Issue2154

When running multiple engines in different threads, the last registered writer is in use for all script executions.

JSR 223 issue

The problem here is that the PySystemState, via PythonInterpreter, is shared across engines, which is what controls the writer output. Need to possibly reconsider this sharing setup, which is managed by PythonInterpreter.threadLocalStateInterpreter.

2.7.1 or later

Florent, I finally had a chance to go through JSR 223 support today. The change is trivial:

    [junit] Running org.python.indexer.TestBase
    [junit] Tests run: 7, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.376 sec
    [junit] Running org.python.jsr223.ScriptEngineIOTest
diff -r e1ae43e28fbf src/org/python/util/PythonInterpreter.java
--- a/src/org/python/util/PythonInterpreter.java        Wed Jan 06 22:36:07 2016 -0700
+++ b/src/org/python/util/PythonInterpreter.java        Wed Jan 06 22:55:58 2016 -0700
@@ -78,7 +78,7 @@
      * @param dict a Python mapping object (e.g., a dictionary) for use as the default namespace
      */
     public static PythonInterpreter threadLocalStateInterpreter(PyObject dict) {
-        return new PythonInterpreter(dict, null, true);
+        return new PythonInterpreter(dict, new PySystemState(), true);
     }

     /**

So I'm going to apply it now because the change makes sense: we should isolate each engine with its own PySystemState. Beta 3 should be a good chance to prove  whether this is truly the case or not.

I would like to add your test to our test suite, so if you could submit a contributor agreement (https://www.python.org/psf/contrib/contrib-form/), I will do add accordingly.

Sorry this has taken so long!

Fixed as of https://hg.python.org/jython/rev/0d15d1513251

Unfortunately Jim's fix is breaking regrtest pretty badly, so I've backed out the change. We'll have to revisit it.

Darjus didn't back out the change ;) and I did commit my fix to test_jsr223. So back in pending.

It could still be the wrong approach, but beta 3 should shake that out.