Use float epsilon instead of double to determine axis alignment
Previous to this CL, double epsilon was used as the zero threshold for
blink::TransformationMatrix::Preserves2dAxisAlignment(), while the
gfx::Transform::Preserves2dAxisAlignment() function used float epsilon
for this purpose. This brings the definitions back together, which
should hopefully fix crbug.com/938007
Bug: 938007
Change-Id: I2d9596c2c58e6a2c86fd1c0bbfee3ec87388bc0d
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1512154
Commit-Queue: Xianzhu Wang <wangxianzhu@chromium.org>
Auto-Submit: Mason Freed <masonfreed@chromium.org>
Reviewed-by: Xianzhu Wang <wangxianzhu@chromium.org>
Cr-Commit-Position: refs/heads/master@{#639213}diff --git a/third_party/blink/renderer/platform/transforms/transformation_matrix.cc b/third_party/blink/renderer/platform/transforms/transformation_matrix.cc
index 2ec1aa32..e325303 100644
--- a/third_party/blink/renderer/platform/transforms/transformation_matrix.cc
+++ b/third_party/blink/renderer/platform/transforms/transformation_matrix.cc
@@ -2030,7 +2030,9 @@
if (has_x_or_y_perspective)
return false;
- constexpr double kEpsilon = std::numeric_limits<double>::epsilon();
+ // Use float epsilon here, not double, to round very small rotations back
+ // to zero.
+ constexpr double kEpsilon = std::numeric_limits<float>::epsilon();
int num_non_zero_in_row_1 = 0;
int num_non_zero_in_row_2 = 0;
