[webkit-dev] Chromium PLT regression

Adam Barth abarth at webkit.org
Wed Oct 5 08:00:29 PDT 2011

On Tue, Oct 4, 2011 at 10:46 PM, Dan Bernstein <mitz at apple.com> wrote:
> On Oct 4, 2011, at 10:37 PM, Adam Barth wrote:
>> About a week ago, the Chromium project measured a PLT regression on
>> Windows, Mac, and Linux:
>> https://bugs.webkit.org/show_bug.cgi?id=69238
>> I don't know whether the regression affects any other ports, but
>> according to the Chromium performance bots, the regression occurred in
>> this range:
>> http://trac.webkit.org/log/trunk/Source?rev=96091&stop_rev=96066&verbose=on
>> Normally, the Chromium project has a bot continuously running the PLT
>> against WebKit trunk, but, due to a configuration error, that bot was
>> been offline for about 10 days, which, unfortunately, includes the
>> time period in question.  The bot is now back online:
>> http://build.chromium.org/f/chromium/perf/linux-release-webkit-latest/intl1/report.html?history=150&rev=-1
>> I'm not able to reproduce the regression locally.  Sometime in the
>> next couple days, when the tree is quiet (which probably means at
>> night PST), I'd like to try to hunt down which revision in that range
>> caused the regression by rolling out the patches.  If I roll out your
>> patch as part of this process, do not worry.  I'll roll it back in
>> shortly thereafter.  My hope is that this experiment will allow us to
>> isolate the exact revision that caused the regression.
>> If anyone is able to reproduce the issue locally, that would certainly
>> be better than me making a mess by rolling patches in and out of the
>> main tree.
> Hi Adam,
> I think that configuring your bot to build and test old the revisions off trunk would be far better. Or, if for some reason that is impossible, then configuring it to work against a local repository. Or, if for some reason that is impossible, and you absolutely have to use svn.webkit.org for this purpose, then configuring it to build and test a branch that you will create for this purpose and on which you will run the experiment.
> It is almost unbelievable that the only way for you to track down a performance regression is to make changes to the source tree.

Ok.  I'l ask the Chromium infrastructure folks if those things are possible.


More information about the webkit-dev mailing list