Closed Bug 413123 Opened 18 years ago Closed 18 years ago

2.5% Ts regression on cb-xserve01 1.8 branch (not trunk) after changing an uppercase F to lowercase

Categories

(Camino Graveyard :: General, defect)

1.8 Branch
All
macOS
defect
Not set
normal

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: mark, Assigned: stuart.morgan+bugzilla)

References

()

Details

(Keywords: fixed1.8.1.12, perf, regression)

Attachments

(2 files)

I don't really believe it either, but Ts on the 1.8 branch regressed when I changed the update menu item text from "Check For Updates…" to "Check for Updates…". The new nib was saved in Interface Builder 2.5.6 (Xcode 2.5) on 10.5.1, while the old nib was probably 2.4.1/10.4.x. The new one might be slightly less efficient to read, at least on 10.4. http://build-graphs.mozilla.org/graph/query.cgi?testname=startup&units=ms&tbox=cb-xserve01.mozilla.com_MOZILLA_1_8_BRANCH&autoscale=1&days=7&avg=1&showpoint=2008:01:17:15:08:15,389 http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=all&branch=MOZILLA_1_8_BRANCH&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=1200607740&maxdate=1200610379&cvsroot=%2Fcvsroot
I backed this out (both places, for consistency) to verify that Ts will go down on the branch; I'll watch the boxen for a while tonight. I also noticed that classes.nib and keyedobjects.nib seemed to have an unusually large number of lines changed, and in classes.nib the formatting seems slightly different. I may also experiment with landing a MainMenu.nib edited in IB 1.5, depending on how time goes.
Unsurprisingly, branch Ts dropped from 394/390ms the two cycles before to 377/380ms the two cycles after the backout. (cb-x1 trunk showed no appreciable change, as expected, and boxset jumped around a bit but remained almost entirely within its current range.) The little-f nib remains backed out at this point.
I just checked in a MainMenu.nib edited in IB 2.4.2 (v364) on 10.3.9. The diff seems much saner (at least the non-binary part). classes.nib seems not to have been changed at all (whereas it was +265/-234 in mento's original checkin). keyedobjects.nib was +47/-162 vs. +155/-160, and info.nib was +3/-3 (two of those were IB version markers, which I could have hacked out) vs +3/-7. Stay tuned for perf numbers.
Attachment #298057 - Attachment is patch: true
Attachment #298057 - Attachment mime type: application/octet-stream → text/plain
keyedobjects is the only significant one at runtime. The +/- numbers are meaningless for binary files. They tend to be meaningless for nibs anyway. Apropos of this bug, f Interface Builder.
So, our 10.3.9 nib's numbers (2 before, 2 after): 379/378ms to 381/383ms. I would have liked to see that 383ms number drop a bit more instead of continuing to rise afterwards, but it's a bit better than the IB 2.5.6 nib. The backout nib was in for four cycles: 377 380 379 378 The 10.3.9 nib after four cycles shows: 381 383 383 390 If someone has a 10.4 with the IB that comes with Xcode < 2.5, we can back the 10.3 nib out and try again to see if we can manage an even better result. I think for now let's leave it in overnight and let it run a more cycles to see if the numbers steady or continue to rise.
I can make a 10.4 nib tomorrow night or Monday.
Attached file 10.4 nib —
Here's a 10.4/IB2.5.4 nib.
In the past 12 hours or so, we've fallen to be more consistently 378-380, which is almost where we were pre-little-f nib. I'm going to go ahead and check in Stuart's 10.4/IB2.5.4 nib right now, for completeness's sake, and let it run a while before I do any of my other landings today.
This seems to have remained in the upper-370s since checkin (379, 378, 379, 374, 377), roughly where we were before the initial little-f checkin, so let's call this FIXED with Stuart's nib.
Assignee: nobody → stuart.morgan
Status: NEW → RESOLVED
Closed: 18 years ago
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: