Prune createVisibleSelectionDeprecated from SelectionAdjuster

third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp

Index: third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp
diff --git a/third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp b/third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp
index e94156d02c1b8b69dc8c1087b694d12f12973d45..173900d49cc8f058319a04032189f0d92cac2fad 100644
--- a/third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp
+++ b/third_party/WebKit/Source/core/editing/SelectionAdjuster.cpp
@@ -208,7 +208,13 @@ void SelectionAdjuster::adjustSelectionInDOMTree(
   const Position& extent = toPositionInDOMTree(selectionInFlatTree.extent());
 
   if (isCrossingShadowBoundaries(selectionInFlatTree)) {
-    *selection = createVisibleSelectionDeprecated(base, extent);
+    DCHECK(base.document());
+
+    // TODO(xiaochengh): The use of updateStyleAndLayoutIgnorePendingStylesheets
+    // needs to be audited.  See http://crbug.com/590369 for more details.
+    base.document()->updateStyleAndLayoutIgnorePendingStylesheets();

[yosin_UTC9, 2016/10/06 04:42:49]

There are only two places, one is test. So, could you hoist
updateStyleAndLayout... to call sites and put DCHECK for clean layout here?

[Xiaocheng, 2016/10/06 05:11:04]

I could not... Last time (crrev.com/2389653002) the call site was hoisted to
FrameSelection::setSelectionAlgorithm, and then it caused a performance
regression (http://crbug.com/652301) and was reverted.

I don't think we can hoist it out of the |if|, as this is the only place
requiring clean layout during the execution of
SelectionEditor::setVisibleSelection (for both overloads).

Maybe I should add a note here about the regression.

[yosin_UTC9, 2016/10/06 05:18:06]

I see, please add TODO(yosin) about regression. My lazy canoncalization should
fix this.

+
+    *selection = createVisibleSelection(base, extent);
     return;
   }