Implementing document.setRootScroller API for main thread scrolling.

third_party/WebKit/Source/core/page/scrolling/RootScroller.cpp

Index: third_party/WebKit/Source/core/page/scrolling/RootScroller.cpp
diff --git a/third_party/WebKit/Source/core/page/scrolling/RootScroller.cpp b/third_party/WebKit/Source/core/page/scrolling/RootScroller.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..e89c0fcc0f39dbfd9a2f37afcd5fb92c43c47e6e
--- /dev/null
+++ b/third_party/WebKit/Source/core/page/scrolling/RootScroller.cpp
@@ -0,0 +1,159 @@
+// Copyright 2016 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.
+
+#include "core/page/scrolling/RootScroller.h"
+
+#include "core/dom/Element.h"
+#include "core/frame/FrameHost.h"
+#include "core/frame/FrameView.h"
+#include "core/layout/LayoutBox.h"
+#include "core/page/Page.h"
+#include "core/page/scrolling/ViewportScrollCallback.h"
+#include "core/paint/PaintLayerScrollableArea.h"
+#include "platform/scroll/ScrollableArea.h"
+
+namespace blink {
+
+namespace {
+
+Document* topDocument(FrameHost* frameHost)
+{
+    DCHECK(frameHost);
+    if (!frameHost->page().mainFrame())
+        return nullptr;
+
+    DCHECK(frameHost->page().mainFrame()->isLocalFrame());

[esprehn, 2016/05/05 17:29:46]

There's no reason to go through the FrameHost like this all the time, just use
the localRoot()

[bokan, 2016/05/05 21:19:48]

Acknowledged.

+    return toLocalFrame(frameHost->page().mainFrame())->document();
+}
+
+ScrollableArea* scrollableAreaFor(Element& element)
+{
+    if (!element.layoutObject() || !element.layoutObject()->isBox())
+        return nullptr;
+
+    LayoutBox* box = toLayoutBox(element.layoutObject());
+
+    if (box->isDocumentElement() && element.document().view())

[esprehn, 2016/05/05 17:29:46]

it's impossible for view() to be null here if the LayoutObject is not null

[bokan, 2016/05/05 21:19:49]

Acknowledged.

+        return element.document().view()->getScrollableArea();
+
+    return static_cast<PaintInvalidationCapableScrollableArea*>(

[esprehn, 2016/05/05 17:29:46]

this needs a toPaintInvalidationCapableScrollableArea() casting function and a
virtual check.

[bokan, 2016/05/05 21:19:49]

This is an upcast though. LayoutBoxModelObject::getScrollableArea()
(confusingly) returns a PaintLayerScrollableArea which derives from
PaintInvalidationCapableScrollableArea.

+        box->getScrollableArea());
+}
+
+} // namespace
+
+RootScroller::RootScroller(FrameHost& frameHost)
+    : m_frameHost(&frameHost)

[esprehn, 2016/05/05 17:29:46]

use a Frame, not the FrameHost.

[bokan, 2016/05/05 21:19:48]

Yah, I hadn't thought this through and had a 1 RootScroller per Page. I think
having 1:1 RootScroller:Frame is a better idea, with only the main frame's
RootScroller applying viewport actions.

+{
+}
+
+DEFINE_TRACE(RootScroller)
+{
+    visitor->trace(m_frameHost);
+    visitor->trace(m_viewportApplyScroll);
+    visitor->trace(m_rootScroller);
+}
+
+bool RootScroller::set(Element& newRootScroller)
+{
+    if (!isValid(newRootScroller))
+        return false;
+
+    DCHECK(m_frameHost);
+
+    Document* document = topDocument(m_frameHost);
+    if (!document)
+        return false;
+
+    ScrollableArea* newRootScrollableArea = scrollableAreaFor(newRootScroller);
+    if (!newRootScrollableArea)
+        return false;
+
+    if (m_rootScroller)
+        m_rootScroller->removeApplyScroll();
+
+    m_rootScroller = &newRootScroller;
+
+    createApplyScrollIfNeeded();
+
+    // Ideally, the scrolling infrastructure would pass this to the callback but
+    // we don't get that today so we set it manually.
+    m_viewportApplyScroll->setScroller(*newRootScrollableArea);
+
+    // Installs the viewport scrolling callback (the "applyScroll" in Scroll
+    // Customization lingo) on the given element. This callback is
+    // responsible for viewport related scroll actions like top controls
+    // movement and overscroll glow as well as actually scrolling the element.
+    // Use disable-native-scroll since the ViewportScrollCallback needs to
+    // apply scroll actions before (TopControls) and after (overscroll)
+    // scrolling the element so it applies scroll to the element itself.
+    m_rootScroller->setApplyScroll(
+        m_viewportApplyScroll,
+        "disable-native-scroll");
+
+    return true;
+}
+
+Element* RootScroller::get() const
+{
+    if (!m_frameHost)
+        return nullptr;
+
+    return m_rootScroller;
+}
+
+void RootScroller::resetToDefault()
+{
+    if (!m_frameHost)
+        return;
+
+    Document* document = topDocument(m_frameHost);
+    if (!document)
+        return;
+
+    if (Element* defaultRootElement = document->documentElement())
+        set(*defaultRootElement);
+}
+
+void RootScroller::didUpdateTopDocumentLayout()
+{
+    if (m_rootScroller && isValid(*m_rootScroller))
+        return;
+
+    resetToDefault();
+}
+
+bool RootScroller::isValid(Element& element) const
+{
+    if (!m_frameHost)
+        return false;
+
+    if (element.document() != topDocument(m_frameHost))

[esprehn, 2016/05/05 17:29:46]

you want to hook Element::removedFrom not layout for this

[bokan, 2016/05/05 21:19:48]

There's other cases that make an Element invalid to be the root scroller. I.e.
if it's not a box.

I didn't add it in this patch, but our initial version of this proposal also has
that only viewport-filling elements (i.e. width:100%, height:100%) can be the
root scroller. This is to prevent strange effects as we increase/decrease the
clip height dynamically on the root scroller as we move the top controls and to
make overscroll glow simpler to implement. For those cases, I think we need to
be hooked into layout, right?

The way this is implemented right now, if the currently set root scroller
becomes invalid after being set, the root scroller will reset to the document
element. It sounds like maybe making it "sticky" would be more natural. i.e.
While its invalid, use the document element, but if it becomes valid again it
will be used. Does that sound better?

+        return false;
+
+    if (!element.isInTreeScope())

[esprehn, 2016/05/05 17:29:46]

the layoutObject check below is a proxy for this, there's no reason to check
both

[bokan, 2016/05/05 21:19:48]

Acknowledged.

+        return false;
+
+    if (!element.layoutObject())

[esprehn, 2016/05/05 17:29:46]

this check doesn't make sense, it means if you display: none the root scroller,
force a layout, and then display: block it again the scroller is reset. I don't
think this feature should work that way.

[bokan, 2016/05/05 21:19:48]

Yah, that was the intention, though I can see that it might be surprising. I'll
do what I mentioned above.

+        return false;
+
+    if (!element.layoutObject()->isLayoutBlockFlow()
+        && !element.layoutObject()->isLayoutIFrame())
+        return false;
+
+    return true;
+}
+
+void RootScroller::createApplyScrollIfNeeded()
+{
+    if (!m_viewportApplyScroll) {
+        TopControls& topControls = m_frameHost->topControls();
+        OverscrollController& overscrollController =
+            m_frameHost->overscrollController();
+        m_viewportApplyScroll =
+            new ViewportScrollCallback(topControls, overscrollController);
+    }
+}
+
+} // namespace blink