Mozilla Status Board

mh+mozilla@glandium.org

Current Status - 2014-10-31

Done:

  • Reviewed bug 1089832, bug 1089855, bug 1045958, bug 1090088, bug 1090512, bug 1039902, bug 883254
  • Reviewed bug 1059797, and addressed review comments myself to speed up the process, because Android builds have been erratic for long enough, especially on automation. Landed.
  • bug 1077148: Straightened up for review. The rooting hazard failures ended up being probably triggered by a combination of bug 1083686 and a race condition that was already there, that the patches for bug 1077148 made worse by changing the build dependencies. Landed.
  • bug 1090107: Define stlport dependencies in moz.build. Landed.
  • bug 1090111: Remove now unnecessary dependency for sqlite in recurse.mk. Landed.
  • bug 1090114: Possible race condition when building logalloc replay on mac. Landed.
  • bug 1090601: Pushed js shell to tooltool.
  • bug 965151: Setup tooltool in valgrind builds. Landed but doesn't work.
  • bug 1090668: "ElfLoader.cpp:692:23: error: 'AT_PHDR' was not declared in this scope" with NDK r10c if target version is greater than 19. Landed.
  • bug 947462: Use mozilla-centos6-x86_64 chroots for linux32 b2g builds.
  • bug 1091118: Subtle intermittent build failure on android after bug 1059797. Landed.
  • bug 1091383: Move delayload logic entirely in moz.build frontend code.
  • bug 1091384: Remove EXPAND_LIBNAME and affiliated
  • Diagnosed bug 1091377 as a NDK breakage.
  • bug 1091505: Run subconfigures in parallel .
  • Blatantly failed to blog anything.

Older Posts

Member of Projects

Team Posts

mh+mozilla@glandium.org today

Done:

  • Reviewed bug 1089832, bug 1089855, bug 1045958, bug 1090088, bug 1090512, bug 1039902, bug 883254
  • Reviewed bug 1059797, and addressed review comments myself to speed up the process, because Android builds have been erratic for long enough, especially on automation. Landed.
  • bug 1077148: Straightened up for review. The rooting hazard failures ended up being probably triggered by a combination of bug 1083686 and a race condition that was already there, that the patches for bug 1077148 made worse by changing the build dependencies. Landed.
  • bug 1090107: Define stlport dependencies in moz.build. Landed.
  • bug 1090111: Remove now unnecessary dependency for sqlite in recurse.mk. Landed.
  • bug 1090114: Possible race condition when building logalloc replay on mac. Landed.
  • bug 1090601: Pushed js shell to tooltool.
  • bug 965151: Setup tooltool in valgrind builds. Landed but doesn't work.
  • bug 1090668: "ElfLoader.cpp:692:23: error: 'AT_PHDR' was not declared in this scope" with NDK r10c if target version is greater than 19. Landed.
  • bug 947462: Use mozilla-centos6-x86_64 chroots for linux32 b2g builds.
  • bug 1091118: Subtle intermittent build failure on android after bug 1059797. Landed.
  • bug 1091383: Move delayload logic entirely in moz.build frontend code.
  • bug 1091384: Remove EXPAND_LIBNAME and affiliated
  • Diagnosed bug 1091377 as a NDK breakage.
  • bug 1091505: Run subconfigures in parallel .
  • Blatantly failed to blog anything.

jseward@acm.org 2 days ago

Done:

  • Tried to make sense of bug 788974 (make LUL unwinding work with elfhack). Finally got a working fix. Also fixed V's EXIDX reader in the process.
  • Investigated Fx/V hangs on OSX. Ironically the hang detector causes V to hang. Instituted fixes in V and workarounds in Fx.
  • Refreshed and committed a memory-map resync reduction fix for V/OSX that had been lying around for a couple of years. With this in place, V can run Fx on OS 10.9 without an enormous performance overhead compared to Linux.
  • Tried and entirely failed to install OSX 10.10. Urrr.
  • Many more V/Mochitest and crashtest runs/investigation/fixes, including trying out e10s:
    • bug 1077031 - uninit value(s) involving IsFloatLiteral. Is a false positive.
    • bug 1074697 - nsDisplaymtdBorder::GetBounds does not set out parameter *aSnap. FIXED.
    • bug 1078275 - With e10s enabled, TargetConfig::orientation is used uninitialised (no fix yet)
    • bug 1078211 - Uninitialised value use in DoUnpremultiplicationCalculation_SIMD (no fix yet)
    • bug 1090220 - Uninitialised value use in APZCCallbackHelper::UpdateRootFrame. FIXED.
  • V: Looked into the possibility of avoiding the 10%-15% performance loss caused by the recent need to use --vex-iropt-register-updates=allregs-at-mem-access.

Next:

  • Finish up bug 788974
  • Continue with uninit value runs and fixes, especially e10s
  • Get OSX 10.10 installed, get V working with that
  • Try out the V PX-overhead-avoidance hack

erahm@mozilla.com 4 days ago

Done:

Last week:

  • Examined opportunities for improving startup performance in j2me.js
  • Found and triaged issue w/ TypedArrayBuffers on m-c on flame
  • Found decompression of classfiles in jars dominated startup time, fixed support for uncompressed jars, found ~20% improvement
  • Improved lookups of localized strings my mapping field IDs to string values directly
  • Looked into processing of embedded strings in classfile constant tables, tested out using TextDecoder directly and found a significant improvement

Last week - 1:

  • Helped identify performance issues w/ the famo.us engine on b2g devices
  • Gave recommendations to use TypedArrays for matrix operations, improve generations of matrix transforms, look into ways not to update css attribues one at a time which can lead to multiple reflows during one requestAnimationFrame call

Next:

  • Continue work on j2me.js performance
  • Talk to overholt and johns about AWSY
  • Write some webaudio memory reporting tests

Coordination:

  • Review from wtc

nfroyd@mozilla.com 4 days ago

Done:

  • Added networking syscalls to x86-64 rr, #1350, #1351, #1353
  • Fixed recording mmap(2) on x86-64, #1352
  • Fixed replaying zero-length reads, #1357
  • Recording x86-64 Firefox mochitests now works with rr!
  • Moved some of the assembly language fiddling ("monkeypatching") to Python, #1357, along with stupid thinko fix, #1358
  • Spent way too long figuring out how to do previous with recursive C++ templates. But learned that the compiler is an excellent detector of needed base cases with recursive templates!
  • Made rr test more robust in the presence of rr bugs, #1359
  • Wrote support for buffering syscalls on x86-64, #1360. This speeds up recording and replaying, but tests indicate that it's not quite right yet, tricky debugging required. Also, we don't yet buffer important syscalls like read(2), write(2), and recv(2).
  • Read a nice series of blog posts about porting a spectral path tracer from C++ to Rust.
  • Reviews, just a handful.

Next:

  • rr debugging.
  • rr syscall buffering for read(2) et al.
  • Try to get back to some Gecko patches.

nnethercote@mozilla.com 5 days ago

Done:

Landed some DMD improvements:

Landed several patches that each reduce heap churn by 10s or 100s of MiBs during start-up and brief browsing, found with DMD:

  • Remove get_content_type_from_mime_type(), and replace calls to it with calls to g_content_type_from_mime_type() (bug 1084193).
  • Reduce heap churn caused by TempAllocator (bug 1085740).
  • Reduce heap churn involving |input_overflow_buf_| (bug 1087834).

Landed some code clean-ups:

Other:

Next:

  • More profiling with DMD, esp. on e10s builds.
  • Finish memory tools documentation update.