Fix user agent overridding with PlzNavigate.

content/browser/frame_host/navigator_delegate.h

 // Copyright 2013 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 CONTENT_BROWSER_FRAME_HOST_NAVIGATOR_DELEGATE_H_
 #define CONTENT_BROWSER_FRAME_HOST_NAVIGATOR_DELEGATE_H_
 
 #include "base/strings/string16.h"
 #include "content/public/browser/invalidate_type.h"
 #include "content/public/browser/navigation_controller.h"
@@ skipping 87 matching lines @@
   virtual WebContents* OpenURL(const OpenURLParams& params) = 0;
 
   // Returns whether to continue a navigation that needs to transfer to a
   // different process between the load start and commit.
   virtual bool ShouldTransferNavigation(bool is_main_frame_navigation);
 
   // Returns whether URLs for aborted browser-initiated navigations should be
   // preserved in the omnibox.  Defaults to false.
   virtual bool ShouldPreserveAbortedURLs();
 
+  // Returns the overriden user agent string if it's set.
+  virtual const std::string& GetNavigationUserAgentOverride();

[nasko, 2017/03/13 18:47:12]

nit: The "Navigation" part of the name doesn't really add much meaning to the
method. We will use the same user agent override for subresources too, so making
it shorter and easier to read is a win IMHO.

[jam, 2017/03/13 19:32:40]

yeah I did it this way because WebContents, which implements, already has a
GetUserAgentOverride. I know I could leave it as is, and the one override will
work for both interfaces, but I find that it's more confusing that way for
readability. wdyt?

[nasko, 2017/03/13 20:35:41]

We have a few cases where the same pure virtual method exist on this delegate
interface and on WebContentsImpl. As long as the function is exactly the same in
both cases, we keep them exactly the same. An example of that is the OpenURL
call above. This case looks identical, so I personally don't find it more
confusing. It is one method on the same object, reachable through different
interfaces.

[jam, 2017/03/13 20:38:25]

ah, if there's already precedent with this interface I'll do the same, thanks.

Personally I didn't find this confusing but I got feedback in reviews before
that it it, so happy to go with this approach :)

+
   // A RenderFrameHost in the specified |frame_tree_node| started loading a new
   // document. This correponds to Blink's notion of the throbber starting.
   // |to_different_document| will be true unless the load is a fragment
   // navigation, or triggered by history.pushState/replaceState.
   virtual void DidStartLoading(FrameTreeNode* frame_tree_node,
                                bool to_different_document) {}
 
   // A document stopped loading. This corresponds to Blink's notion of the
   // throbber stopping.
   virtual void DidStopLoading() {}
@@ skipping 12 matching lines @@
   // wants to see passed to the corresponding URLRequest on the IO thread.
   // In the case of a navigation to an interstitial, no call will be made to the
   // embedder and |nullptr| is returned.
   virtual std::unique_ptr<NavigationUIData> GetNavigationUIData(
       NavigationHandle* navigation_handle);
 };
 
 }  // namspace content
 
 #endif  // CONTENT_BROWSER_FRAME_HOST_NAVIGATOR_DELEGATE_H_