Fix main thread top controls scrolling to mirror CC.

third_party/WebKit/Source/core/frame/VisualViewport.cpp

 /*
  * Copyright (C) 2013 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above
@@ skipping 456 matching lines @@
 IntPoint VisualViewport::maximumScrollPosition() const
 {
     return flooredIntPoint(maximumScrollPositionDouble());
 }
 
 DoublePoint VisualViewport::maximumScrollPositionDouble() const
 {
     if (!mainFrame())
         return IntPoint();
 
     // FIXME: We probably shouldn't be storing the bounds in a float. crbug.com/422331.

[majidvp, 2016/04/01 17:09:36]

The actual bug for this TODO is marked as WontFix. Does this need to be updated?

[bokan, 2016/04/05 17:11:40]

Good catch, this was the wrong bug. Updated.

     FloatSize frameViewSize(contentsSize());
 
     if (m_topControlsAdjustment) {
         float minScale = frameHost().pageScaleConstraintsSet().finalConstraints().minimumScale;
         frameViewSize.expand(0, m_topControlsAdjustment / minScale);
     }
 
     frameViewSize.scale(m_scale);
     frameViewSize = FloatSize(flooredIntSize(frameViewSize));
 
     FloatSize viewportSize(m_size);
-    viewportSize.expand(0, m_topControlsAdjustment);
+    viewportSize.expand(0, ceilf(m_topControlsAdjustment));

[majidvp, 2016/04/01 17:09:36]

Can you point out where in CC we are ceiling the top controls adjustment?

It is a bit odd that we ceil the top controls adjustment when computing 
viewportSize and floor it when computing frameViewSize. Ideally we 
only clamp it at the end of the computation. For example with a 
scale of 1 the current behaviour can adjust the max scroll offset
by one pixel though it was not necessary. 

Should we instead floor maxPosition instead?

[bokan, 2016/04/05 17:11:40]

https://code.google.com/p/chromium/codesearch#chromium/src/cc/trees/property_...

We ceil the bounds after adding the bounds_delta (bounds_delta is basically the
CC equivalent of topControlsAdjustment) when computing the clip size. I agree
it's odd that we're doing this but I'd rather match CC's quirks here as the CC
path is much better tested. We can adjust both sides in a follow up, though I'm
also weary of changing this code that isn't showing any obvious bugs.

 
     FloatSize maxPosition = frameViewSize - viewportSize;
     maxPosition.scale(1 / m_scale);
     return DoublePoint(maxPosition);
 }
 
 IntPoint VisualViewport::clampDocumentOffsetAtScale(const IntPoint& offset, float scale)
 {
     if (!mainFrame() || !mainFrame()->view())
         return IntPoint();
@@ skipping 243 matching lines @@
     } else if (graphicsLayer == m_rootTransformLayer) {
         name =  "Root Transform Layer";
     } else {
         ASSERT_NOT_REACHED();
     }
 
     return name;
 }
 
 } // namespace blink