Electrolysis/Meetings/2016-04-21

« previous week | index | next week »

billm

• Working on bug 1262671 to further address OOM issues in IPC code.

bsmedberg

• Stability dashboard data has been signed off by Kairo.
• He believes that this is no longer a blocker for deploying e10s to beta "permanently" (all 6 weeks of the cycle)
• Final production-quality dashboards have not been built, but there are a bunch of prototypes which can be used to answer questions now see https://public.etherpad-mozilla.org/p/kairo-telemetry-notes and https://sql.telemetry.mozilla.org/dashboard/stability-dashboard-for-release-tracking
• In particular for e10s, see https://sql.telemetry.mozilla.org/queries/177 - e10s crash rate is still 3x control
• For 47 beta, really want bug 1265015 - may need help from EPMs getting proper urgency/traction

gw280

• Mostly working on using firefox.exe for everything. Should have a patch by the end of today or tomorrow.
• Per-process texture pool: have some preliminary patches and some good feedback from nical and mattwoodrow. Working on this in the background.

haik

• Landed bug 1258925 ([e10s] Browser window is resized when click)
• Working on bug 1248683 ([e10s] Find bar sometimes gets keypresses in the wrong order when system is under heavy load)

jimm

• bug 1263951 (crash in mozilla::ipc::TransferHandleToProcess (GMP related)) - patch posted, waiting on review.

mconley

• M9 bug 1263887 - [e10s] Select dropdowns in e10s mode do not notice changes to the <select> after it has opened.
  • Patch up, waiting on review
• Perf P1 - Bug 1251699 - 5.7% TART regression on OS X with APZ disabled
  • This has gone away. The remaining TART regression is APZ-related, and a price we have to pay for APZ. See bug 1253412
  • barbara signed off on TART over email
• In discussions with a number of folks regarding the remaining UI Smoothness Criteria, and whether we want to block on what's left.
  • I think "no", but being conservative and assuming the group will say "yes", and diving full-on into the remaining regressions.
• Perf P2 bug 1252768 - 2–7% e10s CART regression (with APZ disabled)
  • Linux64 only, and very mysterious. I have about 12 try builds going right now with various experiments.
• Perf P2 bug 1174770 - tpaint regressions (3%-8%) in e10s mode compared to non-e10s mode
  • The big win will be in bug 1261842 - Make initial browser remote sooner if we're defaulting to using remote tabs.
    • Initial data suggests significant wins over non-e10s for both tpaint and ts_paint.
• Perf P2 bug 1096013 - [e10s] Improve perceived session restore duration
  • Got this working yesterday, where we restore favicons and tab labels way faster, which is nice
  • Also makes it so that background tabs, when crashing, don't lose their favicons, which is even nicer
  • Currently trying to find the best way to do it "right", since my initial solution was hacky at best.

chutten

bug 1247443 - providing analysis support. Apparently basically no one on Nightly sets browser.tabs.animate to a non-default value.

bug 1265000 - we now report dom.max_script_run_time on 47+

kanru

• bug 1253956 NEW:: [e10s] Windows debug browser/base/content/test/general application crashed [@ mozilla::LinkedList<js::WeakMapBase>::~LinkedList<js::WeakMapBase>()] (Assertion failure: isEmpty(), at LinkedList.h:332) on shutdown
  • Initial investigation
• bug 1264642 NEW:: Reduce the size of PBrowser::Msg_AsyncMessage IPC messages
  • study the StructuredClone js API and commented the findings
• bug 1265680 NEW:: crash in mozalloc_abort | NS_DebugBreak | mozilla::ipc::FatalError | mozilla::dom::PBrowserChild::FatalError | mozilla::dom::PBrowserChild::OnMessageReceived
  • Initial investigation.

cervantes

• bug 1258317 NEW:: crash in mozilla::ipc::TransferHandleToProcess (PBackground related)
  • Updating the diagnostic patch, landed
• bug 1259480 NEW:: [e10s] FatalError crash in PContentChild::Read from PContentChild::SendReadPrefsArray

wcpan

• bug 1259641 NEW:: [e10s] MediaFeatureValuesChangedAllDocuments takes tons of time when resizing e10s window
  • Request review, got r+

ting

• bug 1219914 NEW:: 25MiB AWSY regression when re-enabling jemalloc 4
  • compared jemall4 on/off with current m-c on awsy, traced jemalloc for dirty pages ratio, am collecting dmd data
  • made a patch to try out 1mb dirty pages limitation, then the diff on page-cache and heap-unclassified are similar, am tring to comapre the talos' scores
  • jemalloc4 regresses talos, ni'd erahm and glandium for whether to land bug 1005844 at first

gabor

• bug 1247420 - (m9 1255138 - Resizing windows from JS is not supported in e10s mode) - I'm struggling with the test. The individual features are complete but the test is quite broken and hard to get right for both e10s and non-e10s to work. If something goes wrong with it (typically only on try) because of a tricky coordinate math random things happen and it's hard to track what went wrong exactly. I'm considering to just rewrite the entire test from scratch, I wasted too much time on it already. Another issue that this is a quite tricky patch, since layout is not particular straightforward when it comes to window manipulation to say the least... and splitting the whole thing in to process did not help. I need someone who can review it and I'm concerned about the uplift a bit.
• bug 1263953 - (Reduce the growth rate of Pickle) - Landed.
• bug 1264383 - (Add-on multiprocess compatible shims stops TabSelect event listeners being removed from the tab container) - requested uplift, we decided not to uplift to beta/esr.
• bug 1262661 - (Many large SessionStore:update messages) - One thing we should do is breaking up big messages automatically, but maybe the recent work on pickle will be enough. Another thing is that a, we always send the whole session store instead of the changes only b, we have a MessageQueue that sends the messages in batches. We could be smarter there too by forcing the queue to fire up a message before the timeout beyond a certain size.
• more addon reviews...