[Webkit-unassigned] [Bug 83411] All Mac bots occasionally fail tons of media/ sputnik/ and svg/ tests

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=83411


Dirk Pranke <dpranke at chromium.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |eric.carlson at apple.com,
                   |                            |tonikitoo at webkit.org




--- Comment #8 from Dirk Pranke <dpranke at chromium.org>  2012-04-09 18:41:41 PST ---
(In reply to comment #7)
> My guess is that there's one test that behaves badly (a media test?), and causes the off-by-one for all the other tests run in the same shard.

I think Simon's hypothesis is a good one. I spot-checked a few logs (from builds 7423, 7428, and 7437 on the Lion Release bot) and they all appear to have one worker that starts failing every test starting with media/nodesFromRect-shadowContent.html, so it might be that test.

If you then look at the results for 7437, http://build.webkit.org/results/Lion%20Release%20(Tests)/r113622%20(7437)/results.html , you can see that the -actual output for nodesFromRect-shadowContent is repeated in the next failing test, media/remove-from-document-before-load.html . And, interestingly, the -expected for that test says "PASS: A crash did not occur when the media element was removed before loading.".

So, I'm guessing that that test is actually crashing, and we're not noticing it, and somehow the internal output buffers in NRWT aren't getting emptied, so we end up reusing the same output twice and everything goes to hell after that.

Note that that is the test that is being moved in r113371, so I think we have a pretty good suspect :).

Maybe try skipping that test and see if the problem goes away? 

There's almost certainly also a bug in NRWT that maybe I introduced a few weeks ago when I was reworking the crash log code, so I can start looking for that now that I have a theory, as well.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.