418810 - Tp Ts Txul regression on bl-bldlnx03_fx-linux-tbox-head on 20080218

Status: RESOLVED INCOMPLETE

(Core :: CSS Parsing and Computation, defect)

Description

[Steve England [:stevee]]

Attachment: Image of graph showing Tp regression

pageload from bl-bldlnx03_fx-linux-tbox-head:

Build start: 20080218 22:12:00 Tp ~241ms
Build start: 20080218 22:26:00 Tp ~245ms

Looks like it was caused by bug 401291

http://build-graphs.mozilla.org/graph/query.cgi?tbox=bl-bldlnx03_fx-linux-tbox-head&testname=pageload&autoscale=1&size=&units=ms&ltype=&points=&showpoint=2008%3A02%3A21%3A03%3A02%3A19%2C248&avg=1&days=3

http://bonsai.mozilla.org/cvsquery.cgi?module=PhoenixTinderbox&date=explicit&mindate=1203396840&maxdate=1203403139

Comment 1

[Steve England [:stevee]]

Attachment: Image of graph showing Txul regression

xulwinopen (bl-bldlnx03_fx-linux-tbox-head)

189ms --> 193ms

http://build-graphs.mozilla.org/graph/query.cgi?testname=xulwinopen&units=ms&tbox=bl-bldlnx03_fx-linux-tbox-HEAD&autoscale=1&days=7&avg=1&showpoint=2008:02:21:04:24:12,193

Comment 2

[Steve England [:stevee]]

Attachment: Image of graph showing Ts regression

Ts (bl-bldlnx03_fx-linux-tbox-head)

845ms --> 850ms

http://build-graphs.mozilla.org/graph/query.cgi?tbox=bl-bldlnx03_fx-linux-tbox-head&testname=startup&autoscale=1&size=&units=ms&ltype=&points=&showpoint=2008%3A02%3A21%3A04%3A27%3A53%2C834&avg=1&days=4

Comment 3

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

Hrm, I'd clicked a whole bunch of graph links and didn't see anything that bad.  Sayrer did point out a slight Txul regression that showed up on the XP and maybe Mac talos boxes (and it looks like the Txul regression is the worst one here).

I think some of this may be unavoidable -- the bug was that we weren't doing enough work; we have to do more work to fix that.  I started looking into which selectors / elements we were actually hitting that case on, but got pulled away.

Comment 4

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

Things to investigate:
 * is the extra time the testing whether we need to restyle or the actual restyling
 * if the latter, what selectors/elements are causing us to take the slow path?
 * if the latter, does setting the flags in SelectorMatches more precisely by using a variable and holding off until the end of the function help?

Comment 5

[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 (specifically the stuff dbaron lays out in comment 4)