FrameTree::Find only searches relative to local frames.

third_party/WebKit/public/web/WebLocalFrame.h

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef WebLocalFrame_h
 #define WebLocalFrame_h
 
 #include <set>
 #include "WebCompositionUnderline.h"
 #include "WebFrame.h"
@@ skipping 92 matching lines @@
   virtual void SetAutofillClient(WebAutofillClient*) = 0;
   virtual WebAutofillClient* AutofillClient() = 0;
   virtual void SetDevToolsAgentClient(WebDevToolsAgentClient*) = 0;
   virtual WebDevToolsAgent* DevToolsAgent() = 0;
 
   // Hierarchy ----------------------------------------------------------
 
   // Get the highest-level LocalFrame in this frame's in-process subtree.
   virtual WebLocalFrame* LocalRoot() = 0;
 
+  // Returns the frame identified by the given name.  This method supports
+  // pseudo-names like _self, _top, and _blank.  It traverses the entire
+  // browsing instance containing this frame, looking for a frame that matches

[dcheng, 2017/05/26 18:28:06]

"browsing instance" isn't quite true today, is it? We still have the fallback to
Page::ordinaryPages() AFAIK

[Łukasz Anforowicz, 2017/05/26 20:04:05]

I've added a disclaimer to the comment above.  Right now the disclaimer talks
about https://crbug.com/718489, but in the follow-up CL I'll tweak the
disclaimer to talk about WebFrameClient::FindFrame extensibility point.

[dcheng, 2017/05/26 20:16:19]

Small nit: the web name for this is "same unit of related browsing contexts", so
probably that's the terminology we should use here.

Overall thoughts: I would just have this say what it does right now (falls back
to looking at all frame trees of WebViews in the same process, with no
consideration of browsing instance) and add a TODO for what we want (we want it
to consider only frames in the same unit of related browsing contexts).

That way, it's clear how this currently behaves and how it will behave in the
future.

[Łukasz Anforowicz, 2017/05/26 20:51:19]

In the end I've stopped talking about "browsing instance" / "unit of related
browsing contexts" and just said that FindFrameByName "performs the same kind of
lookup that |window.open(..., name)| would".  This also let's me avoid having to
explain how the browsing instance boundaries are sometimes violated by
extensions (or by current process-wide lookup).

+  // the given name.
+  virtual WebFrame* FindFrameByName(const WebString& name) = 0;
+
   // Navigation Ping --------------------------------------------------------
 
   virtual void SendPings(const WebURL& destination_url) = 0;
 
   // Navigation ----------------------------------------------------------
 
   // Runs beforeunload handlers for this frame and returns the value returned
   // by handlers.
   // Note: this may lead to the destruction of the frame.
   virtual bool DispatchBeforeUnloadEvent(bool is_reload) = 0;
@@ skipping 442 matching lines @@
   // to call these on a WebLocalFrame.
   bool IsWebLocalFrame() const override = 0;
   WebLocalFrame* ToWebLocalFrame() override = 0;
   bool IsWebRemoteFrame() const override = 0;
   WebRemoteFrame* ToWebRemoteFrame() override = 0;
 };
 
 }  // namespace blink
 
 #endif  // WebLocalFrame_h