466038 - Tp, TDHTML Regression on 11/18/2008

Status: RESOLVED FIXED

(Firefox :: General, defect)

Description

[Shawn Wilsher :sdwilsh]

There was a Tp regression only on Windows XP on 11/18/2008.  There are only three checkins in range:
http://hg.mozilla.org/mozilla-central/rev/4f32d50c10bc
http://hg.mozilla.org/mozilla-central/rev/5b9f7cddadcb
(both are part of the same bug)
12:13

And then the tracemonkey merge that happened after that.
top @ http://hg.mozilla.org/mozilla-central/rev/e8ed5d4bf531
bottom @ http://hg.mozilla.org/mozilla-central/rev/ebb3e752cb34
14:11

No other platform shows this regression.

Comment 1

[Shawn Wilsher :sdwilsh]

Tp_nochrome also shows this:
http://graphs.mozilla.org/#show=395008,395020,395048,1431867&sel=1226996681,1227107549

Comment 2

[Shawn Wilsher :sdwilsh]

TDHTML also shows a regression, moreso on XP than vista:
http://graphs.mozilla.org/#show=395032,395040,395060&sel=1227033590,1227057584 (XP)
http://graphs.mozilla.org/#show=787166,787165,787152&sel=1227000567,1227109779 (XP)

It's possible that bz's checkin for bug 445004 was also in the range for all of this:
http://hg.mozilla.org/mozilla-central/rev/b2fbc259ee2b

Comment 3

[Boris Zbarsky [:bzbarsky]]

Bug 445004 only affects document.write (and only which base URI is used for it, just replaces one virtual function call by another) and javascript: loads (a bit more cost for creating the javascript: URI, not much other difference).  It also added one more QI on the document channel during document load, and then a single hashtable lookup.  I'd be very suprised if any of this affected Tp, much less Tdhtml.

Comment 4

[Shawn Wilsher :sdwilsh]

Right, and mano's bug just added some stuff to an xbl binding (tabbrowser) but it's only ran for drag and drop code.  It also seems unlikely, but those two bugs are going to be easier to backout and see if they caused a regression than backing out the tracemonkey merge.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Ah, the bizarre diff on the merge commit had confused me into thinking there was more to mano's change.

If backing out bug 445004, need to back out both of the diffs there, the followup patch that got landed for bug 465752, and disable tests that got landed for bug 49312 and bug 132255.  Those tests will fail otherwise.

Comment 6

[Shawn Wilsher :sdwilsh]

Tp3 has dropped below pre-regression levels:
http://graphs.mozilla.org/#show=395008,395020,395048,1431867&sel=1226996816,1227950771

As has TDHTML
http://graphs.mozilla.org/#show=395032,395040,395060&sel=1227033005,1227966371 (XP)
http://graphs.mozilla.org/#show=787166,787165,787152&sel=1226998788,1227975048 (Vista)

I think it might be noise we were looking at.  Close this as WORKSFORME?

Comment 7

[Shawn Wilsher :sdwilsh]

No responses, so WORKSFORME

Comment 8

[Mike Beltzner [:beltzner, not reading bugmail]]

Agreed. We need a better way of tracking these things and perhaps of marking them as noise, as a way of later being able to systemically determine what was noise and what wasn't.