Perl old-run-webkit-tests issue seen on the build bots after a crash: Use of uninitialized value in numeric lt (<) at Tools/Scripts/old-run-webkit-tests line 1778.
Line 1778: # The crash log must have been created after this script started running. $crashLog = $newestCrashLog if -M $newestCrashLog < 0;
A little more context: # We assume that the newest crash log in matching the glob is the one that corresponds to the crash that just occurred. if (my $newestCrashLog = findNewestFileMatchingGlob($glob)) { # The crash log must have been created after this script started running. $crashLog = $newestCrashLog if -M $newestCrashLog < 0; } So $newestCrashLog itself is defined. I guess -M is returning undefined. The docs say this will happen if the file doesn't exist. Maybe the file was deleted between when it was found by findNewestFileMatchingGlob and when we tested it with -M? Joe, which bot did this happen on? Do you have a link to the test output that shows the error?
findNewestFileMatchingGlob can early return, which I think returns the undefined value. I saw this on the Qt bot.
Ah, that makes sense. Should be easy to fix then.
We should probably make captureSavedCrashLog bail if we don't have a $glob (which should be the case on Qt).
(In reply to comment #3) > findNewestFileMatchingGlob can early return, which I think returns the undefined value. I saw this on the Qt bot. Hm, but if findNewestFileMatchingGlob returns undefined, we shouldn't even get to line 1778, since $newestCrashLog will be undefined.
Ah, the "return unless @paths" check in findNewestFileMatchingGlob is buggy.
Created attachment 84423 [details] Don't try to capture crash logs on platforms that haven't implemented that feature
Comment on attachment 84423 [details] Don't try to capture crash logs on platforms that haven't implemented that feature View in context: https://bugs.webkit.org/attachment.cgi?id=84423&action=review r=me with that fix! Thanks! > Tools/Scripts/old-run-webkit-tests:1793 > + return unless $#paths; "$#paths" returns the last index in the array. So if there is 1 item, the last index will be 0, and this might still bail. As you mentioned on IRC, an alternative to getting the size of the array is "scalar(@paths)". That seems better.
Committed r80152: <http://trac.webkit.org/changeset/80152>
http://trac.webkit.org/changeset/80152 might have broken GTK Linux 64-bit Debug The following tests are not passing: fast/css-generated-content/005.html fast/css-generated-content/beforeAfter-interdocument.html fast/css-generated-content/no-openclose-quote.html fast/invalid/residual-style.html