Message11402

Author stefan.richthofer
Recipients jamesmudd, jeff.allen, stefan.richthofer, zyasoft
Date 2017-05-24.16:37:06
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1495643827.61.0.934639309485.issue2536@psf.upfronthosting.co.za>
In-reply-to
Content
Okay, I investigated a bit more and found that test_isinstance yields StackOverflowError too. All in all we'd have to exclude the following tests:

test_json:
From Lib/json/tests/test_recursion.py:
test_highly_nested_objects_decoding
test_highly_nested_objects_encoding
test_endless_recursion


test_isinstance:
test_subclass_recursion_limit
test_isinstance_recursion_limit

I will @skip these and push it soon. After that no more deadlocks of this kind should be happening in regrtests. If we continue to see such stuff, we didn't understand this issue as well as we thought...

Secondly:
For Jython 2.7.2/Jython 3 we should consider to use a Jython-specific Guava fork. We can maintain it in a manner that allows us to steadily pull updates from Guava trunk. Also we can remove all classes that Jython does not actually need - guava.x.jar is > 2.5 MB and I suspect the stuff we actually use is about 180kB or so. So as a side effect this would thin Jython dist a bit (also a frequent request IIRC).
Then we could add a public way to reset locks. Whenever in PyTableCode.call a Throwable is caught that is not an Exception, we would reset locks of certain relevant caches/dicts in Jython.
History
Date User Action Args
2017-05-24 16:37:07stefan.richthofersetmessageid: <1495643827.61.0.934639309485.issue2536@psf.upfronthosting.co.za>
2017-05-24 16:37:07stefan.richthofersetrecipients: + stefan.richthofer, zyasoft, jeff.allen, jamesmudd
2017-05-24 16:37:07stefan.richthoferlinkissue2536 messages
2017-05-24 16:37:06stefan.richthofercreate