Platform/2010-04-20

« previous week | index | next week »

Notices / Schedule

• Firefox 3.6.4 beta build 1 went out late Friday, April 16th
• Schedule is at risk due to newly found issues (see blocker section)
• Firefox 3.5.10 is getting final submissions this week
• Will likely want a build of 3.5.10 and 3.6.4 on Monday April 26th (after RelEng Toronto work week)

Blocker Report

• There are 8 Firefox 3.6.4 open blockers. One was a missed fix when merging with Lorentz, the rest are newly found OOPP issues. Hopefully most/all will be resolved for a build Monday
• There are 2 Firefox 3.5.10 open blockers. They both have patches and should be landed this week

Browser / Front End

You know how I always say that planet firefox has the details? Still does!

Some interesting tidbits:

• Some of the JEPs the team has committed to implementing are already nearing completion, including Context Menu and Page Workers.
• Speaking of Jetpack, Felipe's got a post about setting up a jetpack development environment.
• Geo-Notification work continues on apace, too.

GFX Update

• • WOFF spec formally submitted to W3C!

Layout Update

• Timothy Nikkel landed lazy frame construction --- significant performance improvements for certain kinds of scripts and benchmarks, mainly those that add and remove lots of DOM elements
• Daniel Holbert has mostly finished <animateMotion>, moving on to adding support for SVG images
• Matthew Gregan blogged about playing Ogg Theora using DSP decoding and GL on the N900: full-screen (800x480) 26 FPS with CPU 80% idle
  • Needs some work to integrate into Gecko, but currently lower priority than other work
• All but one of the reproducible sg:critical bugs have patches

• Matt Woodrow's retained path patches are being reviewed; small win on path-heavy benchmark on Mac, hopefully much larger wins on Windows/D2D (working on that now)
• Michael Ventnor landed major performance improvements for box-shadows with blur
• Last week, Ehsan landed lazy editor initialization; speeds up loading of pages with lots of text controls

Content Update

• bnewman refactoring XPCWrappedNative::CallMethod (bug 559962).
• bent wrote ChromeWorker, a DOM worker usable from chrome that has access to jsctypes!
• Honza Bambas (mayhemer) is back after dealing with hand surgery, still recovering, but able to start working again.
• Peterv working on linking xpconnect in with the layout library in order to make stuff faster.

JS

• obsolete getter/setter syntaxes removed, killed 200 lines of code and about 1000 lines of tests for stuff nobody used (except Jesse)
• nice speedups observed for iteration work, now facing the fuzz-test gauntlet
• fuzzing of new ES5 work started
• more tracing wins TM and M-C, more v8 wins landed on JM

Electrolysis

bsmedberg maintaining a couchdb of all crashes for beta/RC versions of Firefox, which allows more advanced queries of crashes and hangs by various criteria. If you have specific questions you'd like to ask it, please send bsmedberg an email.

Faster shared-memory memory-mapped drawing now being used for Fennec. Other work continues apace.

Initial API schema for multi-process jetpack posted to the jetpack mailing list. Not a lot of response yet on the technical details. JEP implementations will have to change significantly, so certain JEPs may be available only in Firefox.next, to avoid implementing each one twice (once for 3.6 and once for FF.next).

Startup Performance

• No major landings or updates
• Performance roundup from last week (4/16)

• View snapshot results for startup and all other tests.
• More details on the project page.

Estimated win	Bug #	Summary	Owner	Status	Notes
~10%	"Dirty Profile" Startup Reduction	Constant startup relative to profile size	Shawn Wilsher
significant	bug 558200	Extension profiler platform api	Steve Fink		There is an easy way to provide feedback to extension developers to help them figure out why their extension is misbehaving. We need to implement that ASAP, so typical firefox installs(with lots of extensions) can be fast

Breakpad

• We may have a new reviewer for the 23 patches in Google's queue (Mac CFI, Mac 64-bit)
• bug 560377 (RHEL 5.4 build problems) landed, allowing fix for...
• bug 558880 (newlines in module names break analyses) to be brought into staging

Tree Management

• Need help with Unittest oranges
  • 10.6 debug builds have oranges - need help with bug 558496 bug 558947 bug 559623 bug 558094 bug 558098
  • fedora12x32, fedora12x64: opt has known oranges - need help with bug 553365
  • until oranges fixed, we're running unittest-on-builders and also unittest-on-minis.
  • 10.6 opt builds green, 10.5 opt builds green, no longer being run on build machines.
• Possible compiler switch for linux (move from GCC4.3.3 -> GCC4.4 / GCC4.5?)
  • which branches? which version? binary compat concerns?
  • please comment in dev.platform or bug 559964

Roundtable

• Test Firefox 3.6.4, especially on older hardware