Well done for finding it - having fiddled around with a debugger for a while and getting nowhere, I was thinking of doing a similar thing myself.
I have absolutely no idea why the tests pass on the VM I set up (I'm running them on the VM again to make sure I wasn't seeing things, and they still pass), I'm seeing the same error as everyone else on my work laptop though. I have also run the tests on Ubuntu 9.04 (Jaunty) on both x86 and x86_64. The 32 bit machine was running Java 1.6.0_13 while the 64 bit machine was running Java 1.6.0_11. The tests passed on both. I can probably find a Windows machine to run a CI server on - is that any use, or does it need to run on official Apache machines? Jon On Sat, May 30, 2009 at 3:22 AM, David Blevins <[email protected]>wrote: > > On May 29, 2009, at 5:14 PM, David Blevins wrote: > > >> On May 29, 2009, at 12:52 PM, Jean-Louis MONTEIRO wrote: >> >> >>> ubuntu jaunty (9.04) 64 bits / JDK Sun 1.6.0_13-b03 64-Bit >>> --> Test fails >>> >>> >>> http://www.nabble.com/file/p23786057/ubuntu-jaunty-64Bit_JDK1.6.0_13-64-64Bit.log >>> ubuntu-jaunty-64Bit_JDK1.6.0_13-64-64Bit.log >>> >> >> >> >> This failure looks like tomcat may have taken too long to start and the >> client timed out. The second log posted definitely looks like what I'm >> seeing. >> >> I tried a route similar to Jon and grabbed a copy of Parallels and set it >> up with WinXP SP2. >> >> cd \test >> set JAVA_HOME=C:\Program Files\Java\jdk1.6.0_14 >> set ANT_HOME=C:\test\apache-ant-1.7.1 >> set PATH=%PATH%;%ANT_HOME%\bin >> ant test:all >> >> >> Got the expected 29 failures. >> >> >> All these CI systems we have are on linux and mac.. we totally need a >> windows setup that runs this script like every day. Stuff like this takes >> so much less time when you can narrow it down to an svn revision. >> > > So on the notion that it's easier to fix if we knew the revision I went > ahead and did a binary search. Basically just started with middle revision > between now and when 3.1 was released and kept splitting it in half till i > pinned down the revision. > > svn up -r $(( ($BEGIN + $END) / 2 )) && mvn clean install $skip -Dassemble > && installjars.sh > > If the tests failed I'd set END to the current version, if they passed I'd > set BEGIN to the current version and then just rerun. > > So anyway, this is the commit that did it: > http://svn.apache.org/viewvc?rev=769395&view=rev > > It claims to be for windows, still checking it out. If I revert the change > on my copy of trunk then all itests pass on windows with no issues. I > suspect the context path we get from Tomcat is URL syntax and not path > syntax, so the fix actually breaks things on windows rather than fixes > anything. I suspect that it's the test case that is faulty. Not sure what > the issue is, but I remember seeing people on Tuscany complain with similar > path issues. > > Still looking at the test case in question.. > > Anyway we should be good to re-roll soon. > > -David > >
