1457325 - Record and expose a timestamp after the parser finishes and the presentation is flushed

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, enhancement, P2)

Description

[Bobby Holley (:bholley)]

Tp6 currently uses time to first non-blank-paint, which is very non-deterministic - see bug 1455115 comment 3.

A potentially-more-reliable metric would be the end of the first style+layout+paint flush after DOMContentLoaded. I'm writing up a prototype for this, which we can then hook up to Talos.

Comment 1

[Bobby Holley (:bholley)]

Attachment: Expose time to DOMContentFlushed on the timeline. v1

I've only tested this lightly but it seems to work roughly as I'd expect.
Suggestions welcome.

MozReview-Commit-ID: E6QPjgfUKdo

Comment 2

[Bobby Holley (:bholley)]

Note that tomorrow is my last day before heading out for a while, so quick turnaround would be appreciated.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=b61a58604a26613990696d56ca1657605ed9d8cc

Comment 3

[Markus Stange [:mstange]]

Comment on attachment 8971438 [details] [diff] [review]
Expose time to DOMContentFlushed on the timeline. v1

Review of attachment 8971438 [details] [diff] [review]:
-----------------------------------------------------------------

This looks good to me. The "delete this;" part isn't very pretty but I don't have a better alternative.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 8971438 [details] [diff] [review]
Expose time to DOMContentFlushed on the timeline. v1

sr=me.  I'm on PTO tomorrow (Friday), and likely not checking email.....

Comment 5

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

(I thought people explicitly wanted to measure first non-blank paints, because that is what is relevant to users. But I guess having two different measurements is good.)

Comment 6

[Bobby Holley (:bholley)]

So, the post-refresh-observer setup ended up causing leaks, because the refresh driver can end up getting torn down, and it doesn't have any way to disconnect or drop the observer, which holds a strong reference to the presshell.

So instead I just implemented this directly in the refresh driver, which seems to work.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=9c58acb3565478caa98664c848693c3924b6d53d

Comment 7

[Bobby Holley (:bholley)]

Attachment: Expose time to DOMContentFlushed on the timeline. v2

MozReview-Commit-ID: E6QPjgfUKdo

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/560f881dffb8
Expose time to DOMContentFlushed on the timeline. r=bholley,r=mstange

Comment 9

[Bobby Holley (:bholley)]

NI sphilp to start tracking this metric in Talos.

Comment 10

[Andreea Pavel [:apavel]]

https://hg.mozilla.org/mozilla-central/rev/560f881dffb8

Comment 11

[Stuart Philp :sphilp]

See bug 1458242 for implementation in talos