Add a border radius property to inherit it from a parent in a fallback content

third_party/WebKit/Source/core/layout/line/InlineBox.cpp

 /*
  * Copyright (C) 2003, 2004, 2005, 2006, 2007 Apple Inc. All rights reserved.
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Library General Public
  * License as published by the Free Software Foundation; either
  * version 2 of the License, or (at your option) any later version.
  *
  * This library is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
@@ skipping 208 matching lines @@
 
 bool InlineBox::nodeAtPoint(HitTestResult& result, const HitTestLocation& locationInContainer, const LayoutPoint& accumulatedOffset, LayoutUnit /* lineTop */, LayoutUnit /* lineBottom */)
 {
     // Hit test all phases of replaced elements atomically, as though the replaced element established its
     // own stacking context.  (See Appendix E.2, section 6.4 on inline block/table elements in the CSS2.1
     // specification.)
     LayoutPoint childPoint = accumulatedOffset;
     if (parent()->lineLayoutItem().hasFlippedBlocksWritingMode()) // Faster than calling containingBlock().
         childPoint = layoutObject().containingBlock()->flipForWritingModeForChild(&toLayoutBox(layoutObject()), childPoint);
 
+    bool inside = lineLayoutItem().hitTest(result, locationInContainer, childPoint);

[pdr., 2016/01/31 21:32:36]

The other nodeAtPoint functions follow a pattern where, for non-rect hit
testing, the HitTestResult is updated if and only if nodeAtPoint returns true.
For example, in LayoutBlock::nodeAtPoint, all of the codepaths that return true
will call updateHitTestResult (except for rect-based tests). In this patch,
you're updating the result (through lineLayoutItem().hitTest) but returning
false from InlineBox::nodeAtPoint... that doesn't seem right. Can you
investigate why a different pattern is used here vs the other nodeAtPoints?

It may also be useful to find what code is working when InlineBox::nodeAtPoint
returns false but the HitTestResult has been updated. I suspect that code is
wrong.

[Miyoung Shin(c), 2016/02/05 23:40:03]

Your right. I found out we have only this issue with missing image.
We create a new element for missing image, but don't inherit the style from the
original IMG.
If we apply the right style, I think we don't need this code.
Thanks for letting me look deep into this.

It's a National holiday until a next week.
After I come back, I will upload the new patchset.

+
     if (lineLayoutItem().style()->hasBorderRadius()) {
         LayoutRect borderRect = logicalFrameRect();
         borderRect.moveBy(accumulatedOffset);
         FloatRoundedRect border = lineLayoutItem().style()->getRoundedBorderFor(borderRect);
         if (!locationInContainer.intersects(border))
             return false;
     }
-
-    return lineLayoutItem().hitTest(result, locationInContainer, childPoint);
+    return inside;
 }
 
 const RootInlineBox& InlineBox::root() const
 {
     if (m_parent)
         return m_parent->root();
     ASSERT(isRootInlineBox());
     return static_cast<const RootInlineBox&>(*this);
 }
 
@@ skipping 162 matching lines @@
 
 void showLineTree(const blink::InlineBox* b)
 {
     if (b)
         b->showLineTreeForThis();
     else
         fprintf(stderr, "Cannot showLineTree for (nil) InlineBox.\n");
 }
 
 #endif