Message11347

Author stefan.richthofer
Recipients jamesmudd, jeff.allen, stefan.richthofer, zyasoft
Date 2017-05-04.21:45:40
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1493934342.85.0.975560034113.issue2536@psf.upfronthosting.co.za>
In-reply-to
Content
I just observed test_pythoninterpreter_jy to be hanging, which smells like another manifestation of this issue (current trunk version, i.e with increased xss). See jstack attached -- "Thread-261" is waiting on a guava internal reentrant lock without apparent reason (The jstack is not well readable due to #2584; I'd appreciate to get that one fixed!).

So, this proves that increased xss does not actually fix this issue, just makes it less likely.

_I recommend to exclude test_endless_recursion from regular test suite._

Testing this kind of edge case potentially leaves the JVM in an insane state as JVM provides insufficient guarantees after catching StackOverflowError. Note that stuff like this is the exact reason why it is strictly recommended not to catch Errors (in contrast to Exceptions). I suppose in context of test_endless_recursion it's actually okay, because we just replace the Error by a Python-style one. However that Python-style Error in turn should not be caught, i.e. the Interpreter/the JVM should not continue regular execution (i.e. for other purpose than printing errors, dumping stacktrace etc) because of potential insane state.

This IMO renders test_endless_recursion an unacceptable member of the test suite. (We can run it in separation from time to time or could move it to the end of the suite.)
History
Date User Action Args
2017-05-04 21:45:42stefan.richthofersetmessageid: <1493934342.85.0.975560034113.issue2536@psf.upfronthosting.co.za>
2017-05-04 21:45:42stefan.richthofersetrecipients: + stefan.richthofer, zyasoft, jeff.allen, jamesmudd
2017-05-04 21:45:42stefan.richthoferlinkissue2536 messages
2017-05-04 21:45:42stefan.richthofercreate