Performance/MemShrink/Meetings/2011-07-19

Announcements

TBPL (XHR) leak was fixed! https://bugzilla.mozilla.org/show_bug.cgi?id=671053
- Did this fix the other sites known to cause zombie compartments?
  - We haven't tested yet
- Maybe there's another bug that still causes TBPL to leak?
"Actually sending memory-pressure notifications when we get low on memory" patch is coming along (jlebar, bug 670967). Need to test on a wider variety of low-end machines.
[pbiggar] jemalloc on Mac (bug 414946).
- RSS regressions on 10.5, hangs on 10.7. Planning to fix by turning it off on 10.5, 10.7, but there are build system issues to deal with. Plan is to update to latest upstream jemalloc, at least for Mac, and see if that helps.
- Turning on decommit on Mac helps memory a lot but destorys perf (not just on benchmarks). Likely to use madvise instead of decommit. Long-term, perhaps "decommit on idle".

bug 563700 - Reduce about:memory's "heap-unclassified" number

DOM memory reporter coming along. Expecting 3-5% (of total memory?). Accounted for 3.8% so far? (nodes, attributes, documents, windows)
Necko cache reporter just landed!
Text runs
Layout arenas

Bug list - 8 bugs

Several user-reported bugs we couldn't reproduce. Decided these would be P3 (but not INCO) for now.

bug 671352, Split chrome into multiple compartments for better accounting of JS memory used by chrome code
- make a separate compartment for each global (one for each XUL window or JS component)
bug 672439, Avoid duplicate names for system compartment reporters
bug 672443, [Jetpack] Too many compartments
- Feeling that Jetpack is lower-quality, different-style, and isolated-people from the rest of Firefox :(
- Ehsan has investigated and will split this into a bunch of bugs
- mrbkap may interact with Jetpack folk more in the future

4 JS, 4 others