Add touchpad pinch zoom support to Blink

Source/web/WebViewImpl.cpp

Index: Source/web/WebViewImpl.cpp
diff --git a/Source/web/WebViewImpl.cpp b/Source/web/WebViewImpl.cpp
index 06e1c3efd3a9d6a137123b869811287b1502407b..6a8a33a707cb14caaa9297aab82a680ce13e6f6e 100644
--- a/Source/web/WebViewImpl.cpp
+++ b/Source/web/WebViewImpl.cpp
@@ -2140,8 +2140,29 @@ bool WebViewImpl::handleInputEvent(const WebInputEvent& inputEvent)
 
     // Unhandled touchpad gesture pinch events synthesize mouse wheel events.
     if (inputEvent.type == WebInputEvent::GesturePinchUpdate) {
-        if (handleSyntheticWheelFromTouchpadPinchEvent(static_cast<const WebGestureEvent&>(inputEvent)))
+        const WebGestureEvent& pinchEvent = static_cast<const WebGestureEvent&>(inputEvent);
+
+        // First, synthesize a Windows-like wheel event to send to any handlers that may exist.
+        if (handleSyntheticWheelFromTouchpadPinchEvent(pinchEvent))
             return true;
+
+        // If the event remains unhandled, use it pinch the viewport.
+        const float magnifyDelta = pinchEvent.data.pinchUpdate.scale;
+        const float oldPageScale = pageScaleFactor();
+        const float newPageScale = clampPageScaleFactorToLimits(magnifyDelta * oldPageScale);
+
+        // Keep the center-of-pinch anchor specified by (x, y) in a stable
+        // position over the course of the magnify.
+        const FloatPoint anchor(pinchEvent.x, pinchEvent.y);

[bokan, 2015/02/03 00:45:12]

Drive-by: Could you please put this logic inside PinchViewport in a method that
takes the magnifyDelta and anchor?

[ccameron, 2015/02/03 02:29:08]

I don't see where the scale factor limits are available in PinchViewport, so I
had the function take the new scale factor, and the caller compute the two (the
caller can early-out and not return true if they are different). LMK if I missed
a way to do this entirely in PinchViewport.

[aelias_OOO_until_Jul13, 2015/02/03 02:37:14]

You can plumb in the information from WebViewImpl by adding an extra method to
ChromeClientImpl (which you can see in this file is reachable from FrameHost). 
I think it would be nicer to have the PinchViewport method take a delta.

[bokan, 2015/02/03 03:05:16]

Right, sorry, forgot about that. You can take Alexandre's suggestion above. 

[ccameron, 2015/02/03 17:03:28]

Done (yeah, I prefer this too -- thanks for pointing out how we get back to
WebViewImpl from PinchViewport).

+        FloatPoint anchorAtOldScale = anchor.scaledBy(1.f / oldPageScale);
+        FloatPoint anchorAtNewScale = anchor.scaledBy(1.f / newPageScale);
+        FloatPoint newLocation(pinchViewportOffset() + anchorAtOldScale - anchorAtNewScale);
+        setPageScaleFactorAndLocation(newPageScale, newLocation);

[aelias_OOO_until_Jul13, 2015/02/03 00:16:58]

Sorry, I noticed another issue.  Please scroll the normal FrameView scroll
offset first, and only apply the unused portion to the location in
setPageScaleFactorAndLocation.  That will match the current logic in
LayerTreeHostImpl::PinchGestureUpdate.  That will allow pinch zooming out to not
get clamped in a weird way, among other things.

[bokan, 2015/02/03 00:45:12]

Though almost dead, please be aware of the old pinch-path and guard necessary
calls behind the pinchVirtualViewportEnabled() flag. As-is
setPageScaleFactorAndLocation will ASSERT in the old-path. I'll be removing it
by the end of this quarter.

pinchViewportOffset() will return 0 in the old path which I think is fine in
this case.

[ccameron, 2015/02/03 02:29:08]

Done (in the PinchVirtualViewport version).

+
+        // TODO(ccameron): Always return that the event was handled. Once the pinch-zoom content scale
+        // handlers have been removed from content, this function may return false if the page scale does
+        // not change.
+        return true;
     }
 
     return false;