On Thu, 10 Sep 2026 04:30:07 GMT, Prasanta Sadhukhan <[email protected]> wrote:
> For me it still fails in HiDPI which is also mentioned in JBS comment. But it > does not fail for 1.25 but fails for 1.75, 2.25 etc so it seems there's some > product issue still to be fixed.. > > ``` > ./jdk/bin/java -Dsun.java2d.uiScale=1.75 OverriddenInsetsTest > Test failed at 0 0 c=ff5050b0 > Dumped res to: OverriddenInsetsTest_res.png > Exception in thread "main" java.lang.RuntimeException: Test FAILED. > at OverriddenInsetsTest.test(OverriddenInsetsTest.java:103) > at OverriddenInsetsTest.main(OverriddenInsetsTest.java:69) > ``` > > ``` > ./jdk/bin/java -Dsun.java2d.uiScale=2.25 OverriddenInsetsTest > Test failed at 99 0 c=ff3030cf > Dumped res to: OverriddenInsetsTest_res.png > Exception in thread "main" java.lang.RuntimeException: Test FAILED. > at OverriddenInsetsTest.test(OverriddenInsetsTest.java:103) > at OverriddenInsetsTest.main(OverriddenInsetsTest.java:69) > ``` I also see this behaviour only in Windows. After more analysis, i see that only Windows supports exact fractional scaling. Based on heavyweight component we pick different path at https://github.com/openjdk/jdk/blob/master/src/java.desktop/share/classes/java/awt/Robot.java#L613 Java2D rasterizes onto a scaled windows surface and for heavyweight components this is mapped to integer device pixels. Then robot screen-capture downsamples this 175x175 pixels to 100x100 logical pixels. This can result in picking different pixels on edges. Capturing the failed output at 1.75/2.25 scaling in Windows shows that only some part of edge pixels are not matching the expected results. This is not some product bug in paint/rendering. Attached output of these images. <img width="100" height="100" alt="OverriddenInsetsTest_res_1_75" src="https://github.com/user-attachments/assets/cabcbb28-0bac-4c84-87d5-3511af630dd4"; /> <img width="100" height="100" alt="OverriddenInsetsTest_res_2_25" src="https://github.com/user-attachments/assets/4e68e821-90d3-48e6-810a-406c7772c06e"; /> Also if we check the history, this test was added under [JDK-6694230](https://bugs.openjdk.org/browse/JDK-6694230). After this, test was never updated to actually use HiDPI. I don't know why this test was completely problemlisted on all platforms just because we saw some failures using HiDPI on windows. This test should be removed from problem list, so that we can have actual coverage for which it was written. I don't see windows specific product issue and this test should ignore the edge pixels. ------------- PR Comment: https://git.openjdk.org/jdk/pull/32779#issuecomment-5616124469
