1502529 - Coalesce touchmove

Status: RESOLVED FIXED

(Core :: DOM: UI Events & Focus Handling, enhancement, P2)

Description

[Olli Pettay [:smaug][[e-mail]]]

I think we should try to coalesce touchmoves similarly to mousemoves.
Chrome is doing that.


http://mozilla.pettay.fi/moztests/touchmove.html
The last line shows average touchmoves per animation frame.
Try using two fingers.

Something to consider here is how to handle
https://w3c.github.io/pointerevents/extension.html#extensions-to-the-pointerevent-interface

Comment 1

[Olli Pettay [:smaug][[e-mail]]]

Looks like on Windows things are a bit different. http://mozilla.pettay.fi/moztests/pointermove.html should work on Edge too.
We don't seem to dispatch so many events.

(but coalescing is needed on mobile at least.)

Comment 2

[Timea Cernea [:tbabos][inactive]]

Reproduced on Lenovo Yoga C630-13Q50 with Windows 10 Home (v1803) on latest Nightly aarch64 build.

Comment 3

[Markus Stange [:mstange]]

It seems like we would also need to implement getCoalescedEvents for touch events. At the moment this seems to be implemented only for pointer events from mouse move events. Do we send pointer events for touch events on Android?
Furthermore, Android has its own touch move event batching, but lets you access the "historical" events of a MotionEvent.

Comment 4

[Olli Pettay [:smaug][[e-mail]]]

Yes, pointer events are supported on multiprocess GeckoView.
(Fennec doesn't have pointere events)

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

One simple way to get at least the IPC level compression would be to add a new IPC message
RealTouchMoveEvent2 or some such. It would behave exactly like RealTouchMoveEvent.
Whenever sender changes the touch stream id of the touches it is sending, it would switch to the other RealTouchMoveEvent* message.
Then both the the messages could use compress in ipdl.

(That wouldn't affect to getCoalescedEvents, but that is a bit separate issue anyhow)

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

A variant of the original testcase showing that if JS listeners are slow, we end up queuing more touchmoves
http://mozilla.pettay.fi/moztests/touchmoveslow.html
http://mozilla.pettay.fi/moztests/touchmove2slow.html

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

ARM64 WIN https://treeherder.mozilla.org/#/jobs?repo=try&revision=3fe891039fd9bc89e7b6426395f646961f7b6186
Android https://treeherder.mozilla.org/#/jobs?repo=try&revision=a3cb6a0ddd0c16f92269c4210e01222f66a31fa5

Comment 8

[Olli Pettay [:smaug][[e-mail]]]

Attachment: Bug 1502529, compress touchmoves on IPC layer, r=masayuki

Comment 9

[Olli Pettay [:smaug][[e-mail]]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=1f8808c41268a70bc1ff3dae17a7e7f7c6447d7a

Comment 10

[Olli Pettay [:smaug][[e-mail]]]

Hmm, looks like I've lost link to the latest tryserver push.
https://treeherder.mozilla.org/#/jobs?repo=try&revision=11557adef33d4850b9554c0e9e1e7fb6504cdebe

Comment 11

[Olli Pettay [:smaug][[e-mail]]]

Aha, had to open the test in a new window. silly me.
https://treeherder.mozilla.org/#/jobs?repo=try&revision=3cca640d3da15797e1cdd24c37b788031e65125f

Comment 12

[Olli Pettay [:smaug][[e-mail]]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=770938ca1ade2b286742f2fe9d9c9cfe198457ca

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/139fa21432f0
compress touchmoves on IPC layer, r=masayuki

Comment 14

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/139fa21432f0