On Sun, 27 Sep 2026 18:47:22 GMT, Christopher Schnick <[email protected]> wrote:
> It seems like long ago there were tiny rendering differences encountered > between the sw and hardware renderer when rendering rounded edges. Instead of > fixing the root cause, a workaround was added to just render the affected > edges using a manual sw renderer approach. It seems like everyong forgot, and > this workaround is still here. As a result, any region with uniform corner > radius 0 < r <= 4 is rendered using a software renderering approach right now. > > Maybe the issue has been fixed somehow, the linked issue is very unspecific. > Even if not, and there are differences between the renderers, they don't > warrant this performance slowdown. I can't see a difference so far with this > workaround disabled. > > You can see this case being logged when being run with > `-Djavafx.pulseLogger=true`. According to the counters at the end of a render > pass, around 10% of nodes were rendered using this ineffecient approach in > our application. > > This case still happens for non-uniform corner radius values, but that is > somewhat expected. > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). Re-reading the initial bug report it seems that the rounding of the corners was only part of the problem. The outermost pixels of the straight horizontal and vertical edges are being drawn with an alpha other than 100% so their colors are shifting compared to the interior. Based on the discussion comments this was causing visible issues along the sides of buttons with Modena and that was one of the driving force for the work-around. Again, not sure this is an issue worth pursuing anymore. I mention it just to ensure we're looking in the right spots when trying to spot regressions. ------------- PR Comment: https://git.openjdk.org/jfx/pull/2329#issuecomment-5916581070
