1944819 - Firefox is 2x slower on artificial "Self-referencing Shadow DOM" testcase

Status: RESOLVED FIXED

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

Description

[Mayank Bansal]

Attachment: Self-referencing shadow DOM.HTML

Open the attached testcase

enter 1000 and press enter

chrome: https://share.firefox.dev/4jDxfq0 (13s)
Nightly: https://share.firefox.dev/3WGviix (24s)

Testcase generated by chatgpt on my prompt. No idea how realistic this testcase is. But i hope that this may potentially highlight some hotspots.
As usual, feel free to mark INVALID/WONTFIX

Comment 1

[Mayank Bansal]

Sean, any thoughts on the profile and if it highlights anything worth optimizing?

Comment 2

[Sean Feng [:sefeng211]]

Nothing really stands out to me. I see we did a bunch of DOM works with calling appendChild, I am not sure how we can optimize anything in the call chain.

Olli, do you see anything?

Comment 3

[Mayank Bansal]

1000 elements
Profile with latest Nightly: https://share.firefox.dev/4kF4M3J (24s)

Comment 4

[Mayank Bansal]

This is what I get today: https://share.firefox.dev/4i7S2AT (19.5s + CC) , which is very different and faster than the previous profile. No idea what changed in the interim period.

Comment 5

[Mayank Bansal]

latest profile: https://share.firefox.dev/47mauTk (22s + cleanup)

Comment 6

[Mayank Bansal]

and this is with adding idlecallback in each iteration: https://share.firefox.dev/3Ln0wbI (70 seconds)

Comment 7

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

We have a flag which can be used here to improve this quite a bit, assuming I read the test and profile correctly.
(testing...)

Comment 8

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

Attachment: Bug 1944819, no need to check if parent is a descendant of a new child, if new child doesn't have possible containers as its children, r=emilio

Comment 9

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

The patch helps ~25%.
There might be still other things to optimize

Comment 10

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/fea6941decfb
https://hg.mozilla.org/integration/autoland/rev/326156160cc8
no need to check if parent is a descendant of a new child, if new child doesn't have possible containers as its children, r=emilio

Comment 11

[Iulian Moraru]

https://hg.mozilla.org/mozilla-central/rev/326156160cc8

Comment 12

[Mayank Bansal]

Latest: https://share.firefox.dev/49NbfGz / https://share.firefox.dev/3LvMipa (17s + 5.5s in cleanup)
Olli, anything else that seems worth fixing?

Comment 13

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

Other things would happen in other bugs. We might figure out some optimization for nsIMutationObserver. And clean up part is being handled elsewhere (possibly no specific bugs quite yet).