Support plumbing LatencyInfo through the old software path.

content/browser/renderer_host/render_widget_host_view_gtk.cc

Index: content/browser/renderer_host/render_widget_host_view_gtk.cc
diff --git a/content/browser/renderer_host/render_widget_host_view_gtk.cc b/content/browser/renderer_host/render_widget_host_view_gtk.cc
index 5279631d53246a0591be181f6fa575ef86ca8765..240d9092e1418269ecbebed3b0730ea4f8be9ea7 100644
--- a/content/browser/renderer_host/render_widget_host_view_gtk.cc
+++ b/content/browser/renderer_host/render_widget_host_view_gtk.cc
@@ -853,8 +853,10 @@ void RenderWidgetHostViewGtk::ImeCompositionRangeChanged(
 void RenderWidgetHostViewGtk::DidUpdateBackingStore(
     const gfx::Rect& scroll_rect,
     const gfx::Vector2d& scroll_delta,
-    const std::vector<gfx::Rect>& copy_rects) {
+    const std::vector<gfx::Rect>& copy_rects,
+    const ui::LatencyInfo& latency_info) {
   TRACE_EVENT0("ui::gtk", "RenderWidgetHostViewGtk::DidUpdateBackingStore");
+  software_latency_info_.MergeWith(latency_info);
 
   if (is_hidden_)
     return;
@@ -1211,6 +1213,9 @@ void RenderWidgetHostViewGtk::Paint(const gfx::Rect& damage_rect) {
       // recorded.
       web_contents_switch_paint_time_ = base::TimeTicks();
     }
+    software_latency_info_.swap_timestamp = base::TimeTicks::HighResNow();
+    render_widget_host->FrameSwapped(software_latency_info_);

[piman, 2013/06/07 02:51:30]

So compared to the hardware path, this can somewhat misleading because this just
sent the command to X but we don't know when it'll be on screen really.
It's probably ok as a first step, but have we measured the delta? so that we can
normalize the typical latency expectation when comparing HW vs SW.

[jbauman, 2013/06/07 07:28:24]

Well, even in the hardware path we don't know when the actual swap to the screen
occurs - particularly when there's an external compositor.

I generally haven't worried too much about the precise amount of latency that we
can't measure, because that's very system-dependent (touchpad controller
latency, compositing window manager vs. not) and there's not that much we can do
about it. However, I think it wouldn't be hard to compare against our
camera-based measurements to see how much is being missed here.

+    software_latency_info_.Clear();
   } else {
     if (window)
       gdk_window_clear(window);