Support vsync-aligned wheel event dispatch

Support vsync-aligned wheel event dispatch

When begin-frame scheduling is enabled, buffer wheel events until a
BeginFrame-triggered flush. Also factor out wheel event logic from
the InputRouter into a standalone WheelEventQueue.

BUG=394562

[jdduke (slow), 2015-10-21 17:27:02]

jdduke@chromium.org changed reviewers:
+ tdresser@chromium.org

[jdduke (slow), 2015-10-21 17:27:02]

Tim, any thoughts on this change? (particularly the content/ bits, the ui/bits
are landing separately in crrev.com/1414533003). It's just a rough prototype, so
looking more for high level feedback.

I have a simpler patch (crrev.com/1391743004) that buffers both scroll and wheel
events, but it does it in such a way that we might incur an extra frame of
latency for every event. Don't think we want that, even if it's conceptually a
simpler model.

[tdresser, 2015-10-22 13:53:53]

This looks like a solid approach to me, at a high level.

It seems like there's a slim possibility that more code sharing between the
touchpad and touchscreen code would improve matters, but it seems pretty
unlikely to me.

I'm happy with going with this approach.

https://chromiumcodereview.appspot.com/1405613002/diff/100001/content/browser...
File content/browser/renderer_host/input/wheel_event_queue.h (right):

https://chromiumcodereview.appspot.com/1405613002/diff/100001/content/browser...
content/browser/renderer_host/input/wheel_event_queue.h:53: return
mouse_wheel_pending_ || coalesced_wheel_events_.empty();
Why does mouse_wheel_pending_ imply that the queue is empty?

[jdduke (slow), 2015-10-22 15:15:05]

https://chromiumcodereview.appspot.com/1405613002/diff/100001/content/browser...
File content/browser/renderer_host/input/wheel_event_queue.h (right):

https://chromiumcodereview.appspot.com/1405613002/diff/100001/content/browser...
content/browser/renderer_host/input/wheel_event_queue.h:53: return
mouse_wheel_pending_ || coalesced_wheel_events_.empty();
On 2015/10/22 13:53:53, tdresser wrote:
> Why does mouse_wheel_pending_ imply that the queue is empty?

It doesn't, this is a bug :)

[jdduke (slow), 2015-10-22 15:17:12]

On 2015/10/22 13:53:53, tdresser wrote:
> It seems like there's a slim possibility that more code sharing between the
> touchpad and touchscreen code would improve matters, but it seems pretty
> unlikely to me.

Yeah, it is a bit unfortunate. I tried a few ways to share code, but the
GestureEventQueue has some nuances that make things tricky. In particularly, it
allows 2 events to be in-flight at the same time (scroll + pinch), which
complicates any kind of shared buffering data structure.

[tdresser, 2016-11-28 15:27:31]

Description was changed from

==========
Support vsync-aligned wheel event dispatch

When begin-frame scheduling is enabled, buffer wheel events until a
BeginFrame-triggered flush. Also factor out wheel event logic from
the InputRouter into a standalone WheelEventQueue.

BUG=394562
==========

to

==========
Support vsync-aligned wheel event dispatch

When begin-frame scheduling is enabled, buffer wheel events until a
BeginFrame-triggered flush. Also factor out wheel event logic from
the InputRouter into a standalone WheelEventQueue.

BUG=394562
==========