By 'Switch the theme, once or twice', I mean the label's theme (choosing a photo border from the slider), not the Drupal theme, sorry for the confusion.
One other small note, I tried to reproduce the issue with an older build of FF (sitting on a VM in the office here, FF 12) and the issue was not present.
Is there any chance you could help narrow the regression range by using http://mozilla.github.com/mozregression/?
Hey Josh, sure I can run the regression (cool tool btw). I started at 2010-01-01 and let the tool run from there. mozregression --good=2010-01-01 [OK] 10.0a1 (2011-09-28) [OK] 17.0a1 (2012-08-10) [BAD] 21.0a1 (2013-01-16) [NOT AS BAD] 19.0a1 (2012-10-28) [OK] 18.0a1 (2012-09-18) [BAD] 18.0a1 (2012-10-08) [BAD] 18.0a1 (2012-09-28) [BAD] 18.0a1 (2012-09-23) [BAD] 18.0a1 (2012-09-20) [BAD] 18.0a1 (2012-09-19) -more targeted mozregression --good=2012-09-17 --bad=2012-9-19 '19.0a1 (2012-10-28)' might be a error in my judgement but 2012-09-18 is clearly good and all nightly builds AFTER have the hang / pause after the request completes.
Thanks! Going by the revisions of the Windows builds for 9/18 and 9/19, that suggests the following regression range: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=0d3b17a88d5f&tochange=80499f04e875
I can reproduce this easily on Fedora, too.
Status: UNCONFIRMED → NEW
Ever confirmed: true
OS: Windows 7 → All
Hardware: x86_64 → All
Figuring out what's happening in the busy loop would be really useful. Would you mind gathering some profiling data by following the steps at https://developer.mozilla.org/en-US/docs/Performance/Reporting_a_Performance_Problem? I'd do it myself, but the profiler doesn't give useful data on Linux systems.
Created attachment 766680 [details] Profiler Run #1 Profiler output from 1st run
It's actually easier to upload the profile results to the server, since you then receive a link you can share in Bugzilla.
Antony, given that I haven't figured out how to use the files you attached yet, could you just do another profiler run and paste the link to the uploaded profile here?
Hey Josh, the profiler is not working right now. On the nightly build it fails to load the url for the profiler-consuming site. From:https://addons.mozilla.org/en-US/firefox/addon/gecko-profiler/?src=api 'Not available for Firefox 25.0' - Nightly is @ 25.01a right now ;( I can't install 24.X right now, try the profiler analyse, download the file: (not explaining that) goto:http://people.mozilla.com/~bgirard/cleopatra/ with FF, and click upload.
(sorry, forgot to mention to select the file in comment #14)
You can use https://github.com/bgirard/Gecko-Profiler-Addon/raw/master/geckoprofiler.xpi instead.
Couldn't get the built-in profile to save/share the profile, also, it doesn't show the red (busy) portion of the request? Anyways, I uploaded the profile files to the link in comment #14 and said 'share' on the bottom left, here are the links to the shared profiles. http://people.mozilla.com/~bgirard/cleopatra/#report=6aba5fc4c84a8d4b42ba3b2cfb813f877b2de035 http://people.mozilla.com/~bgirard/cleopatra/#report=4395f6920521c10c45f351f7e319e77b79cee242
Perfect! The profiles show us spending about ~40% of time performing layout (and the timeline shows lots of synchronous layout calls occurring, which would make sense), especially font shaping, and about ~5% of time executing Dupal AJAX code. Alice, would you be able to bisect the regression range any further for us?
..I was looking at the profiler output and noticed that @ 49% of the time was spent in NtUserWaitMessage called from nsBaseAppShell::OnProcessNextEvent(nsIThreadInternal *,bool,unsigned int). I don't know much about the code though...
Regression window(m-c) Good: http://hg.mozilla.org/mozilla-central/rev/2e429432490c Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20120918023002 Bad: http://hg.mozilla.org/mozilla-central/rev/e4757379b99a Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20120918044902 Pushlog: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=2e429432490c&tochange=e4757379b99a Regression window(m-i) Good: http://hg.mozilla.org/integration/mozilla-inbound/rev/27f122beb4f5 Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20120917154902 Bad: http://hg.mozilla.org/integration/mozilla-inbound/rev/49bc46738815 Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20120917164302 Pushlog: http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=27f122beb4f5&tochange=49bc46738815 In local build Last Good: e016faeb6b15 First Bad: 1d8ff0faf0fa Triggered by: 1d8ff0faf0fa Mats Palmgren — Bug 791601. r=roc
Component: General → Layout
Keywords: regressionwindow-wanted → perf, regression
Version: 21 Branch → 17 Branch
Assignee: nobody → matspal
You need to log in before you can comment on or make changes to this bug.