On Wed, 10 Feb 2021 10:33:03 GMT, Prasanta Sadhukhan <psadhuk...@openjdk.org> wrote:
> Test hardcodes bufferedimage and frame bounds so it does not capture correct > size if scaling factor is used. > The solution is to always use the same uiscale=1, an updated test still can > reproduce the JDK-8015085 issue without fix. Additionally, moved the frame to > centre of screen for better mach5 stability. > Mach5 job running for several iterations on all platforms is ok. Link in JBS. I think the test does something wrong. Now I see how `uiScale` affects rendering. The test does not paint JLabel directly to the image but rather in its overridden `paintComponent`. So `uiScale` controls the size of the component or rather where the break occurs. For me on Windows with 200% scaling, in the first image ‘…’ is painted over the transparent background whereas in the second image ‘…’ is over the components opaque background. The test passes with `uiScale=1` but I see no ‘…’ at all. Both images are clipped at “.1”. As far as I understand, the purpose of the test is to make sure combining diacritics does not affect string clipping, and it occurs at the same position whether a composite character is used of a decomposed one. Yet no clipping occurs at all with `uiScale=1`. This looks wrong even though the test passes successfully. And when the test fails with `uiScale=2`, the images suggest there's a product bug because the string is clipped differently. ------------- PR: https://git.openjdk.java.net/jdk/pull/2502
