Fix FlatTreeTraversal for a slot in a document tree

third_party/WebKit/Source/core/dom/shadow/FlatTreeTraversal.cpp

 /*
  * Copyright (C) 2012 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Neither the name of Google Inc. nor the names of its
@@ skipping 41 matching lines @@
                                          direction);
   }
   return resolveDistributionStartingAt(direction == TraversalDirectionForward
                                            ? node.firstChild()
                                            : node.lastChild(),
                                        direction);
 }
 
 Node* FlatTreeTraversal::resolveDistributionStartingAt(
     const Node* node,
     TraversalDirection direction) {

[yosin_UTC9, 2016/10/14 01:44:40]

BTW, how about using template function to avoid |TraversalDirection| with
ternary operator in if-condition and for-condition?

[hayato, 2016/10/14 03:05:35]

That could be possible. We should do it in another CL.

   if (!node)
     return nullptr;
   for (const Node* sibling = node; sibling;
        sibling = (direction == TraversalDirectionForward
                       ? sibling->nextSibling()
                       : sibling->previousSibling())) {
     if (isHTMLSlotElement(*sibling)) {
       const HTMLSlotElement& slot = toHTMLSlotElement(*sibling);
+      if (slot.isInDocumentTree()) {
+        const_cast<HTMLSlotElement&>(slot)

[yosin_UTC9, 2016/10/14 01:44:40]

Can we integrate collecting distributed nodes into update distribution phase?

[hayato, 2016/10/14 03:05:35]

Yeah, that is one of TODO items. That could not be done easily.
You can find TODO description in this file. :)

+            .updateDistributedNodesInDocumentTree();

[yosin_UTC9, 2016/10/14 01:44:40]

It is better that using |collectDistributionNodesInDocumentTree()| or another
which implies populating |m_distributedNoes| instead of
|updateDistributedNodesInDocumentTree()|, since |updateXXX()| looks like
updating flat tree and term "update" is too generic(?).
Or, |updateDistributedNodeCache()| to make using |const_cast| to feel sane?

[hayato, 2016/10/14 03:05:35]

I think using *update* is consistent within this file.

+      }
       if (Node* found = (direction == TraversalDirectionForward
                              ? slot.firstDistributedNode()
                              : slot.lastDistributedNode()))
         return found;
       continue;
     }
     if (node->isInV0ShadowTree())
       return v0ResolveDistributionStartingAt(*sibling, direction);
     return const_cast<Node*>(sibling);
   }
@@ skipping 297 matching lines @@
 
 Node& FlatTreeTraversal::lastWithinOrSelf(const Node& node) {
   assertPrecondition(node);
   Node* lastDescendant = lastWithin(node);
   Node& result = lastDescendant ? *lastDescendant : const_cast<Node&>(node);
   assertPostcondition(&result);
   return result;
 }
 
 }  // namespace blink