On Mon, 2 May 2022 19:53:44 GMT, Alisen Chung <ach...@openjdk.org> wrote:
>> Changed the drawing area to be increased by 0.5 on the left side to prevent >> clipping > > Alisen Chung has updated the pull request incrementally with one additional > commit since the last revision: > > updated test I was worried that had missed something when I tested it yesterday and I re-built and ran the tests once again. The test still fails for me, more on it later. I commented out disposing of the frame and I see issues with border rendering too. The border renders correctly at 150% and 175%. Yet at 200%, some parts of the border are rendered as 1-pixel line and some are 2-pixel line. That's the problem we're trying to fix. As for the failing test, I noticed that the saved image is different from what I see on the screen especially when the uiScale is different from 1.5 which is used in the test. I can't explain it; it could be the result of rendering the frame on main thread after it was laid out on the EDT. I played around with border rendering myself. To get exact lines, I always use 1-pixel stroke. At first I used two rectangles to render the border. It's simple and it gives good enough result, however it leaves some pixels painted with background colour. Then I reimplemented painting to the way browsers render `groove` and `ridge` border. ([Try it](https://developer.mozilla.org/en-US/docs/web/css/border-style#try_it).) I believe, @lukeu mentioned it in one of his comments. The border uiScale=4.0 looks like this:  I pushed the code for it to [alisen-8279614-border](https://github.com/aivanov-jdk/jdk/commits/alisen-8279614-border) branch. You can get it for testing using these commands: git fetch https://github.com/aivanov-jdk/jdk alisen-8279614-border git checkout alisen-8279614-border You can also [compare my branch to this PR](https://github.com/openjdk/jdk/compare/pr/7449...aivanov-jdk:alisen-8279614-border), which should give you a quick overview of the differences. At 150% and 175%, the border looks like this:   The cyan colour is the background colour of the child panel in the test. It highlights rounding off to coordinates. The fourth panel in the 175% case lacks the yellow highlight on the right; at the same time, there are two cyan pixels. The highlight is likely painted outside of the panel clip. It may happen to bottom edge too. When I set `uiScale` to 1.0, the frame is displayed like this initially:  Weird, isn't it? After minimizing and restoring the window, it's painted correctly:  I can't understand why it happens… As for the test, I reformatted the code to use 4-space indents, changed the colours to stand out (white is too close to the grey background colour). The still fails with my implementation of border painting. It could be because there are some cyan background pixels where the test expects border… I haven't looked into it. The test needs updating. For each panel, it should verify _each edge_ has the correct thickness of both shadow and highlight colours. I still think the test can be made headless. Paint `content` panel to buffered image rather than the frame. Since we've seen different artefacts with different scales, the test should verify rendering at 1.0, 1.25, 1.5, 1.75, 2.0 rather than 1.5 only. ------------- PR: https://git.openjdk.java.net/jdk/pull/7449
