It fails, or doesn't fail, depending on which tests ran prior to it. But more importantly, the test is mostly useless. I cannot recall a time when it failed in a manner that helped us pinpoint a bug. But I can recall many times that it failed spuriously, leading to bot redness and false alarms. I will skip it on the Mac port. But I strongly believe that the better solution is to eliminate this test from our repertoire.
Most recent failure, which I don't feel like investigating because I already know it'll be a red herring: http://build.webkit.org/results/Apple%20Lion%20Debug%20WK1%20(Tests)/r122159%20(765)/fast/js/global-constructors-pretty-diff.html
I thought Arv had already posted some patches to remove these. But I may be confused.
Skipped on Mac in http://trac.webkit.org/changeset/122165
I've personally been responsible for regressions in this test in the past -- but I still think it has value. It should prevent unintended additions of globally visible constructors. Is there a less fragile way to get the same result?
We should either remove it or pick a handful of constructors to test.
It looks like many ports now ignore failures in this test (certainly mac, chromium, win, wincairo, qt). I just added a new global constructor (bug 96295) and got no failures in EWS as a result.
This test was deleted in bug 110573.