1410464 - Janky window resize on testcase similar to the webconsole UI with lots of flex children

Status: RESOLVED INVALID

(Core :: Layout, defect, P3)

Description

[Brian Grinstead [:bgrins]]

STR: 

* Open https://console-html-500-logs-addition-every-1s-scroll-to-bottom.glitch.me/
* Resize window

The resize is pretty slow / janky. This is a test case extracted from the webconsole, so the same problem affects us there.

Here's a profile: https://perfht.ml/2yVuNbg

Comment 1

[Brian Grinstead [:bgrins]]

Daniel: do you think this one can be closed out after Bug 1490890, or is there more work to do here?

Comment 2

[Daniel Holbert [:dholbert]]

I don't think this one's fixed (or at least, not by Bug 1490890).

I just profiled the resize-operation in Nightly 2018-09-14, and got this profile with a series of reflows in the 25ms-52ms range:
  https://perfht.ml/2DtmpCG

And I profiled today's Nightly and got basically the same results (25-45ms long reflows):
 https://perfht.ml/2NtNfzw

So I think this one (particularly the resize operation) is distinct from what was going on in bug 1490890 and the console scenario that was addressed there.

Comment 3

[Daniel Holbert [:dholbert]]

Profile in Chrome 83 (focusing on Marker chart to show Layout events): https://share.firefox.dev/2XYpsLz
Profile in Firefox Nightly 79 (from today): https://share.firefox.dev/3huBaFF

We're looking approximately as good (or as bad) as Chrome here, actually!

• Both of us have a mix of reflow durations, with lots of them in the 10-20ms range.
• Both of us have some longer reflows on the order of 40ms
• Chrome has some extra-long reflows on the order of 130-150ms (We don't have any of those, in my trial; yay us!)

So it doesn't look like there's a Firefox-specific performance bug here; this content seems similarly-slow in Firefox as well as Chrome.

This may be something we need to work around in the webconsole (which I guess falls under the bug 1156747 bucket).

But: comparing this particular scenario to Chrome, it's not clear that we could do much better here on the layout-performance side.

--> Closing as INVALID for now, based on this comparison.