Marquee width should not change a fixed width of its container.

Source/core/layout/LayoutBlock.cpp

Index: Source/core/layout/LayoutBlock.cpp
diff --git a/Source/core/layout/LayoutBlock.cpp b/Source/core/layout/LayoutBlock.cpp
index c717f16cd731914f84449fa5e7454a712bf67b39..212549aa523f1d262696a76279d4f4d542d1cbec 100644
--- a/Source/core/layout/LayoutBlock.cpp
+++ b/Source/core/layout/LayoutBlock.cpp
@@ -2007,6 +2007,12 @@ void LayoutBlock::computeIntrinsicLogicalWidths(LayoutUnit& minLogicalWidth, Lay
 
     maxLogicalWidth = std::max(minLogicalWidth, maxLogicalWidth);
 
+    if (!styleRef().autoWrap() && layer() && isHTMLMarqueeElement(node())) {

[mstensho (USE GERRIT), 2015/06/23 09:33:30]

What's with the layer() check?

Also not sure why we need the !styleRef().autoWrap() check.
HTMLMarqueeElement.js controls sets the white-space property on the MARQUEE
element (nowrap for horizontal marquees). If the document overrides this to get
a broken marquee, who cares?

[changseok, 2015/06/24 06:15:58]

This is garbage code. This change is inspired by webkit code. It checks if a
render block is a marquee through RenderLayer so it needs to check a layer
existence first. But we don't need to do because we manipulate the marquee with
a shadow root as you said, the layer has no way to check if a layout object is
marquee or not.
You're right, That was a redundant check for direction.

+        AtomicString direction = toElement(node())->getAttribute(directionAttr);

[mstensho (USE GERRIT), 2015/06/23 09:33:31]

I think I'd like to see this logic being moved to HTMLMarqueeElement. Add an
HTMLMarqueeElement::isHorizontalMarquee()?

[changseok, 2015/06/24 06:15:58]

OK

+        if (direction != "down" && direction != "up")
+            minLogicalWidth = 0;
+    }
+
     if (isTableCell()) {
         Length tableCellWidth = toLayoutTableCell(this)->styleOrColLogicalWidth();
         if (tableCellWidth.isFixed() && tableCellWidth.value() > 0)