On Fri, 27 May 2022 18:02:51 GMT, Alexander Zuev <kiz...@openjdk.org> wrote:
>> Detect the situation where we do need to perform interpolation during >> ImageIcon >> painting and set a hint to the rendering to perform bicubic approximation so >> image details are preserved during transition. > > Alexander Zuev has updated the pull request incrementally with one additional > commit since the last revision: > > Use small icon pointer for icons less than 24 pixels wide - this way we > get the 16x16 icon which is better for scaling purposes. I have fixed the 16x16 icon generation on Windows, but this will not help the generic problem that i demonstrated with the test case, on the magnification factors like 125% or 175% without the proposed fix for the ImageIcon the result looks trashy - jagged lines, missing details. I have done testing with the modified regression test case to see how my change affects performance. Notice that this is the worst case scenario, when all of the paintings trigger the fix (which is only triggered for MultiResolutionImage based icons and only when this MRI does not have the exact match for the requested resolution and scaling is required - this is like 0.01% of overall cases because for majority of image icons they are either single image based or MRI has the variant requested ready for display. And even in this worst case scenario on 1000 iterations test does not show any difference. On 10000 iterations the difference is measurable in tens of milliseconds between test runs. On 100000 iterati ons test execution with BICUBIC approximation is noticeable (like test run takes more than a second longer). We can limit the maximum size of icon we process by some reasonable number, like do not process icons bigger than 256 pixel wide or high so we do not cause any noticeable slowdown, but i still think this needs to be done - otherwise on these odd magnifications we look nothing alike the native applications (windows looks like using some sort of bicubic interpolation for their icons) and just plain bad. ------------- PR: https://git.openjdk.java.net/jdk/pull/7805
