Pipe the devTools FrameId from blink into the browser for headless

headless/lib/browser/headless_web_contents_impl.h

 // Copyright 2015 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 HEADLESS_LIB_BROWSER_HEADLESS_WEB_CONTENTS_IMPL_H_
 #define HEADLESS_LIB_BROWSER_HEADLESS_WEB_CONTENTS_IMPL_H_
 
 #include <list>
 #include <memory>
 #include <string>
@@ skipping 82 matching lines @@
   HeadlessBrowserImpl* browser() const;
   HeadlessBrowserContextImpl* browser_context() const;
 
   void set_window_tree_host(std::unique_ptr<HeadlessWindowTreeHost> host) {
     window_tree_host_ = std::move(host);
   }
   HeadlessWindowTreeHost* window_tree_host() const {
     return window_tree_host_.get();
   }
 
+  // Returns the devtools frame id corresponding to the |frame_tree_node_id|, if
+  // any.  Note this relies on an IPC sent from blink during navigation.
+  std::string GetUntrustedDevToolsFrameIdForFrameTreeNodeId(
+      int process_id,
+      int frame_tree_node_id) const;
+
+  int GetRenderProcessId() const;

[Charlie Reis, 2017/04/26 19:52:56]

A WebContents can have multiple processes, so we shouldn't introduce a new API
suggesting it has only one.  (WebContents::GetRenderProcessHost() is on its way
out in https://crbug.com/666525.)

We should at least rename it to GetMainFrameProcessId() (and implement it using
GetMainFrame()->GetProcess()->GetID()).

Or actually, this appears to only be used in tests?  Maybe there's a way to
avoid exposing it, or make it specific to tests?

[alex clarke (OOO till 29th), 2017/04/27 08:49:44]

Done.

+
  private:
   // Takes ownership of |web_contents|.
   HeadlessWebContentsImpl(content::WebContents* web_contents,
                           HeadlessBrowserContextImpl* browser_context);
 
   void InitializeScreen(const gfx::Size& initial_size);
   using MojoService = HeadlessWebContents::Builder::MojoService;
 
   std::unordered_map<content::RenderFrameHost*, std::string>
       render_frame_host_to_devtools_agent_host_id_;
   std::unordered_map<std::string, int> devtools_agent_id_to_frame_tree_node_id_;
 
   class Delegate;
   std::unique_ptr<Delegate> web_contents_delegate_;
   std::unique_ptr<HeadlessWindowTreeHost> window_tree_host_;
   std::unique_ptr<content::WebContents> web_contents_;
   scoped_refptr<content::DevToolsAgentHost> agent_host_;
   std::list<MojoService> mojo_services_;
   std::unique_ptr<HeadlessTabSocketImpl> headless_tab_socket_;
 
   HeadlessBrowserContextImpl* browser_context_;      // Not owned.
   content::RenderProcessHost* render_process_host_;  // Not owned.

[Charlie Reis, 2017/04/26 19:52:55]

This has a similar problem.  It looks like HeadlessWebContentsImpl could have
lots of problems when out-of-process iframes are present on a page (e.g., not
observing subframe processes)?

[alex clarke (OOO till 29th), 2017/04/27 08:49:44]

Acknowledged.

 
   using ObserverMap =
       std::unordered_map<HeadlessWebContents::Observer*,
                          std::unique_ptr<WebContentsObserverAdapter>>;
   ObserverMap observer_map_;
 
   DISALLOW_COPY_AND_ASSIGN(HeadlessWebContentsImpl);
 };
 
 }  // namespace headless
 
 #endif  // HEADLESS_LIB_BROWSER_HEADLESS_WEB_CONTENTS_IMPL_H_