379476 - Ts regression on 2007-05-01

Status: RESOLVED INCOMPLETE

(Core :: General, defect)

Description

[Adam Guthrie]

There was a small regression in Ts around midday on all three platforms. Here's the Bonsai URL for the Mac tinderbox:

http://bonsai.mozilla.org/cvsquery.cgi?module=PhoenixTinderbox&date=explicit&mindate=1178045700&maxdate=1178049239

The regression range for the Linux tinderbox is a bit tighter:

http://bonsai.mozilla.org/cvsquery.cgi?module=PhoenixTinderbox&date=explicit&mindate=1178047320&maxdate=1178050379

Mac seems to have been hit the hardest with a 5% increase in Ts. My guess is bug 221428 is the culprit.

Comment 1

[Boris Zbarsky [:bzbarsky]]

So... the problem is that I have no way to debug this -- the startup tests have 40%+ noise over here, so no matter what I do all I can tell is the numbers suck.

And I don't see anything in the CSS changes that would have slowed down Ts, short of actually getting the sheet principals (which is absolutely necessary)....

I guess I'll try profiling startup once I have a profiling-enabled build with that change, but that might be weeks.

If someone can actually reproduce this regression locally, I'd love to hear about it.

Comment 2

[Boris Zbarsky [:bzbarsky]]

OK, so the amount of time spent under nsCSSStyleSheet::nsCSSStyleSheet()
goes from 3 hits to 38 hits with that patch...  The total number of hits during startup is on the order of 42000.

I suppose this could be a lot worse on mac if null principal creation is more expensive, but the Linux tinderbox _does_ show a pretty clear Ts increase...

I assume the times involved are pretty much correct?

Comment 3

[Shawn Wilsher :sdwilsh]

I'm going through and marking old performance regression bugs as INCOMPLETE that are likely too old to be valid or get any traction on them.

Please re-open if you have more information or can demonstrate the regression still exists.