Implementation of device metrics emulation in render view.

content/common/view_messages.h

 // Copyright (c) 2012 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.
 
 // IPC messages for page rendering.
 // Multiply-included message file, hence no include guard.
 
 #include "base/memory/shared_memory.h"
 #include "base/process/process.h"
 #include "base/strings/string16.h"
@@ skipping 1650 matching lines @@
 IPC_STRUCT_BEGIN(ViewHostMsg_BeginPinch_Params)
   IPC_STRUCT_MEMBER(bool, zoom_in)
   IPC_STRUCT_MEMBER(int32, pixels_to_move)
   IPC_STRUCT_MEMBER(int32, anchor_x)
   IPC_STRUCT_MEMBER(int32, anchor_y)
 IPC_STRUCT_END()
 
 IPC_MESSAGE_ROUTED1(ViewHostMsg_BeginPinch,
                     ViewHostMsg_BeginPinch_Params /* params */)
 
+// Sent to make host accept image of a certain expected size instead of current
+// view size. This is essential to emulate screen metrics, where render widget
+// renders only a part of the hosting view. Empty |size| means that default view
+// size is expected.
+IPC_MESSAGE_ROUTED1(ViewHostMsg_SetExpectedSize,

[aelias_OOO_until_Jul13, 2013/09/27 22:51:14]

Hmm, this SetExpectedSize thing still feels wrong.  The problem is that resize
logic is already complex and deadlock-prone and I don't want to make it worse,
and the override is also kind of unprincipled.  Thinking about it further, I
think this can be solved entirely on the render process side with negligible new
code after all.

You can set the main-frame root layer to masksToBounds in
RenderLayerCompositor::ensureRootLayer() when your metrics emulation is active
(or even better, just make it masksToBounds for all platforms except Android
using a WebSetting), so that the content isn't drawn outside Blink's idea of the
viewport.  Then just give the true physical_backing_size to CC but the emulated
DIP size to Blink.  You would also need to tell renderer CC that the background
color is white if that's not already the case, so that it fills in the area
that's clipped out.

I think that should work.  My only concern is that the tile management logic may
get a bit confused by the overly large physical backing size, but I think it
probably won't cause any visible issues, and we have to make that work for OS X
overscroll anyway.  Let me know if you run into any problems with this approach.

[dgozman, 2013/10/01 06:22:16]

This works when setting childrenTransform instead of transform on
rootGraphicsLayer. Added "rootLayerClipsContent" web setting.

+                    gfx::Size /* size */)
+
 IPC_MESSAGE_ROUTED0(ViewHostMsg_Focus)
 IPC_MESSAGE_ROUTED0(ViewHostMsg_Blur)
 
 // Message sent from renderer to the browser when focus changes inside the
 // webpage. The parameter says whether the newly focused element needs
 // keyboard input (true for textfields, text areas and content editable divs).
 IPC_MESSAGE_ROUTED1(ViewHostMsg_FocusedNodeChanged,
     bool /* is_editable_node */)
 
 IPC_MESSAGE_ROUTED1(ViewHostMsg_SetCursor,
@@ skipping 706 matching lines @@
 // synchronously (see crbug.com/120597). This IPC message sends the character
 // bounds after every composition change to always have correct bound info.
 IPC_MESSAGE_ROUTED2(ViewHostMsg_ImeCompositionRangeChanged,
                     gfx::Range /* composition range */,
                     std::vector<gfx::Rect> /* character bounds */)
 #endif
 
 // Adding a new message? Stick to the sort order above: first platform
 // independent ViewMsg, then ifdefs for platform specific ViewMsg, then platform
 // independent ViewHostMsg, then ifdefs for platform specific ViewHostMsg.