[webkit-dev] Getting more buildbots green
ddkilzer at webkit.org
Sat Jan 3 17:00:23 PST 2009
> To me it seems like the result of some tests have impact on the others.
This usually indicates some state isn't being reset (in DumpRenderTree or in WebKit) between tests for the GTK+ port. Individual test results should not rely on running previous tests. (The --random and --reverse flags on run-webkit-tests stress these conditions. I don't think anyone has tried running the tests with these flags recently, though.)
> I'm not sure if we would win anything by making the buildbot green...
It's a psychological issue. If the buildbots are always red, developers start expecting them to be red and then stop paying attention to the details of why they're red.
From: Holger Freyther <zecke at selfish.org>
To: webkit-dev at lists.webkit.org
Sent: Saturday, January 3, 2009 3:42:52 PM
Subject: Re: [webkit-dev] Getting more buildbots green
> I'd love to set up the skipped lists properly and fix enough bugs so
> that all our bots are green. Can you help? Maybe there are already bug
> reports about some of these.
Regarding the skipped list on Gtk+. There are some tests that work on my
machine that are failing on the buildbot. I would like to understand the
difference before adding them to the skipped list. On the other hand we end up
in mess when skipping some of the tests from within fast. To me it seems like
the result of some tests have impact on the others. E.g. when starting to Skip
one dom test we might need to add ~5 more and could unskip some others.
I'm committed to find the difference and get it green but time is the issue as I
have university stuff first, then paid work with varying degree of load and then
webkit... I'm not sure if we would win anything by making the buildbot
and I have no idea why (and the TZ on the buildbot is set to Cupertino time
and the LC_ALL to C).
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the webkit-dev