commit e3deee131979b5e2c6ee07cad59a06c247a40d91
author Brian Osman <brianosman@google.com> Tue Nov 20 11:10:15 2018 -0500
committer Skia Commit-Bot <skia-commit-bot@chromium.org> Tue Nov 20 16:42:04 2018 +0000
tree 9e98dd0d623ed530bb6b99ca38362656d91146b9
parent 343063f3387b5c972c56fea614e31271f37c05e5

Use a smaller tolerance when chopping conics to quads

In most places we were using 0.25 (maximum distance between the actual
and approximated curves). In these few places, we were using 0.5, which
is enough to produce visible errors.

Added a GM that demonstrated the problem as reported - many of the arcs
with radius ~8 were previously closer to round-rects than circles.

Bug: chromium:888453
Change-Id: I7d22e27773f56174861526dd0223f52a93bf48eb
Reviewed-on: https://skia-review.googlesource.com/c/172060
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Greg Daniel <egdaniel@google.com>

src/gpu/ops/GrAAConvexPathRenderer.cpp

diff --git a/src/gpu/ops/GrAAConvexPathRenderer.cpp b/src/gpu/ops/GrAAConvexPathRenderer.cpp
index 624c261..2921b3e 100644
--- a/src/gpu/ops/GrAAConvexPathRenderer.cpp
+++ b/src/gpu/ops/GrAAConvexPathRenderer.cpp
@@ -316,7 +316,7 @@
                 m.mapPoints(pts, 3);
                 SkScalar weight = iter.conicWeight();
                 SkAutoConicToQuads converter;
-                const SkPoint* quadPts = converter.computeQuads(pts, weight, 0.5f);
+                const SkPoint* quadPts = converter.computeQuads(pts, weight, 0.25f);
                 for (int i = 0; i < converter.countQuads(); ++i) {
                     update_degenerate_test(&degenerateData, quadPts[2*i + 1]);
                     update_degenerate_test(&degenerateData, quadPts[2*i + 2]);