Browser tests for starting a drag-and-drop out of an OOPIF.

content/public/test/test_frame_navigation_observer.h

Index: content/public/test/test_frame_navigation_observer.h
diff --git a/content/test/test_frame_navigation_observer.h b/content/public/test/test_frame_navigation_observer.h
similarity index 93%
rename from content/test/test_frame_navigation_observer.h
rename to content/public/test/test_frame_navigation_observer.h
index 9e40ae71fc9c2bbc722c377e9f462b59a1030420..cf36fbbf89f1b3ff48cae06be40b43c6e6ecd278 100644
--- a/content/test/test_frame_navigation_observer.h
+++ b/content/public/test/test_frame_navigation_observer.h
@@ -17,8 +17,7 @@
 class GURL;
 
 namespace content {
-class FrameTreeNode;
-class WebContents;
+class RenderFrameHost;
 
 // For content_browsertests, which run on the UI thread, run a second

[ncarter (slow), 2016/11/11 20:56:36]

This comment needs to be updated slightly, since it's now safe to use this
outside of content_browsertests

[Łukasz Anforowicz, 2016/11/14 17:53:22]

Ooops.  Done.

I got rid of the UI thread comment, but to make up for it added DCHECKs to the
implementation.

 // MessageLoop and quit when the navigation in a specific frame (and all of its
@@ -26,8 +25,8 @@ class WebContents;
 class TestFrameNavigationObserver : public WebContentsObserver {
  public:
   // Create and register a new TestFrameNavigationObserver which will track
-  // navigations performed in the specified |node| of the frame tree.
-  explicit TestFrameNavigationObserver(FrameTreeNode* node);
+  // navigations performed in the specified |frame|.

[ncarter (slow), 2016/11/11 20:56:36]

Navigation might actually destroy |frame| (if it transfers to a new RFH), so
this comment should make it clear that that's actually okay.

[Łukasz Anforowicz, 2016/11/14 17:53:22]

Done (+ tried to wordsmith in a way that distinguishes between "frame tree node"
VS "frame" / "RenderFrameHost").

+  explicit TestFrameNavigationObserver(RenderFrameHost* frame);

[ncarter (slow), 2016/11/11 20:56:36]

Could this be a ToRenderFrameHost? That would allow it to accept a FrameTreeNode
too, which is actually a more appropriate parameter.

[Łukasz Anforowicz, 2016/11/14 17:53:22]

Good idea.  Done.

 
   ~TestFrameNavigationObserver() override;