Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

2017-09-28

NECKO WEEKLY MEETING NOTES

Thurs Sept 28th, 2017

STICKY

  • Triage duty: nwgh
  • new untriaged bugs: https://tinyurl.com/zhk9yjf (https://mzl.la/2ylkidL)
  • NECKO-ACTIVE bugs: https://tinyurl.com/j8pj35u (https://mzl.la/2he1C85)
  • NECKO NEXT bugs: https://tinyurl.com/zjwjfee (https://mzl.la/2yaD1aO)
  • NECKO-BACKLOG bugs: https://tinyurl.com/hpkcxx3 (https://mzl.la/2xcYzEL)
  • NECKO-WOULD-TAKE bugs: https://mzl.la/2hdnQXy
  • Triage Notes
    • Android crashes on infra are currently not symbolicated: https://bugzilla.mozilla.org/show_bug.cgi?id=1389805 (makes triaging/diagnosing android crash bugs hard/impossible)

Round Table:

  • (15:48:00) annevk: if y’all have meetings, could you bring up fixing https://bugzilla.mozilla.org/show_bug.cgi?id=1363284 as part of some Quantum effort
  • (15:48:28) annevk: we even have Google devrel volunteering to draw up graphics how it makes Firefox faster than Chrome
  • (15:49:05) annevk: https://twitter.com/annevk/status/901493567855710208

Nick

Honza

Michal

  • reviews
  • RCWN
  • 1403010 Intermittent toolkit/components/aboutmemory/tests/test_memoryReporters.xul | application crashed [@ arena_salloc]
  • 1401224 Assertion failure: aLength > 0 (Writing a zero-sized chunk), at image/SourceBuffer.cpp:348
  • 1394564 Version 55 breaks sending large amounts of data over web sockets.

Patrick

last week: mpendingq, 256fd telemetry, 0.9 telemetry, namespaces, neat tls, alt-svc, immutable bug, some non-necko stuff this week: neat tls, alt-svc, immutable bug, some non-necko stuff. more reviews.

Daniel

  • NEAT project work
  • Bug 1395914 - IPv6 network identification (Linux version reviewed OK, mac version coming)

Valentin

Bug 1403154 - Crash in mozilla::dom::PerformanceNavigationTiming::Type

Bug 1403926 - PerformanceNavigationTiming fields are not correct

Bug 1392739 - IPC: unknown address crash [@CharAt] - IPC fuzzing with errors

Bug 1403137 - Crash in ContinueProcessResponse2 - null mTransactionPump

Dragana

TFO

rel=preload and no-cache

Gary

Bug 1373708 - Convert Jar channels to stop using main thread I/O

  • fixing tests, some patches are submitted.

Junior

Bug 870460 - Lazy load of cookie service blocks main thread while cookie database loads

  • r? intend to implement: remove legacy upgrade: cookie.txt, DELETE null base basedomain

Amy

Bug 1376009 - Crash in mozalloc_abort | NS_DebugBreak | mozilla::ipc::LogicError | mozilla::net::PCookieServiceChild::SendGetCookieString

  • uplift to beta Bug 1360603 - Mark Save as… channels used for download as throttleable

  • f+

  • Have to ask dom peer to review this patch Bug 1386746 - HTTP response throttling: experiment with limiting the amount of data we read during the short don’t-block-read period

  • Started to work on this bug.

Kershaw

Bug 1377208 - Move background transactions over h2 to the background group

  • landed

Bug 1395472 - Kill switch for moving background transactions over h2 to the background group

  • landed

Bug 1247843 - All s are downloaded during the critical path

  • WIP
  • Fix test failures

SC

[PBg-Http]

  • performance analysis
  • First Paint time for visible elements in HTML document is improved because HTML parser will start to create DOM element earlier. The difference can be observed from TALOS tp6 amazon/youtube. The average first paint time is improved around 50ms.
  • However for facebook no improvement is observed because there is no visible element in the HTML document. In the HTML document there is only for loading CSS/JS scripts and executing
  • I tried to dispatch StopRequestEvent runnables into high priority queue. The TALOS result shows changes dramatically. Performance test is either improved a lot or regress a lot. All the tp6 test cases are seeing huge improves (around 80~100ms for amazon/youtube/facebook, 30ms for google). However, the tpaint, tp5o_scroll, and several tp5o metrics are regressed a lot. My impression is this patch changes the event sequence too much, so that the performance landscape is changed too much. TALOS result: https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=fd6dceb87671ff98461f935a0be00678f017264b&newProject=try&newRevision=e5cfac6814cfe32fa7ad6a8bf084bb874d4d7fd7&framework=1&showOnlyImportant=0&showOnlyConfident=1
  • I did scrolling smoothness test on facebook timeline page with Taipei Hasal team. No obvious performance change after PBg-Http is enabled because the jank is caused by JS code in web page and rendering. This is because facebook uses React framework, which is using javascript to manage components and virtual DOM tree.
  • QA in taipei also help record a comparison video of scrolling smoothness on instagram. With PBg-Http enabled the scrolling speed and smoothness is improved [comparison video https://youtu.be/rlWsig8YA58]. Looked into the corresponding gecko profile, instagram is use less heavy JS to do the incremental loading. Thus, the benefit of OMT is easier to be observed.
  • trying to get CC/GC log from treeherder for investigating bug 1353829
  • major difference is because different set of decoded image is hold by Gecko. This is highly related to the change of loading sequence.