1088231 - 3.3% Linux32 Dromaeo DOM regression on Inbound (v.36) Oct 22 from push 59be1e005220

Status: RESOLVED WONTFIX

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

Description

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

graph server confirms a regression in dromaeo dom:
http://graphs.mozilla.org/graph.html#tests=%5B%5B73,131,33%5D%5D&sel=1413933686624.3416,1414036963282.818&displayrange=7&datatype=running

here is a view on tbpl:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=ede829d6aa61&tochange=7f49fd6cc976&jobname=Ubuntu%20HW%2012.04%20mozilla-inbound%20talos%20dromaeojs

here is the changeset:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=59be1e005220

this doesn't seem to affect other platforms.

Comment 1

[Boris Zbarsky [:bzbarsky]]

None of the code I removed in bug 1019191 ran during dromaeo-dom.

How was the changeset in "here is the changeset" determined?  The tbpl view shows a number of js engine changes, right?

Comment 2

[Boris Zbarsky [:bzbarsky]]

Also, the chart doesn't show an obvious regression over the entire highlighted range...

Comment 3

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

if you look at the tbpl view, you can click on the 'd' and see the value for dromaeo DOM.  we have a noticeable drop when we get to 59be1e005220.  Let me retrigger some to help highlight or disprove it:)

Comment 4

[Boris Zbarsky [:bzbarsky]]

I would be somewhat, but not incredibly, surprised if removing dead code changes performance characteristics... ;)

Comment 5

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

the retriggers show a regression with this push (~900 before, ~880 after), but I see a win (~890) from this push (2 later):
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=d6d519cc677b

as this appears to be linux32 and the net regression is very small (if any at all now), I recommend closing this as wontfix.

Comment 6

[Boris Zbarsky [:bzbarsky]]

Worksforme.  Especially since the changeset the win is from doesn't change any code used by dromaeo-dom either.  :(