On Fri, 28 Aug 2026 18:15:32 GMT, Martin Fox <[email protected]> wrote:
> This PR attempts to improve LCD text rendering on Windows and Linux. Changes > include: > > - (Windows only) When setting up DirectWrite the code now uses the > NATURAL_SYMMETRIC rendering mode except for very small glyphs where it uses > NATURAL. Using NATURAL_SYMMETRIC avoids distorted glyphs at specific pixel > sizes (see [JDK-8389632](https://bugs.openjdk.org/browse/JDK-8389632)) and > retains the curves along the top and bottom of the glyphs. Using NATURAL at > small sizes avoids glyphs turning very fuzzy and light. > > - The code is now consistently converts the colors from sRGB to a linear > space (more or less), composites them, and then converts the result back to > sRGB. > > - The shader applies a contrast equation to the LCD glyph mask which helps > emphasize the stems. The same equation is used by Skia and probably added by > Microsoft when they cleaned up text rendering for Chromium. BTW it’s just the > equation for a parabola that goes through points (0, 0) and (1, 1). > > My testing was mostly done on a 27 inch display with a resolution of > 2560x1440 and a screen scale of 150%. This was low enough to notice a > difference. Resolutions higher than that (like full-on Retina) tend to hide a > lot of sins. > > I recommend reading “The Raster Tragedy in Skia” which is concise but covers > a lot of ground. It contains a section on the challenges of compositing text > in sRGB space and also the issues getting LCD text to look dark enough > without inflating the stems. I wish I had found this earlier in the process. > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). I've added a manual test (tests/manual/text/TextCascade.java). In each block the left cascade is LCD and the right is grayscale. I use it mostly to compare differences in perceived stem widths depending on colors and anti-aliasing settings. ------------- PR Comment: https://git.openjdk.org/jfx/pull/2284#issuecomment-5481001349
