Handle large rects better.

cc/trees/layer_tree_host_common_unittest.cc

Index: cc/trees/layer_tree_host_common_unittest.cc
diff --git a/cc/trees/layer_tree_host_common_unittest.cc b/cc/trees/layer_tree_host_common_unittest.cc
index e43a988097021b8298e141632795dc6187f7a5db..50312cdd47c4805e3dda7d320456c30bfddcb497 100644
--- a/cc/trees/layer_tree_host_common_unittest.cc
+++ b/cc/trees/layer_tree_host_common_unittest.cc
@@ -1124,9 +1124,14 @@ TEST_F(LayerTreeHostCommonTest, LayerFullyContainedWithinClipInTargetSpace) {
                                  page_scale_layer, inner_viewport_scroll_layer,
                                  outer_viewport_scroll_layer);
 
-  // Mapping grand_child's bounds to screen space produces an empty rect so
-  // grand_child should be hidden.
-  EXPECT_EQ(gfx::Rect(), grand_child->visible_layer_rect());
+  // Mapping grand_child's bounds to screen space produces an empty rect, but
+  // only because it is turned sideways.  The entire rect is contained inside
+  // the clip, and is only empty so land as the numerical precision of the

[Peter Mayo, 2017/03/28 20:15:40]

s/land/long/

Done.

+  // transform is effectively perfect.  Currently we do the calculation the
+  // way around, and the Projection of the screen space clip into layer space

[jaydasika, 2017/03/28 20:11:48]

nit : the 'other' way around ?

[Peter Mayo, 2017/03/28 20:15:40]

Done.

+  // includes the entire bounds.
+  EXPECT_EQ(gfx::Rect(grand_child->bounds()),
+            grand_child->visible_layer_rect());
 }
 
 TEST_F(LayerTreeHostCommonTest, TransformsForDegenerateIntermediateLayer) {