards, Sergey.
