<Regression> Mozilla-Inbound-Non-PGO - TResize - Ubuntu HW 12.04 - 18.3%

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - TResize - Ubuntu HW 12.04 - 18.3% increase
--------------------------------------------------------------------------------
Previous: avg 18.148 stddev 1.439 of 12 runs up to revision 7ccebf05488c
New : avg 21.470 stddev 0.279 of 12 runs since revision 5a86fde943e6
Change : +3.322 (18.3% / z=2.309)
Graph : http://mzl.la/14lwha6

Changeset range: http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=7ccebf05488c&tochange=5a86fde943e6

Changesets:
* http://hg.mozilla.org/integration/mozilla-inbound/rev/6490e2abb8b3
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Only use JSOPTION_UNROOTED_GLOBAL for DOM JSContexts. r=mccr8

We don't cycle collect any other kind, so there's no difference between marking
them gray and letting the JS engine mark them black. This also gets rid of that
nasty code which reset the flag, which billm theorizes has to do with faulty
logic in ContextHolder when creating ephemeral sandbox cxes. The assertion in
this patch should catch us if anything goes wrong.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/001f423a94e8
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Add an API to determine if we're marking gray. r=billm

This allows us to assert that certain callbacks get invoked during the gray
phase of a marking GC.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/c800b53263d2
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Explicitly trace outer windows, rather than doing it via JSContext iteration. r=mccr8
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/bb9646daf4c3
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Explicitly traverse outer windows, rather than doing it via JSContext iteration. r=mccr8

Note that this machinery will go away in further patches. But we want to be
deliberate about each of these steps.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/6ca662ba7c55
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Add an API to opt out of default compartment objects. r=luke

We need this explicit API because otherwise the default compartment object will
get set implicitly during the first call to InitClasses.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/14df512b792a
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Stop depending on the default compartment object for DOM JSContexts. r=mccr8
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/8e4bd1c50df4
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Have nsGlobalWindow go through nsJSContext to access the outer. r=mccr8
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/6a1f5a4594ae
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Stop storing the window proxy on the JSContext. r=mccr8

We need to simultaneously report the new Heap Pointer to the cycle collector,
and remove the old machinery for cycle-collecting the default object via the
JSContext.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/cd8526f29c2f
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Remove JSOPTION_UNROOTED_GLOBAL. r=luke,mccr8

\o/
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/32c995dfd080
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Remove xpc_UnmarkGrayContext. r=mccr8

Default compartment objects are never gray anymore.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/156e8e401a83
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Remove check for non-WindowProxy objects on an scx. r=mccr8
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/dd6d206a9a8a
: Bobby Holley <bobby...@gmail.com> - Bug 899367 - Make GetNativeGlobal consumers call GetWindowProxy and rm the former. r=mccr8

Some of the uses of this are a bit nonsensical now, but that's a problem for
another day.
: http://bugzilla.mozilla.org/show_bug.cgi?id=899367

* http://hg.mozilla.org/integration/mozilla-inbound/rev/0e9fc39801fc
: Georg Fritzsche <georg.f...@googlemail.com> - Bug 899080 - Let all plugins except Flash default to click-to-play. r=bsmedberg
: http://bugzilla.mozilla.org/show_bug.cgi?id=899080

* http://hg.mozilla.org/integration/mozilla-inbound/rev/46a5c7cf1ec4
: Georg Fritzsche <georg.f...@googlemail.com> - Bug 899080 - Test fixup part 1 - Always enable test-plugins for reftests. r=jmaher
: http://bugzilla.mozilla.org/show_bug.cgi?id=899080

* http://hg.mozilla.org/integration/mozilla-inbound/rev/3bcbfbc9591b
: Georg Fritzsche <georg.f...@googlemail.com> - Bug 899080 - Test fixup part 2 - dom/plugins. r=bsmedberg
: http://bugzilla.mozilla.org/show_bug.cgi?id=899080

* and 10 more

Bugs:
* http://bugzilla.mozilla.org/show_bug.cgi?id=911771 - Unable to start Firefox under gdb on Mac
* http://bugzilla.mozilla.org/show_bug.cgi?id=899367 - Stop Cycle Collecting JSContexts
* http://bugzilla.mozilla.org/show_bug.cgi?id=899080 - Make plugins default to click-to-play
* http://bugzilla.mozilla.org/show_bug.cgi?id=906601 - Improper tag parsing causing crash
* http://bugzilla.mozilla.org/show_bug.cgi?id=904533 - APZ panning on about:start tab is wonky
* http://bugzilla.mozilla.org/show_bug.cgi?id=902075 - Add telemetry for usage patterns of click-to-play plugins

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - Paint - Ubuntu HW 12.04 - 9.65% increase
------------------------------------------------------------------------------
Previous: avg 164.670 stddev 3.467 of 12 runs up to revision 7ccebf05488c
New : avg 180.558 stddev 5.220 of 12 runs since revision 5a86fde943e6
Change : +15.887 (9.65% / z=4.583)
Graph : http://mzl.la/14lwhXB

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - Tab Animation Test - Ubuntu HW 12.04 x64 - 44.3% increase
-----------------------------------------------------------------------------------------------
Previous: avg 3.732 stddev 0.013 of 12 runs up to revision 7ccebf05488c
New : avg 5.385 stddev 0.057 of 12 runs since revision 5a86fde943e6
Change : +1.653 (44.3% / z=131.802)
Graph : http://mzl.la/14lyB0S

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - SVG-ASAP - Ubuntu HW 12.04 x64 - 10.7% increase
-------------------------------------------------------------------------------------
Previous: avg 488.291 stddev 6.621 of 12 runs up to revision 7ccebf05488c
New : avg 540.636 stddev 11.483 of 12 runs since revision 5a86fde943e6
Change : +52.345 (10.7% / z=7.905)
Graph : http://mzl.la/14lyAtB

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - SVG-ASAP - Ubuntu HW 12.04 - 8.38% increase
---------------------------------------------------------------------------------
Previous: avg 611.212 stddev 4.903 of 12 runs up to revision 7ccebf05488c
New : avg 662.451 stddev 7.960 of 12 runs since revision 5a86fde943e6
Change : +51.239 (8.38% / z=10.450)
Graph : http://mzl.la/14lyDFX

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - Tab Animation Test - Ubuntu HW 12.04 - 33.4% increase
-------------------------------------------------------------------------------------------
Previous: avg 4.149 stddev 0.039 of 12 runs up to revision 7ccebf05488c
New : avg 5.533 stddev 0.127 of 12 runs since revision 5a86fde943e6
Change : +1.384 (33.4% / z=35.890)
Graph : http://mzl.la/14lyDFN

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - TResize - Ubuntu HW 12.04 x64 - 23.4% increase
------------------------------------------------------------------------------------
Previous: avg 15.149 stddev 0.889 of 12 runs up to revision 7ccebf05488c
New : avg 18.699 stddev 0.200 of 12 runs since revision 5a86fde943e6
Change : +3.550 (23.4% / z=3.993)
Graph : http://mzl.la/14lyF0r

[Matt Brubeck]

Several benchmarks regressed significantly in the changeset range below:
tresize, tpaint, tsvg, tart. It looks like the regressions may be
Linux-only. Does anyone have any idea which patches might be responsible?

Here's some documentation of the individual benchmarks:
https://wiki.mozilla.org/Buildbot/Talos/Tests

[Matt Brubeck]

On 9/5/2013 8:46 AM, Matt Brubeck wrote:
> Several benchmarks regressed significantly in the changeset range below:
> tresize, tpaint, tsvg, tart. It looks like the regressions may be
> Linux-only. Does anyone have any idea which patches might be responsible?

Botond pushed to Try to determine whether his patch is at fault:
https://tbpl.mozilla.org/?tree=Try&rev=d1bad1790560

[Matt Brubeck]

On 9/5/2013 9:53 AM, Matt Brubeck wrote:
> Botond pushed to Try to determine whether his patch is at fault:
> https://tbpl.mozilla.org/?tree=Try&rev=d1bad1790560

Based on this Try push, it looks like bug 904533 was at fault:
https://bugzilla.mozilla.org/show_bug.cgi?id=904533#c43