[Blink] Unify pre-snapping of <video> to match <iframe>

This CL changes the pre-snapping policy of <video> so that only the size
of the contents is pre-snapped, while the location will exactly match the
content box's final location.

BUG=730284

Change-Id: Id10bd0fbd8021a11b7db4a2950a908f3d1b50ac0
Reviewed-on: https://chromium-review.googlesource.com/1141328Reviewed-by: Chris Harrelson <chrishtr@chromium.org>
Commit-Queue: Tien-Ren Chen <trchen@chromium.org>
Cr-Commit-Position: refs/heads/master@{#576317}

third_party/blink/renderer/core/layout/layout_embedded_content.cc

@@ -292,19 +292,16 @@ CursorDirective LayoutEmbeddedContent::GetCursor(const LayoutPoint& point,
 }
 
 LayoutRect LayoutEmbeddedContent::ReplacedContentRect() const {
+  LayoutRect content_rect = ContentBoxRect();
+  // IFrames set as the root scroller should get their size from their parent.
+  if (ChildFrameView() && View() && RootScrollerUtil::IsEffective(*this))
+    content_rect = LayoutRect(LayoutPoint(), View()->ViewRect().Size());
+
   // We don't propagate sub-pixel into sub-frame layout, in other words, the
   // rect is snapped at the document boundary, and sub-pixel movement could
   // cause the sub-frame to layout due to the 1px snap difference. In order to
   // avoid that, the size of sub-frame is rounded in advance.
-  LayoutRect size_rounded_rect = ContentBoxRect();
-
-  // IFrames set as the root scroller should get their size from their parent.
-  if (ChildFrameView() && View() && RootScrollerUtil::IsEffective(*this))
-    size_rounded_rect = LayoutRect(LayoutPoint(), View()->ViewRect().Size());
-
-  size_rounded_rect.SetSize(
-      LayoutSize(RoundedIntSize(size_rounded_rect.Size())));
-  return size_rounded_rect;
+  return PreSnappedRectForPersistentSizing(content_rect);
 }
 
 void LayoutEmbeddedContent::UpdateOnEmbeddedContentViewChange() {

third_party/blink/renderer/core/layout/layout_replaced.cc

@@ -636,6 +636,11 @@ LayoutRect LayoutReplaced::ReplacedContentRect() const {
   return ComputeObjectFit();
 }
 
+LayoutRect LayoutReplaced::PreSnappedRectForPersistentSizing(LayoutRect rect) {
+  rect.SetSize(LayoutSize(RoundedIntSize(rect.Size())));
+  return rect;
+}
+
 void LayoutReplaced::ComputeIntrinsicSizingInfo(
     IntrinsicSizingInfo& intrinsic_sizing_info) const {
   if (ShouldApplySizeContainment()) {

third_party/blink/renderer/core/layout/layout_replaced.h

@@ -63,6 +63,12 @@ class CORE_EXPORT LayoutReplaced : public LayoutBox {
   // This function returns the local rect of the replaced content.
   virtual LayoutRect ReplacedContentRect() const;
 
+  // This is used by a few special elements, e.g. <video>, <iframe> to ensure
+  // a persistent sizing under different subpixel offset, because these
+  // elements have a high cost to resize. The drawback is that we may overflow
+  // or underflow the final content box by 1px.
+  static LayoutRect PreSnappedRectForPersistentSizing(LayoutRect);
+
   bool NeedsPreferredWidthsRecalculation() const override;
 
   // These values are specified to be 300 and 150 pixels in the CSS 2.1 spec.

third_party/blink/renderer/core/layout/layout_video.cc

@@ -171,10 +171,7 @@ LayoutRect LayoutVideo::ReplacedContentRect() const {
   if (ShouldDisplayVideo()) {
     // Video codecs may need to restart from an I-frame when the output is
     // resized. Round size in advance to avoid 1px snap difference.
-    // TODO(trchen): The way of rounding is different from LayoutEmbeddedContent
-    // just to match existing behavior. This is probably a bug and We should
-    // unify it with LayoutEmbeddedContent.
-    return LayoutRect(PixelSnappedIntRect(ComputeObjectFit()));
+    return PreSnappedRectForPersistentSizing(ComputeObjectFit());
   }
   // If we are displaying the poster image no pre-rounding is needed, but the
   // size of the image should be used for fitting instead.