1949094 - Nightly is 3.25x slower than Chrome to paste text on https://quilljs.com/playground/snow

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, task, P2)

Description

[Mayank Bansal]

Attachment: reasonable css.txt

Go to https://quilljs.com/playground/snow
Copy-paste the attached sample text.

Nightly: https://share.firefox.dev/4hHeMqV (26s)
Chrome: https://share.firefox.dev/41nuRwc (8s)

Comment 1

[Jan de Mooij [:jandem]]

A very large chunk of this is Ion JIT code for the uA function and it's a little surprising we're that much slower. I'll take a quick look.

Comment 2

[Jan de Mooij [:jandem]]

Mayank, are you sure those numbers are right? Can you reproduce that also without the profiler?

What I see here on Mac is ~8 seconds in Chrome, ~14 seconds in Nightly. Slower but not 3x slower.

Comment 3

[Mayank Bansal]

This is what i get on a frehs profile : https://share.firefox.dev/4k152JL
I also timed it using a stopwatch and disabling the profiler and got roughly the same 20s.

Comment 4

[Jan de Mooij [:jandem]]

I looked at the code Ion generates for this uA function and we generate fairly decent code for most of the hot blocks - I didn't see any obvious low hanging fruit. Some of these blocks are executed > 1.8 billion times so small codegen issues might have a larger effect than you'd normally expect. This function also seems to walk a large tree or graph of JS objects so some of the overhead might be from cache misses.

            function uA(e) {
                e: for (;;) {
                    for (; null === e.sibling;) {
                        if (null === e.return || u$(e.return)) return null;
                        e = e.return
                    }
                    for (e.sibling.return = e.return, e = e.sibling; 5 !== e.tag && 6 !== e.tag && 18 !== e.tag;) {
                        if (2 & e.flags || null === e.child || 4 === e.tag) continue e;
                        e.child.return = e, e = e.child
                    }
                    if (!(2 & e.flags)) return e.stateNode
                }
            }

Comment 5

[Nicolas B. Pierron [:nbp] {away}]

Jan, is the uA function manipulating DOM elements? Maybe this bug is a hint to look at DOM field access.

Comment 6

[Iain Ireland [:iain]]

Looking at the profile, we're spending 97% of our time in uA itself, not in anything it calls. So I think it's unlikely that it's touching DOM elements.

Comment 7

[Mayank Bansal]

Not sure what changed, but this takes 2s on nightly: https://share.firefox.dev/41sgUwV .
Chrome takes 300ms, but now the hotspots are in editor/Focus/Mutation thingies, for which numerous bugs already exist.
CAlling this WFM.

Comment 8

[Iain Ireland [:iain]]

Looks like the webpage was rewritten; the script containing the slow uA function (https://quilljs.com/_next/static/chunks/framework-98741467467a0898.js) doesn't show up anywhere in the latest profile.