1965186 - Intermittent remoteAsyncEvents TEST-UNEXPECTED-TIMEOUT | /dom/events/scrolling/scroll-cross-origin-iframes.html | Verify sibling cross-origin iframes can wheel-scroll. - Test timed out

Status: RESOLVED FIXED

(Core :: Panning and Zooming, defect, P2)

Description

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Even with bug 1848957 fixed and now dispatching widget events for wheel scroll through Marionette this particular test still times out. I did a quick test and noticed that when synthesizing a wheel scroll for whether frame 1 or frame 2 it works. But synthesizing to both of them in that loop makes the test to timeout because we do not get a scrollend event for the second frame. Instead the scrolling seems to be performed in the first frame again for a second time.

To see the issue run the following command:

mach wpt --setpref="remote.log.truncate=false" --setpref="remote.log.level=Trace" --setpref="remote.events.async.wheel.enabled=true" testing/web-platform/tests/dom/events/scrolling/scroll-cross-origin-iframes.html

Dan, could you maybe have a look?

Comment 1

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Actually adding a sleep of >500ms before synthesizing the seconds event to the other frame or waiting this amount of time before returning from .scroll() fixes the issue. As such I think this will as well depend on bug 1963550.

Comment 2

[Henrik Skupin [:whimboo][⌚️UTC+1]]

It may be related to the mousewheel.transaction.timeout value which we set to 500ms for the wpt tests under /dom/events/scrolling/.

Hiro will take a look.

Comment 3

[Dan Robertson (:dlrobertson)]

(In reply to Henrik Skupin [:whimboo][⌚️UTC+2] from comment #2)

It may be related to the mousewheel.transaction.timeout value which we set to 500ms for the wpt tests under /dom/events/scrolling/.

Hiro will take a look.

Exactly what I was thinking. Wheel transactions work across iframes. To confirm you could run with dom.event.wheel-event-groups.enabled=false. I'll double check this tomorrow.

Comment 4

[Hiroyuki Ikezoe (:hiro)]

Okay, so As Henrik pointed out Marionette tries to end the active wheel transition via calling ChromeUtils.endWheelTransaction, but it does just try to end the transaction on the main-thread.

We need to invoke WheelBlockState::EndTransaction.

This is indeed an issue in APZ.

Comment 5

[Hiroyuki Ikezoe (:hiro)]

Attachment: Bug 1965186 - End the wheel transaction for APZ in ChromeUtils::EndWheelTransaction as well as the transaction on the main-thread. r?#ipc-reviewers!,whimboo!,botond!

A caveat here is that even though ChromeUtils.endWheelTransaction is a
test-only API, PWebRenderBridge.EndWheelTransaction IPC call can not be
sync because it ends up deferring the ending the transaction task to the
controller thread.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/388c2bdd14af
End the wheel transaction for APZ in ChromeUtils::EndWheelTransaction as well as the transaction on the main-thread. r=whimboo,botond,ipc-reviewers,webdriver-reviewers,nika

Comment 7

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/388c2bdd14af