Issue1639

Symptom:
Deploying a web app in JBoss 5.x (and later, I think), the classpath is ['__classpath__', '__pyclasspath__/'] but it should also contain $JBOSS_HOME/{DEPLOYMENT}/WEB-INF/lib/Lib and $JBOSS_HOME/{DEPLOYMENT}/WEB-INF/lib/jython.jar/Lib. Therefore, one is unable to import any libraries and one receives the error: "ImportError: No module named ..."

Reason for the error:
JBoss 5.x introduces the vfszip protocol for jar files. org.python.core.PySystemState uses the jar:file protocol name in an attempt to parse a string that contains the jar filename out of a string that looks like "jar:file:/some/path/jython.jar!/org/python/core/PySystemState.class" in other containers. JBoss returns "vfszip:/some/path/jython.jar/org/python/core/PySystemState.class". Note the missing "!" and the protocol name is different.

The fix:
The fix is to also test for the vfszip: string and parse the resulting string in a way that extracts the jar filename. A patch for 2.5.2b1 is attached.

Additional Note:
It looks like JBoss is staying with vfszip, so this issue is not likely confined to version 5. The issue was originally spotted in Jython version 2.5.1, so that is a confirmed affected version as well.

Thanks for the patch - i'll take care of it!

promised to fix rigth after JavaOne

Eric,

on Windows, the URL contains a leading slash right after vfszip:, e.g.:
vfszip:/C:/Bison/lchload2/JBoss5/bin/jboss/server/bison/deploy/bison.ear/extlib/jython.jar/org/python/core/PySystemState.class

This needs to be removed in order to get a working path.

Can you confirm that everything after vfszip: belongs to a valid path on *nix (like your patch suggests) ?

Yep, its all valid path. On the Solaris 10 system I'm using and the Fedora Linux 12 system, they have numerous log file references that speak to vfszip:/a/valid/path. So, we would want everything after the ":", minus the /org/python/core/PySystemState.class, just as the patch does.

So, a question for you. Are you concerned about the leading "/" before the "C:"? I didn't make an exhaustive search of PySystemState.py, but I see the result of the method call being used at least in findRoot which uses it in a File class. At least on 1.5.0_22, a leading slash is okay.

Test:
File testFile = new File("/D:/tmp/Foo.txt");
if(testFile.isFile() && testFile.canRead())
  System.out.println("It is a readable file");
BufferedReader input = new BufferedReader(new FileReader(testFile));
System.out.println(input.readLine());

Outputs:
It is a readable file
This is the contents of Foo.txt  (which is what is in Foo.txt, so it works fine)

Thanks for your care and attention to this issue.

Oh, I re-read your message and see you are concerned about it. Based on my test, do you still think it is an issue?

Eric,

many thanks for all your tests!

In short: yes i am still concerned about the leading slash on windows.
Our Jython embedding in JBoss 5 did not work this way, i had to remove the slash. Sadly i don't exactly remember why, and i have no possibility to verify it until in 2 weeks.

My suggestion is to distinguish between windows and other platforms for the moment

The attached 1639patch.txt fixes the problem - to the best of my knowledge

fixed in revision 7141

Thanks Oti. Looks great!