Platform/2010-08-03

From MozillaWiki

« previous week | index | next week »

Announcements

• Focus on blockers
• Do not check in performance regressions

Notices / Schedule

Previous release post-mortems

• Firefox 3.6.7/8: 2010-08-03 @ 1:00 pm PST, duration 45 mins
  • details are here
• Firefox 3.6.4/6: 2010-08-10 @ 1:00 pm PST, duration 1 hour
  • The details on how the post-mortem will be run will be sent out today

Firefox (Supported Branches)

• Schedules for Firefox 3.6.9/3.5.12 are posted
  • Code freeze is currently scheduled for Thursday August 12 @ 11:59 pm PST

Firefox (Beta)

• missed code freeze for Firefox 4 Beta 3 due to leaks in Firefox Sync
  • 23 blockers changed from beta3+ to another milestone after July 30 – this isn’t a sustainable pattern
  • the complete changelist for this beta

Firefox mobile (alpha)

• aiming for code freeze on 8/10.
• Bug list

Blocker Report

Firefox 3.6.9

• There are 22 open blockers (+0 w/w)
  • Code freeze is currently scheduled for Thursday August 12 @ 11:59 pm PST
  • Let’s try to not get them all in on the last day. I’ll be bugging people furiously this week and the start of next
• Only approving blockers at this point

Firefox 3.5.12

• There are 13 open blockers (+0 w/w)
• Code freeze is tied to 3.6.9

Firefox 4 Beta

• a handy list of triage queries is available for all!
  • Beta 4: 34 blockers
  • Beta N: 204 blockers
  • Final: 222 blockers (462 total)
  • nominations: 462 nominations (248 OPEN)

Firefox Development

(from our goals):

• [NEW] Feature complete Firefox 4
  • [DONE] Switch to Tab
  • [ON TRACK] Extension Manager – Functionality there, lots of bugs to fix.
  • [AT RISK] Account Manager – Partial work, restaffing to get this one done.
  • [ON TRACK] Notification UI – Geo done, WIP patches on others.
  • [AT RISK] TabCandy – Initial reviews. Big patch, might take iteration, but getting traction now
  • [ON TRACK] Integrated Sync – Landed, activation caused leaks, chasing as we speak.
  • [ON TRACK] New Theme – Substantially done on Windows and Mac, Linux lagging.
  • [ON TRACK] App Tabs – Basic UI in place, several functionality follow ups to get to feature complete
  • [AT RISK] Inspector
  • [AT RISK] Web Console
  • [ON TRACK] Silent updates on Windows

• [NEW] Dirty profile startup within 20% of clean profile startup (modulo extensions, plugins; on windows)
  • Shawn has a blog post with preliminary data
  • Session Restore is the biggest single culprit
  • Bugs filed for session restore bug 582005 and excessive cookie i/o bug 572223

Platform

GFX

• Going to make enabling Direct2D a Beta 4 blocker, with the aim of blocking beta 5 on hardware accelerated layers (D3D9/OpenGL).
• Direct2D gradient issues from last week are resolved.
• libjpeg-turbo integration shows up to 2.5x (+?) speedups of the JPEG decoding code. bug 573948 (jlebar)

JS

• JS fatvals landed
• ES5
  • Landings
    • Function.prototype.apply updated to accept any vaguely array-looking arguments, not just arrays or arguments alone
    • Back to ignoring a second argument to eval
    • Changes/refactoring of parseInt in bug 577536, but after further discussion the one semantic change has been reversed in bug 583925 (now back to status quo, but refactorings remain)
    • We now reject “0x” as valid numeric-literal syntax (an ancient bug)
  • Current work
    • Making getter/setter syntax require an exact number of arguments (zero/one) in bug 536472
    • Adding poison-pill properties for Function.prototype.caller and Function.prototype.arguments when the function in question is a strict-mode function; doing likewise for arguments.caller and arguments.callee in bug 514563 shortly after

Layout

• Retained layers
  • The most obvious retained-layers regressions fixed for beta3
  • Regression involving invalidation during reflow: may need to implement display-list-based invalidation for reflow to fix cleanly (brings perf wins too)
• Content/chrome rendering integration
  • Timothy needs help
• -moz-font-feature-settings has landed (“Land CSS control of lowlevel font features” done)
  • Needs Harfbuzz enabled to work

Video

• Improved software scaling ready to land
• Standards compliance fixes
  • ‘autobuffer’ -> ‘preload’
  • New load algorithm

Windows 7 Test Status

• Same remaining bugs as last week
• 2 out of 10 test suites with oranges
• The following bugs had no activity: bug 562955 (jimm) and bug 581734 (no one – smontagu suggested)
• We also have bug 582315 (felipe)

Tree Management

Roundtable

• Supported Hardware & OS list for Firefox 4 (beltzner)
  • many questions from OSX, JS and build teams about this
  • beltzner gathering data about usage in bug 582848 as well as from secondary sources
  • current proposals are:
    • drop OSX 10.4 (ease of coding, speedups from using better APIs)
    • drop PPC (no OOPP support planned, no JM support planned)
    • maintain W2K (no impact to coding schedule, not a lot of additional work)
    • add OSX 64-bit (OOPP support planned, major speedup increase)
    • do not add Windows 64-bit (no OOPP support planned)
    • Linux 64?
    • SSE2 is still a question
    • beltzner to make concrete proposals to dev-planning
    • beltzner to gather data on SSE2 usage, report to dev-planning separate from previous
• Question for Tree Management: (zwol)
  • Try server pushes are consistently taking 13-14 hours from push to last result email. Try server does not *appear* to be under heavier load than m-c (push rate eyeballs about the same); why is there such a huge difference?
    • Try push load for July is ~65% higher than mozilla-central
    • Build or test requests cannot be consolidated together on try, but can be consolidated during peak load on mozilla-central.
    • All builds on TryServer are full-clobber builds, not incremental builds like on mozilla-central
    • No fast build machines for try. Shuffling from mozilla-central and other project branches would slow those down.
    • If we ordered new hardware today, it would be at least 4 weeks to be racked up, installed, etc. No hardware has been ordered as yet.
    • Not all available testing slaves are currently enabled due to snow-ball effect during high jobs load. bug 563458