AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix lost precision case in SkComputeRadialSteps.

Browse Source 
With large offsets, we can end up with a tiny change in angle at each
radial step, so that rotSin and rotCos can get truncated to 0 and/or 1,
respectively. This means that we won't sweep the entire radial distance,
throwing future calculations off. Instead in this case we should fail.

Bug: oss-fuzz:21826
Change-Id: I24f5b1ac315a3697d2ea00ee4bd73652ec3e16dd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/392377
Commit-Queue: Jim Van Verth <jvanverth@google.com>
Reviewed-by: Kevin Lubick <kjlubick@google.com>
This commit is contained in:
Jim Van Verth 2021-04-02 15:42:35 -04:00 committed by Skia Commit-Bot
parent 3144a1e698
commit ffeee9bfde
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/utils/SkPolyUtils.cpp
View File

@ -494,6 +494,11 @@ bool SkComputeRadialSteps(const SkVector& v1, const SkVector& v2, SkScalar offse
SkScalar dTheta = steps > 0 ? theta / steps : 0;
*rotSin = SkScalarSin(dTheta);
*rotCos = SkScalarCos(dTheta);
// Our offset may be so large that we end up with a tiny dTheta, in which case we
// lose precision when computing rotSin and rotCos.
if (steps > 0 && (*rotSin == 0 || *rotCos == 1)) {
return false;
}
*n = steps;
return true;
}