Electrolysis/Meetings/2016-03-03

« previous week | index | next week »

jimm

• bug 1232181 (M8, plugin window captures) - fixed. filed follow up (bug 1252874 on GTK. The simple solution involves a call that is very slow, a more performant solution involves intricate changes to plugin rendering.
• currently going through mochitest-chrome tests I have to sing off on
• cleared out needinfos

chutten

• Jank Release Criteria - completed. As far as I can tell, we're a go.
• Plugin Jank Release Criteria - the criteria we were hoping to use doesn't actually measure anything in e10s, so we lack data. These criteria should be abandoned. bug 1245575 is for cleaning up those probes for the future.

mconley

• Perf P1 bug 1251699 - Investigate e10s TART regression on osx and windows
  • I think we can chalk this one up to APZ. Waiting on retriggers to come in for Win 8 and OS X, but the rest look compelling.
  • General tip for perf spelunkers: try disabling APZ before you do anything else.
• bug 1251732 - See if e10s regresses tpaint for windows opened from the parent process
  • Turns out the answer is "no, except maybe a negligible amount on OS X".
  • I believe slow paths will be hit when content causes a new tab or window to open
  • We only measure the window case with tpaint, and this is probably less frequent than opening tabs due to default prefs
  • Unsure how to proceed.
• Many, many reviews. Lots of tests are getting activated for e10s.
• I'm traveling next week for a D&I Working Group thing I'm involved in. Unlikely to make next Thursday's stand-up.

poiru

• bug 1250716 Submission rate analysis

 submit attempts per main crash            0.19       0.17
 submit attempts per content crash         0.00       0.12

• Tried and failed to figure out why the absolute crash rate number differs for different build IDs. Note that the relative difference between e10s and non-e10s is pretty similar across different build IDs.

gabor

• bug 1248806 - (p1 crash - Recursive initialization of nsLayoutStatics when capability.policy.policynames is set) - fixed + landed
• APZ related regressions: https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&originalRevision=a4929411c0aa&newProject=try&newRevision=850e39c8f668&framework=1
• filed some perf bugs but still there are perf regressions without matching bugs based on https://treeherder.allizom.org/perf.html#/e10s
• bug 1194761 - (p1 perf - PCookieService should probably not use sync messages) - tried reproducing it but could not, all my profiled showed cookie service getters to be insignificant. I still see no reason to work on this or the other PCokkieService bug.
• bug 1210842 - (p1 perf - Long tab-switch times on hilton.com website) - did some profiling it seems that we call a lot more often restyles with e10s.
• bug 1252822 - (Scrolling www.theguardian.com is really janky with e10s) - I find this one particularly disturbing.

felipe

• bug 1250744 - Enable blocking of e10s by add-ons on beta/release
• bug 1252338 - Disallow E10s on ESR45
• bug 1248130 - Crashing user in wrong experiment branch
• bug 1249845 - E10s staged rollout
• Tests

bobowen

• bug 1252889 - On Windows use a 10 second pause instead of trying to invoke the JIT debugger with MOZ_DEBUG_CHILD_PROCESS - on inbound.

RyanVM

• Softvision team working on formally documenting the e10s test plan
• Working on getting e10s tests green enough on OSX to enable in CI
• Found owners for various performance-related release criteria

aklotz

• bug 1236108 - Crash report memory annotations not implemented for child processes (content processes) - Under review

gw280

• Not really much to report as I was on PTO for half of last week.
• bug 1114647 - rename plugin-container. patch up for review.
• bug 1186585 - tps regression. digging into this, eliminated a few possibilities (multiple active tabs, unsuppressed displayport).
• bug 1179732 - tresize regression. closed this now as per approval from jgriffiths! \o/ filed a follow up for WinXP only bug 1252947