1297367 - Call iCC slice and forgetSkippable right after paint

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

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

Attachment: iccslice_forgetskippable_right_after_paint.diff

...so that we'd have as much as possible time before the next refreshdriver tick.

The patch also moves NotifyDidPaint() calls a bit later so that DidRefresh callbacks have chance to run before the slow operations.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=e7449fcbdfc3

Comment 1

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

https://treeherder.mozilla.org/#/jobs?repo=try&revision=d065c4db93fe
https://treeherder.mozilla.org/perf.html#/comparechooser?newProject=try&newRevision=d065c4db93fe

Comment 2

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

Comment on attachment 8783934 [details] [diff] [review]
iccslice_forgetskippable_right_after_paint.diff

I think I'd like to try this. We should push more stuff to happen right after painting, and not at random time between paints, so that it is more likely that we don't miss any paints.
And GC already does something like this.
Copy-pasting timer initialization isn't perhaps the nicest but adding some helper wouldn't help much since in this case we want to re-init timers only if timers haven't been released during ICCTimerFired/CCTimerFired calls. So the code isn't exactly the same as in the other place where the timers are initialized.


We should also add some telemetry probes to detect when GC or CC is running when refreshdriver should tick, in some other bug.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/649c62d13b57
Call iCC slice and forgetSkippable right after paint, r=mccr8

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/649c62d13b57

Comment 5

[Ryan VanderMeulen [:RyanVM]]

Backed out per bug 1297977.
https://hg.mozilla.org/mozilla-central/rev/7963ebdd52b9

Comment 6

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

Attachment: iccslice_forgetskippable_right_after_paint_2.diff

https://treeherder.mozilla.org/#/jobs?repo=try&revision=3264043f1265

Comment 7

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

Comment on attachment 8789617 [details] [diff] [review]
iccslice_forgetskippable_right_after_paint_2.diff

Doesn't seem to regress per tryserver talos

interdiff works quite well here.
https://bugzilla.mozilla.org/attachment.cgi?oldid=8783934&action=interdiff&newid=8789617&headers=1

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/32c103f71eff
Call iCC slice and forgetSkippable right after paint, r=mccr8

Comment 9

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/32c103f71eff

Comment 10

[Pulsebot]

Backout by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/16e3043a7ab7
backout because of tart regression, r=backout

Comment 11

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

Attachment: v3

This would remove gc/cc slices from tart, but apparently avih doesn't want that.

Comment 12

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

https://treeherder.mozilla.org/#/jobs?repo=try&revision=8c91a90b60c6 is anyhow talos run for that, but I'll re-land v2.

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/c4e8525415a0
Call iCC slice and forgetSkippable right after paint, r=mccr8

Comment 14

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/c4e8525415a0