979875 - dromaeo dom regression on mozilla beta on Feb 27th

Status: RESOLVED WONTFIX

(Core :: DOM: Core & HTML, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

regression found on dev.tree-management:
https://groups.google.com/forum/#!topic/mozilla.dev.tree-management/hdb9WWTpVZs

here is a view on graph server:
http://graphs.mozilla.org/graph.html#tests=[[73,53,21]]&sel=none&displayrange=30&datatype=running

I did some retriggers and it shows a steady change in the numbers:
https://tbpl.mozilla.org/?tree=Mozilla-Beta&jobname=Rev4%20MacOSX%20Snow%20Leopard%2010.6%20mozilla-beta%20talos%20dromaeojs&fromchange=3d2e069fb893&tochange=860e76ec9fe5

There are 6 different patches on here (2 seem completely unlikely).  Any thoughts on what might be the cause?

Comment 1

[Boris Zbarsky [:bzbarsky]]

Did this happen on other platforms?  http://perf.snarkfest.net/compare-talos/index.html?oldRevs=525d9af770e7&newRev=eb3b0ef6c34d&submit=true seems to suggest not?

It's theoretically conceivable that bug 976920 might have affected this somehow...

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

I don't see any issues on graph server related to other platforms, this is only seen on osx 10.6 from my point of view.

Comment 3

[Randell Jesup [:jesup] (needinfo me)]

The Webrtc items in here can't affect Dromaeo (DataChannels changes and Android OpenSL changes)

Comment 4

[Joel Maher ( :jmaher ) (UTC -8)]

that leaves:
https://hg.mozilla.org/releases/mozilla-beta/rev/8934420cbdba – Bug 834678 - Ensure correct update of lastPC_ for MInstructions which add OOL code. r=jandem
https://hg.mozilla.org/releases/mozilla-beta/rev/8483f824c151 – Bug 951827 - Revert the key piece of bug 932100 due to external code interactions
https://hg.mozilla.org/releases/mozilla-beta/rev/dc567976528c -  Bug 976920 - Mostly back out bug 932322 for now; only define the unforgeable properties on the window object itself.
https://hg.mozilla.org/releases/mozilla-beta/rev/078261348763 - Bug 946502 - Rollup of patches for bugs 946502, 967587, 969354 and 976412

bz, you mentioned "Bug 976920 - Mostly back out bug 932322 for now; only define the unforgeable properties on the window object itself." might be a cause, did bug 932322 give us a performance win on dromaeo dom?

Comment 5

[Boris Zbarsky [:bzbarsky]]

Not that I recall, no.

Comment 6

[(Away)]

> https://hg.mozilla.org/releases/mozilla-beta/rev/8483f824c151 – Bug 951827 -
> Revert the key piece of bug 932100 due to external code interactions

FWIW that change moves some work from one point in the startup path to a slightly later point in the startup path; that work happens before Dromaeo starts in either case.

Comment 7

[Joel Maher ( :jmaher ) (UTC -8)]

Here is a view on datazilla:
https://datazilla.mozilla.org/?start=1392866979&stop=1393723405&product=Firefox&repository=Mozilla-Beta&os=mac&os_version=OS%20X%2010.6.8&test=dromaeo_dom&project=talos

It appears we only affect the attr.html test:
http://hg.mozilla.org/build/talos/file/e2dd770e2d4c/talos/page_load_test/dromaeo/dom-attr.html

Maybe that will help narrow this down.

Comment 8

[Joel Maher ( :jmaher ) (UTC -8)]

as this was on mozilla-beta, when we did merge day our numbers got much better:
http://graphs.mozilla.org/graph.html#tests=[[73,53,21]]&sel=none&displayrange=30&datatype=running

sadly though this made it to release.  I am going to mark this bug as wontfix, since we have already released the code and made significant improvements in this test on future branches.