Add fast/inline to virtual/slimmingpaint and make the tests pass

Source/core/paint/InlinePainter.cpp

Index: Source/core/paint/InlinePainter.cpp
diff --git a/Source/core/paint/InlinePainter.cpp b/Source/core/paint/InlinePainter.cpp
index d8708834faad5f7bcb49ed0a10d38556b6e63374..8f79fd7f755b05a959b119fd6b92707afd8a403f 100644
--- a/Source/core/paint/InlinePainter.cpp
+++ b/Source/core/paint/InlinePainter.cpp
@@ -6,6 +6,7 @@
 #include "core/paint/InlinePainter.h"
 
 #include "core/paint/BoxPainter.h"
+#include "core/paint/DrawingRecorder.h"
 #include "core/paint/LineBoxListPainter.h"
 #include "core/paint/ObjectPainter.h"
 #include "core/rendering/GraphicsContextAnnotator.h"
@@ -29,6 +30,13 @@ void InlinePainter::paintOutline(PaintInfo& paintInfo, const LayoutPoint& paintO
     if (!styleToUse->hasOutline())
         return;
 
+    LayoutRect bounds;
+    if (RuntimeEnabledFeatures::slimmingPaintEnabled()) {
+        bounds = m_renderInline.linesVisualOverflowBoundingBox();

[leviw_travelin_and_unemployed, 2014/11/14 22:27:59]

This is going to be really expensive to calculate for every inline. We don't
need this to wrap perfectly. We should just use the cached visual overflow from
our containing block.

[chrishtr, 2014/11/14 22:31:43]

It suffices to use a bound that includes the inline, so use the bounds of the
entire line box list?

[fs, 2014/11/17 16:13:20]

I've used Levi's suggestion for now. I'm not sure how the bounds of the entire
line box list (assuming that means root box list?) would be cheaper - could you
elaborate?

+        bounds.moveBy(paintOffset);
+    }
+    DrawingRecorder recorder(paintInfo.context, &m_renderInline, paintInfo.phase, bounds);
+
     if (styleToUse->outlineStyleIsAuto()) {
         if (RenderTheme::theme().shouldDrawDefaultFocusRing(&m_renderInline)) {
             // Only paint the focus ring by hand if the theme isn't able to draw the focus ring.