Implement NavigationThrottle::BLOCK_REQUEST_AND_COLLAPSE.

content/browser/frame_host/frame_tree_node.h

Index: content/browser/frame_host/frame_tree_node.h
diff --git a/content/browser/frame_host/frame_tree_node.h b/content/browser/frame_host/frame_tree_node.h
index ac46de23c5737bedbf5e9fb7a2f4839ef1c1ab23..8c4c5f29344391148fc6bd5fa914781e2f8aef8d 100644
--- a/content/browser/frame_host/frame_tree_node.h
+++ b/content/browser/frame_host/frame_tree_node.h
@@ -147,6 +147,20 @@ class CONTENT_EXPORT FrameTreeNode {
     return has_committed_real_load_;
   }
 
+  // Returns whether the frame's owner element in the parent document is
+  // collapsed, that is, removed from the layout as if it did not exist, as per
+  // request by the embedder (of the content/ layer).
+  bool is_collapsed() const { return is_collapsed_; }
+
+  // Sets whether to collapse the frame's owner element in the parent document,
+  // that is, to remove it from the layout as if it did not exist, as per
+  // request by the embedder (of the content/ layer). Cannot be called for main
+  // frames.
+  //
+  // This only has an effect for <iframe> owner elements, and is a no-op when
+  // called on sub-frames hosted in <frame>, <object>, and <embed> elements.

[dcheng, 2017/05/10 06:59:46]

Out of curiosity, why limited only to iframe?

[engedy, 2017/05/15 13:35:51]

For historical reasons, we ignore `display: none` for <frame> elements (starting
with the initial revision [1] until today [2]), so I wanted to be consistent
with that and implement collapsing for <iframe> elements only.

Not sure what the original rationale was, but this makes some sense considering
that a <frameset> represents a complete partitioning of a page, so collapsing
one <frame> would affect all others in ill-defined ways.
 
[1]:
https://chromium.googlesource.com/chromium/src/+blame/d869b93fe74f4d6cb2dd6f6...
[2]:
https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/html/HTML...

[dcheng, 2017/05/18 09:35:55]

Probably makes sense to do this for <object> and <embed> though doesn't it?
Otherwise, switching to either one of those would trivially circumvent this. I'm
OK with poking at this in a followup, but it seems inconsistent to only do it
for one type of embeddable frame.

[engedy, 2017/05/18 09:40:06]

That's a good point and I agree. Let me address that in a follow-up.

+  void SetCollapsed(bool collapsed);
+
   // Returns the origin of the last committed page in this frame.
   // WARNING: To get the last committed origin for a particular
   // RenderFrameHost, use RenderFrameHost::GetLastCommittedOrigin() instead,
@@ -398,6 +412,9 @@ class CONTENT_EXPORT FrameTreeNode {
   // about:blank page.
   bool has_committed_real_load_;
 
+  // Whether the frame's owner element in the parent document is collapsed.
+  bool is_collapsed_;
+
   // Track information that needs to be replicated to processes that have
   // proxies for this frame.
   FrameReplicationState replication_state_;