Fixing color bleeding when the canvas is stretched.

third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp

Index: third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp
diff --git a/third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp b/third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp
index 2a5a8ff5939da3b58adbbd60f3f5670202277781..5a262fb7603cc2239223fa1b5168535578e97a20 100644
--- a/third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp
+++ b/third_party/WebKit/Source/core/html/HTMLCanvasElement.cpp
@@ -323,12 +323,21 @@ void HTMLCanvasElement::didFinalizeFrame() {
   // Propagate the m_dirtyRect accumulated so far to the compositor
   // before restarting with a blank dirty rect.
   FloatRect srcRect(0, 0, size().width(), size().height());
-  m_dirtyRect.intersect(srcRect);
+
   LayoutBox* ro = layoutBox();
   // Canvas content updates do not need to be propagated as
   // paint invalidations if the canvas is accelerated, since
   // the canvas contents are sent separately through a texture layer.
   if (ro && (!m_context || !m_context->isAccelerated())) {
+    // If ro->contentBoxRect() is larger than srcRect the canvas's image is
+    // being stretched, so we need to account for color bleeding caused by the
+    // interpollation filter.
+    if (ro->contentBoxRect().width() > srcRect.width() ||
+        ro->contentBoxRect().height() > srcRect.height()) {
+      m_dirtyRect.inflate(0.25);

[Justin Novosad, 2016/10/25 14:30:15]

Are you sure 0.25 is always enough? Even at very large scale factors? 0.5 seems
more logical to me.

+    }
+
+    m_dirtyRect.intersect(srcRect);
     LayoutRect mappedDirtyRect(enclosingIntRect(
         mapRect(m_dirtyRect, srcRect, FloatRect(ro->contentBoxRect()))));
     // For querying PaintLayer::compositingState()