Message11351

@Stefan Yes I think we have probably seen a build failure on Travis due to this deadlock in test_pythoninterpreter_jy as well https://travis-ci.org/jythontools/jython/jobs/228908723
It's good to know its not fixed by the increased Xss that was very puzzling to me!

I think excluding tests where they attempt to cause this situation, like test_endless_recursion is probably a reasonable solution. Maybe we could just have one test at the end of the suite which does this just to check the error is correctly propagated back to the python code.

I wonder if this could be solved... Maybe when we catch the StackOverflowError we could go and remove the Guava maps which might contain still locked locks and replace them with new instances. It was my feeling these are "only" caches to improve jython performance so while emptying them is generally bad if it avoids this error it might be worth considering? Its an exceptional case anyway so degrading performance might be a acceptable trade off? However I do feel we are working against the JVM on this, they really don't want you to try and recover from these situations and we might just be getting into more trouble.