Changing between multicol and regular block shouldn't recreate all renderers.

Source/core/rendering/style/RenderStyle.cpp

Index: Source/core/rendering/style/RenderStyle.cpp
diff --git a/Source/core/rendering/style/RenderStyle.cpp b/Source/core/rendering/style/RenderStyle.cpp
index a4cb88432c2c76f68e13a4f1be202f7770eaf843..b7148fc6e63ea42c92ac321aee21d213127bec4b 100644
--- a/Source/core/rendering/style/RenderStyle.cpp
+++ b/Source/core/rendering/style/RenderStyle.cpp
@@ -180,7 +180,6 @@ StyleRecalcChange RenderStyle::stylePropagationDiff(const RenderStyle* oldStyle,
     if (oldStyle->display() != newStyle->display()
         || oldStyle->hasPseudoStyle(FIRST_LETTER) != newStyle->hasPseudoStyle(FIRST_LETTER)
         || oldStyle->columnSpan() != newStyle->columnSpan()

[esprehn, 2014/04/09 18:19:08]

We still need to reattach if the span changes though? What's the benefit of
avoiding a reattach when going between multi column and not.

It's not clear what you gain from this patch. Do people really dynamically add
multi column?

[mstensho (USE GERRIT), 2014/04/09 18:58:27]

In the old implementation, yes. In the new implementation, it may be feasible to
avoid it, but I'll put that on hold until the old implementation is gone.
It's probably not very common, but if we use the multicol implementation to do
paged overflow on the viewport (overflow:-webkit-paged-x on BODY, for instance),
it should be better to do this rather than reattaching RenderView (which, last
time I checked, seemed like a problematic thing to do).

-        || oldStyle->specifiesColumns() != newStyle->specifiesColumns()

[esprehn, 2014/04/09 18:19:08]

I assume we check this bit somewhere else too?

[mstensho (USE GERRIT), 2014/04/09 18:58:27]

Yes, we do.

         || !oldStyle->contentDataEquivalent(newStyle)
         || oldStyle->hasTextCombine() != newStyle->hasTextCombine())
         return Reattach;