Move download messages from Renderer to Frame filter.

third_party/WebKit/Source/web/WebLocalFrameImpl.cpp

Index: third_party/WebKit/Source/web/WebLocalFrameImpl.cpp
diff --git a/third_party/WebKit/Source/web/WebLocalFrameImpl.cpp b/third_party/WebKit/Source/web/WebLocalFrameImpl.cpp
index 0893b67b0d59e7013896a1dce6d7647599e72786..8782e4be1a1dc8f0dd50232f2857590db45fc7b3 100644
--- a/third_party/WebKit/Source/web/WebLocalFrameImpl.cpp
+++ b/third_party/WebKit/Source/web/WebLocalFrameImpl.cpp
@@ -120,6 +120,7 @@
 #include "core/frame/RemoteFrame.h"
 #include "core/frame/Settings.h"
 #include "core/frame/UseCounter.h"
+#include "core/frame/VisualViewport.h"
 #include "core/html/HTMLAnchorElement.h"
 #include "core/html/HTMLCollection.h"
 #include "core/html/HTMLFormElement.h"
@@ -1355,6 +1356,36 @@ bool WebLocalFrameImpl::selectionStartHasSpellingMarkerFor(int from, int length)
     return frame()->spellChecker().selectionStartHasSpellingMarkerFor(from, length);
 }
 
+void WebLocalFrameImpl::copyImageAt(const WebPoint& posInViewport)
+{
+    HitTestResult result = hitTestResultForVisualViewportPos(posInViewport);
+    if (!isHTMLCanvasElement(result.innerNodeOrImageMapImage()) && result.absoluteImageURL().isEmpty()) {
+        // There isn't actually an image at these coordinates.  Might be because
+        // the window scrolled while the context menu was open or because the page
+        // changed itself between when we thought there was an image here and when
+        // we actually tried to retreive the image.
+        //
+        // FIXME: implement a cache of the most recent HitTestResult to avoid having
+        //        to do two hit tests.
+        return;
+    }
+
+    frame()->editor().copyImage(result);
+}
+
+void WebLocalFrameImpl::saveImageAt(const WebPoint& posInViewport)
+{
+    Node* node = hitTestResultForVisualViewportPos(posInViewport).innerNodeOrImageMapImage();
+    if (!node || !(isHTMLCanvasElement(*node) || isHTMLImageElement(*node)))
+        return;
+
+    String url = toElement(*node).imageSourceURL();
+    if (!KURL(KURL(), url).protocolIsData())
+        return;
+
+    m_client->saveImageFromDataURL(url);
+}
+
 WebString WebLocalFrameImpl::layerTreeAsText(bool showDebugInfo) const
 {
     if (!frame())
@@ -1758,6 +1789,15 @@ void WebLocalFrameImpl::loadJavaScriptURL(const KURL& url)
         frame()->loader().replaceDocumentWhileExecutingJavaScriptURL(scriptResult, ownerDocument);
 }
 
+HitTestResult WebLocalFrameImpl::hitTestResultForVisualViewportPos(const IntPoint& posInViewport)
+{
+    IntPoint rootFramePoint(frame()->host()->visualViewport().viewportToRootFrame(posInViewport));

[dcheng, 2016/06/08 16:00:35]

+bokan, is this the right thing to do for a local frame with a remote parent?

[bokan, 2016/06/08 16:13:27]

Hmm...I'm not sure how VisualViewport behaves in OOPIF scenarios. Suppose we
pinch-zoom in and move around such that the non-OOPIF VisualViewport would have
a scaleFactor and scrollOffset. In OOPIF, does the VisualViewport in the child
process reflect the scaleFactor and scrollOffset seen by the user? Or does each
process have its own independent VisualViewport?

If the former, and we've done the work to sync the VisualViewport values between
processes, than this would be ok w.r.t. passed in points that are actually in
"viewport coordinates". That is, relative to the window origin on desktop.

VisualViewport aside, the local frame doesn't have access to the entire Widget
hierarchy does it? I think rootFrameToContents below will be problematic since
it needs to walk through all the parent widgets and adjust by their locations
and scroll offsets. At a quick glance, it doesn't look like that code has been
OOPIF'd so remote frames will be problematic there.

[alexmos, 2016/06/08 16:40:33]

Note that I recently did some work to sync the VisualViewport size in
https://codereview.chromium.org/1834913002/.  I'm not sure about scaleFactor and
scrollOffset though.

[brettw, 2016/06/09 23:41:01]

I didn't make any changes here. If I should change something please just let me
know. I wrote this code based on what was happening elsewhere and what I could
figure out that worked.

+    IntPoint docPoint(frame()->view()->rootFrameToContents(rootFramePoint));
+    HitTestResult result = frame()->eventHandler().hitTestResultAtPoint(docPoint, HitTestRequest::ReadOnly | HitTestRequest::Active);
+    result.setToShadowHostIfInUserAgentShadowRoot();
+    return result;
+}
+
 static void ensureFrameLoaderHasCommitted(FrameLoader& frameLoader)
 {
     // Internally, Blink uses CommittedMultipleRealLoads to track whether the