I think we should push this fix, although there are concerns we can't
immediately demonstrate
that Math.ceil may be wrong in some theoretical case.
Also Sergey, please add a comment in the bug report about the
alternative fix you mentioned
that was discussed previously so that can be re-visited if we
subsequently find problems as a result of this fix.
-phil.
On 09/19/2017 07:58 AM, Semyon Sadetsky wrote:
On 09/18/2017 12:01 PM, Sergey Bylokhov wrote:
Hi, Semyon.
Is it possible to write a test case for this issue? It will be
helpful to understand the bug and a fix.
Writing the test is hard because those inner interfaces are too deep
inside. You can use SwingSet2 demo to see the described artifacts at
fractional scales before the fix.
On 9/18/17 09:36, Semyon Sadetsky wrote:
Hello,
Please review fix for JDK10 (in Swing and Java2D):
bug: https://bugs.openjdk.java.net/browse/JDK-8187367
webrev: http://cr.openjdk.java.net/~ssadetsky/8187367/webrev.00/
Swing apps may have artifacts on HiDPI screens with fractional
scales. There are several issues which may cause various artifacts
but the current fix only eliminates one type of artifacts the
vertical and horizontal lines on painted surfaces (other issues are
addressed by [1] and [2]). This issue was introduced after 8073320
and then incorrectly fixed in 8163193.
The root cause is the painter pattern is drawn on a wrongly sized
and scaled image in case the image is an off-screen volatile image.
The painter doesn't take into account the volatile image inner scale
transformation which is not an identity in case of HiDPI , so the
resulting image is drawn on of off-screen surface that in the scale
times bigger than it is actually necessary. Since such images are
cached this was waste of RAM. Also, the bounds sent to the painter
class don't take into account the transformation rounding error the
latter caused artifacts on the drawing edges.
--Semyon
[1] https://bugs.openjdk.java.net/browse/JDK-8187585
[2] https://bugs.openjdk.java.net/browse/JDK-8187586
