This all looks good to me. It looks like we may be able to remove the workaround in the plugin where we have to update the size of the CALayer on a window resize. I knew something was wrong in the AWT but wasn't sure what.
-- Scott On Oct 23, 2012, at 5:45 AM, Sergey Bylokhov <[email protected]> wrote: > New version of the fix: > http://cr.openjdk.java.net/~serb/7198229/webrev.01/ > > 23.10.2012 16:34, Anthony Petrov wrote: >> On 10/23/2012 4:32 PM, Sergey Bylokhov wrote: >>>> src/macosx/classes/sun/lwawt/macosx/LWCToolkit.java >>>>> 503 public synchronized void run() { >>>> >>>> Why does this method have to be synchronized? Do you expect the >>>> Runnable.run() to be called many times from different threads? >>> I get this pattern from CDropTargetContextPeer.flushEvents(). This code can >>> be safely removed later. >> >> Unless there's a clear reason for this method to be synchronized, I suggest >> to remove this keyword from the method signature. >> >> >>>> I'm not an expert in CA layers, but I think changing from a layer backed >>>> view to a layer hosting view is a big change. What tests did you run to >>>> verify there's no regressions with rendering? >>> At least awt jck/regression tests passed. But sure some regressions could >>> be introduced. >> >> Thanks. The fix looks safe then. >> >> -- >> best regards, >> Anthony >> >>>> >>>> -- >>>> best regards, >>>> Anthony >>>> >>>> On 10/23/2012 3:41 PM, Sergey Bylokhov wrote: >>>>> Hi Everyone, >>>>> Please review the fix. >>>>> This is an attempt to make our resizing better. I try to eliminate all >>>>> unnecessary repaint actions. >>>>> 1 "Layer backed view" was changed to "Layer hosting view" - unnecessary >>>>> call from drawRect() was eliminated. Now we post Paint event in the >>>>> resize callback when necessary. >>>>> 2 CALayer options were changed, so now it don't force setNeedsDisplay. >>>>> 3 canDrawInCGLContext() was added to CGLLayer.m to eliminate unnecessary >>>>> call to drawInCGLContext() when our 2d texture is not ready. >>>>> 4 CPlatformWindow.deliverMoveResizeEvent() make an attempt to expose >>>>> LWWindowPeer synchronously. (LWCToolkit.invokeAndWait is used) >>>>> >>>>> After the fix we call Layer.setNeedsDisplay only in the Java2D Queue >>>>> Flusher threads. It means that we have no possibility to accelerate >>>>> resizing even more from awt side. >>>>> I'll make addition research of this question in the following CR. >>>>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7129082 >>>>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=8001213 >>>>> >>>>> Bug: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7198229 >>>>> Webrev can be found at: http://cr.openjdk.java.net/~serb/7198229/webrev.00 >>>>> >>> >>> > > > -- > Best regards, Sergey. >
