Fix index used to join results of DOM.getDocument and CSS.getLayoutTreeAndStyles

third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp

Index: third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp
diff --git a/third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp b/third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp
index c02151fa90407b36ee8d2c2c3a97e1f41638a982..c61f5800baf916ec85bda88ba6fee43923337ed6 100644
--- a/third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp
+++ b/third_party/WebKit/Source/core/inspector/InspectorCSSAgent.cpp
@@ -2375,8 +2375,13 @@ int InspectorCSSAgent::getStyleIndexForNode(
       CSSComputedStyleDeclaration::create(node, true);
 
   Vector<String> style;
+  // WTF::HashMap refuses to allow you to insert an empty vector :(
+  style.append("dummy value");

[Sami, 2016/10/26 17:15:23]

Would it be hard to fix this in WTF::HashMap? (I wonder if it's by design or
not). Alternatively could we avoid inserting when we have nothing to insert?

[dgozman, 2016/10/26 17:24:39]

There should be a possibility with some magic boolean flags combination on the
traits, but I cannot advise which one...

[alex clarke (OOO till 29th), 2016/10/26 21:18:49]

Possibly, I'm not sure how though.  Honestly WTF::HashMap is a nightmare to use,
I spent an hour trying to avoid the ugly dummy value with no success :(

It would be trivial to fix this with std::unordered_map...

   for (const auto& pair : cssPropertyWhitelist) {
-    style.append(computedStyleInfo->getPropertyValue(pair.second));
+    String value = computedStyleInfo->getPropertyValue(pair.second);
+    if (value.isEmpty())
+      continue;

[dgozman, 2016/10/26 17:24:39]

If you skip values in resulting array, how does client match the values to the
properties they requested?

[alex clarke (OOO till 29th), 2016/10/26 21:18:49]

We send the key as well as the value, see:

styleProperties->addItem(protocol::CSS::CSSComputedStyleProperty::create()
 .setName(cssPropertyWhitelist[i - 1].first)
 .setValue(style[i])
.build());

+    style.append(value);
   }
 
   ComputedStylesMap::iterator it = styleToIndexMap.find(style);
@@ -2387,9 +2392,10 @@ int InspectorCSSAgent::getStyleIndexForNode(
   std::unique_ptr<protocol::Array<protocol::CSS::CSSComputedStyleProperty>>
       styleProperties =
           protocol::Array<protocol::CSS::CSSComputedStyleProperty>::create();
-  for (size_t i = 0; i < style.size(); i++) {
+  // Skip the dummy value...
+  for (size_t i = 1; i < style.size(); i++) {
     styleProperties->addItem(protocol::CSS::CSSComputedStyleProperty::create()
-                                 .setName(cssPropertyWhitelist[i].first)
+                                 .setName(cssPropertyWhitelist[i - 1].first)
                                  .setValue(style[i])
                                  .build());
   }
@@ -2434,10 +2440,12 @@ void InspectorCSSAgent::visitLayoutTreeNodes(
     if (!layoutObject)
       continue;
 
-    int backendNodeId = DOMNodeIds::idForNode(node);
+    int nodeId = m_domAgent->boundNodeId(node);

[dgozman, 2016/10/26 17:24:39]

This requires the node to be already pushed to the client. Is that what you
want?

[alex clarke (OOO till 29th), 2016/10/26 21:18:49]

What I want to do is join the results of DOM.getDocument (with depth of -1) and
CSS.getLayoutTreeAndStyles 
With that in mind this seems ok.

Still I don't really mind how we do that, and if theres a better way I'd be
happy to explore it.

+    if (!nodeId)
+      continue;
     std::unique_ptr<protocol::CSS::LayoutTreeNode> layoutTreeNode =
         protocol::CSS::LayoutTreeNode::create()
-            .setBackendNodeId(backendNodeId)
+            .setNodeId(nodeId)
             .setStyleIndex(getStyleIndexForNode(
                 node, cssPropertyWhitelist, styleToIndexMap, computedStyles))
             .setBoundingBox(buildRectForFloatRect(