1348848 - FF52 Performance Regression

Status: NEW

(Core :: Layout, defect, P3)

Description

[irena.kull]

Attachment: FF52.zip

User Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; Trident/7.0; .NET4.0C; .NET4.0E; .NET CLR 2.0.50727; .NET CLR 3.0.30729; .NET CLR 3.5.30729; rv:11.0) like Gecko

Steps to reproduce:

Extract the attachment and open the index.html file in FF52 and measure time to load


Actual results:

Time to load is around 10s


Expected results:

Expected time to load is a few ms

Comment 1

[Alice0775 White]

Roughly regression range is between 46 and 47.

Comment 2

[Alice0775 White]

Regression window:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=5f6c9ac2919a7592af3e24c9142cac39dd79a1d5&tochange=4e08f29820a35936be6142b86350318e3246633d

Regressed by:
4e08f29820a3	Boris Zbarsky — Bug 1263845. When a parent changes from auto height to non-auto height or vice versa, a percentage height non-block child needs to realize it's doing a vertical resize. r=dbaron

Comment 3

[Boris Zbarsky [:bzbarsky]]

Right, that fix makes us do a relayout in more cases...  I'll look at this testcase to see whether we can avoid them in some cases in it.

Alice0775, do you mind requesting needinfo from me in the future on regressions from my patches, so they don't get lost in the other bugmail?

Comment 4

[Alice0775 White]

(In reply to Boris Zbarsky [:bz] (still a bit busy)
> 
> Alice0775, do you mind requesting needinfo from me in the future on
> regressions from my patches, so they don't get lost in the other bugmail?

I will do.

Comment 5

[Boris Zbarsky [:bzbarsky]]

OK, so this was slow before bug 1209994, then was sped up when we introduced bug 1263845 (which did not perform relayout in some cases when it needed to happen), then slowed down again when we fixed that.

I have various ideas for how we could make the table case specifically faster here, but the more general solution is fixing bug 1351924.  If that gets stuck, then I can probably implement something specific here.

Comment 6

[Gerry Chang [:gchang]]

It's too late for 54. Mark 54 won't fix and see if we can fix it in 55.

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: reporter's testcase as a single file