Regression: Tp4 (Private Bytes) increase 4.15% on Linux Firefox
Previous results: 243605000.0 from build 20091118042303 of revision 4efe8dde9f52 at 2009-11-18 04:39:00 on talos-rev2-linux03 run # 0
New results: 253707000.0 from build 20091118043757 of revision c61a7e1570c0 at 2009-11-18 05:25:00 on talos-rev2-linux15 run # 0
Suspected checkin range: from revision 4efe8dde9f52 to revision c61a7e1570c0
_______________________________________________
dev-tree-management mailing list
dev-tree-...@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-tree-management
Improvement: Ts, MED Dirty Profile decrease 0.70% on Vista Firefox
Previous results: 498.263 from build 20091118062958 of revision 4efe8dde9f52 at 2009-11-18 08:47:00 on talos-rev2-vista10 run # 0
New results: 494.789 from build 20091118051406 of revision c61a7e1570c0 at 2009-11-18 07:08:00 on talos-rev2-vista04 run # 0
That's pretty odd, since all the code Henri is touching is preffed off
by default, no?
-Boris
> This seems to have been caused by your latest push, Henri.
This is very odd. The non-test code in that push is off by default and shouldn't get run on tp4 at all. The only bits that run with the old parser are checks for the pref and a couple of no-op stubs on nsParser. The no-op stubs only run when a script element is executed, so it seems implausible that no-op virtual methods run upon <script> could cause a 4.15% increase.
Thank you for bringing this to my attention. I'll start right away figuring out what the cause of the slowdown is.
> On 2009-11-18, at 4:32 PM, Nobody wrote:
>
>> Regression: Tp4 (Private Bytes) increase 4.15% on Linux Firefox
>>
>> Previous results: 243605000.0 from build 20091118042303 of revision 4efe8dde9f52 at 2009-11-18 04:39:00 on talos-rev2-linux03 run # 0
>>
>> New results: 253707000.0 from build 20091118043757 of revision c61a7e1570c0 at 2009-11-18 05:25:00 on talos-rev2-linux15 run # 0
>>
>> Suspected checkin range: from revision 4efe8dde9f52 to revision c61a7e1570c0
--
Henri Sivonen
hsiv...@iki.fi
http://hsivonen.iki.fi/
Note that this is not a slowdown. It's a 4% increase in memory used.
-Boris
I've pushed a patch that avoids the creation of the HTML5 parser thread
when the HTML5 parser hasn't been enabled. (I had prepared the patch as
part of investigating this issue, but I landed it to fix an orange
situation.)
http://hg.mozilla.org/mozilla-central/rev/2a0941f8bef9
Until the Talos boxes cycle, I'm unable to tell if this patch alone
fixes the tp4 (Private Bytes) regression. However, based on my TryServer
testing, it seems that at least taken together with an implementation of
the proposal at
http://groups.google.com/group/mozilla.dev.platform/browse_thread/thread/
3a3785a83571bfeb/44d733fd112ce92e
it fixes the regression.
(I implemented that proposal locally in order to investigate this issue.)
> I've pushed a patch that avoids the creation of the HTML5 parser thread
> when the HTML5 parser hasn't been enabled. (I had prepared the patch as
> part of investigating this issue, but I landed it to fix an orange
> situation.)
> http://hg.mozilla.org/mozilla-central/rev/2a0941f8bef9
It seems that avoiding the thread creation fixed the problem:
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1258636981.125863800
5.796.gz