The fix looks good for me.
Just a minor comment. It seem that the condition in
DefaultDesktopManager: "if (!floaterCollision) { .. }
if(floaterCollision) {...}"
is the same as "if (!floaterCollision) { ... } else {...} ".
Thanks,
Alexandr.
On 1/31/2014 11:55 AM, Anton V. Tarasov wrote:
Hi Petr,
On 30.01.2014 18:16, Petr Pchelko wrote:
Hello, Anton.
Great, you are removing the dirty JViewPort hack)
I have one question: is JViewport a single place where we use the
"blit" rendering?
Yes, there's one more place - in DefaultDesktopManager. It BLITs on
dragging a JInternalFrame instance but only in FASTER_DRAG_MODE which
is not its default mode. I hadn't test it, but I tried. I switched the
mode and what I see is that w/o any modifications a dragged JIF is, as
expected, not repainted. I've put the notifications into the code.
This made a JIF instance repaint on drag. However, there's an issue
with the shadow. The shadow is not repainted, but background artifacts
instead. I didn't yet find the reason, I suspect there's some async
repaint which I probably don't catch.
I suggest not to spend time in DefaultDesktopManager for now, as a)
this functionality has less priority for SwingNode (than JViewport) b)
it works fine in the default mode. So, I'll file a P5 against it.
Please, look at the new webrev:
http://cr.openjdk.java.net/~ant/JDK-8033233/webrev.1
Besides modifications to DDM, I've put additional bounds constraint to
JLF/RepaintListener.repaintPerformed.
Also, could you please update the copyright years.
Sure, I did.
Also, may be we could replace the RepaintListener instantiation in
JLWF with a lambda? What do you think?
Sure, we can.
Thanks,
Anton.
With best regards. Petr.
On 30.01.2014, at 17:49, "Anton V.
Tarasov"<[email protected]> wrote:
Hi all,
Please review the fix.
jira:https://bugs.openjdk.java.net/browse/JDK-8033233
webrev:http://cr.openjdk.java.net/~ant/JDK-8033233/webrev.0
Here I'm duplicating JIRA:
The point of the fix is that it introduces a RepaintListener which
can be added to a RepaintManager in order to get back notifications
of repaints performed as BLITs. JLightweightFrame registers such a
listener to its RM instance. JViewport is the source of those
notifications. Now it works in default BLIT_SCROLL_MODE.
Shortly, the mechanism of repainting of a JLF is the following. Once
a JLF's child component is requesting repaint, an appropriate
repaint runnable is scheduled by the RepaintManager (RM). The
runnable is then gets dispatched by the RM which calls the paint()
method of the root component, that is the JLF. JLF overrides this
method in the way that after all the painting is done (super.paint)
it initiates a pixel bits transfer to the host application (e.g.
SwingNode). In case of JViewport, when it works in the default BLIT
scroll mode, scrolling of the JViewport doesn't lead to a repaint
runnable being dispatched. Instead, JViewport immediately repaints
its content (via blitting + repainting a dirty area) and then tells
the RM there's nothing to repaint (so the runnable scheduled is just
skipped). As the result, the JLF doesn't get any notification of the
update.
As a workaround to this problem, JViewport had been forcibly
switched to the BACKINGSTORE scroll mode, in which case it passes
the whole repainting cycle.
Regards,
Anton.
