Fix vector normalization in epaint/tessellator/add_line_loop. #5364
Conversation
|
Preview available at https://egui-pr-preview.github.io/pr/5364-normalization-fix |
|
Seems like this causes a couple small differences in the snapshot tests, the biggest one is visible in the rendering test: Bildschirmaufnahme.2024-11-11.um.14.28.34.movAnd some minor differences in the bezier test: Bildschirmaufnahme.2024-11-11.um.14.29.26.movNot sure which versions are more correct though, just wanted to share the differences I noted. Here is a link to the snapshot test results: https://github.com/emilk/egui/actions/runs/11778645510/artifacts/2171312779 @PangolinGiLA You'll need to run the tests with |
| @@ -470,7 +470,7 @@ impl Path { | |||
| self.add_point(points[i], n1c / n1c.length_sq()); | |||
| } else { | |||
| // miter join | |||
| self.add_point(points[i], normal / length_sq); | |||
| self.add_point(points[i], normal.normalized()); | |||
There was a problem hiding this comment.
Dividing with length-squared is what gets the correct widths for a miter-joins. If this fails for some angles, let's figure out why. The CUT_OFF_SHARP_CORNERS is meant to catch this problem.
There was a problem hiding this comment.
Oh sorry, I didn't realize that, I just changed it thinking normal vectors always have unit length, and this fixed my problem with rendering.
The problem appeared when I tried to render points distributed this way (this is the bottom part of the image above):
There was a problem hiding this comment.
Oh I just read the issue linked in the code. I guess i should have done a little bit more research before opening the PR.
|
I would love to have a proper solution to this problem, but this PR isn't it |
In epaint/tessellator/add_line_loop normal vectors of added points were not normalized correctly, they were divided by length squared, what caused their length to explode for some points close together.
Before:
After:
