Bug 56109 – run-webkit-tests sometimes times after fast/text/large-text-composed-char-dos.html fails (due to PrettyPatch hanging?)

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Adam Roben (:aroben)]

run-webkit-tests is sometimes hanging and getting killed by buildbot on Windows XP Debug (Tests). It seems to (always?) happen after fast/text/large-text-composed-char-dos.html fails. Here are some examples of it happening on all three of our slaves:

http://build.webkit.org/builders/Windows%20XP%20Debug%20%28Tests%29/builds/26168/steps/layout-test/logs/stdio
http://build.webkit.org/builders/Windows%20XP%20Debug%20%28Tests%29/builds/26182/steps/layout-test/logs/stdio
http://build.webkit.org/builders/Windows%20XP%20Debug%20%28Tests%29/builds/26172/steps/layout-test/logs/stdio

[Jessie Berlin]

<rdar://problem/9114976>

[Adam Roben (:aroben)]

It looks like the pretty diff output file is getting held open: http://build.webkit.org/builders/Windows%20XP%20Debug%20%28Tests%29/builds/26164/steps/archive-test-results/logs/stdio

This indicates that prettify.rb is hanging.

[Adam Roben (:aroben)]

The fact that the test results are failing to get uploaded is covered by bug 55581.

[Adam Roben (:aroben)]

I logged into apple-windows-13 and ruby.exe was indeed still running.

[Adam Roben (:aroben)]

I can reproduce on my own machine.

[Adam Roben (:aroben)]

Killing ruby gave me this output:

./Websites/bugs.webkit.org/PrettyPatch/diff.rb:115:in `find_match': Interrupt
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:114:in `each'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:114:in `find_match'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:110:in `upto'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:110:in `find_match'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:88:in `recursively_find_matching_blocks'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:83:in `matching_blocks'
        from ./Websites/bugs.webkit.org/PrettyPatch/diff.rb:38:in `operations'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:729:in `initialize'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:728:in `each'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:728:in `initialize'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:726:in `each'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:726:in `initialize'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:767:in `new'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:767:in `parse'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:767:in `collect'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:767:in `parse'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:509:in `initialize'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:583:in `new'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:583:in `parse'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:583:in `collect'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:583:in `parse'
        from ./Websites/bugs.webkit.org/PrettyPatch/PrettyPatch.rb:15:in `prettify'
        from Websites/bugs.webkit.org/PrettyPatch//prettify.rb:26
.Use of uninitialized value $convertedPath in substitution (s///) at ../Internal/Tools/Scripts/../../../OpenSource/Tools/Scripts/old-run-webkit-tests line 1872, <inFH> line 5.

[Adam Roben (:aroben)]

We should probably bail on finding intra-line diffs on extremely large lines so we don't hang like this.

[Adam Roben (:aroben)]

Created attachment 85367 [details]
Don't search for intra-line diffs in really long lines

[David Kilzer (:ddkilzer)]

Comment on attachment 85367 [details]
Don't search for intra-line diffs in really long lines

r=me

[Adam Roben (:aroben)]

Committed r80741: <http://trac.webkit.org/changeset/80741>

[Adam Roben (:aroben)]

Looks like this can happen on the Release bots, too.