Message10673

Author	nickmbailey
Recipients	nickmbailey
Date	2016-02-01.22:41:08
SpamBayes Score	-1.0
Marked as misclassified	Yes
Message-id	<1454366469.11.0.128909592643.issue2457@psf.upfronthosting.co.za>
In-reply-to

Content
There is an issue with the synchronization in PySystemStateCloser. It's synchronizing on the class object which is problematic because there will be many instances of that object and the synchronization lock will also need to be acquired during GC. I think we just need to switch to synchronizing on individual instances of the object. I have a script to demonstrate the issue here: https://gist.github.com/nickmbailey/979614eae0f9607616e9 I think the patch in that gist is required to make GC slow enough to see the issue or something like that. The script won't reproduce the issue until you add the patch.

There is an issue with the synchronization in PySystemStateCloser. It's synchronizing on the class object which is problematic because there will be many instances of that object and the synchronization lock will also need to be acquired during GC. I think we just need to switch to synchronizing on individual instances of the object. I have a script to demonstrate the issue here:

https://gist.github.com/nickmbailey/979614eae0f9607616e9

I think the patch in that gist is required to make GC slow enough to see the issue or something like that. The script won't reproduce the issue until you add the patch.

History
Date	User	Action	Args
2016-02-01 22:41:09	nickmbailey	set	recipients: + nickmbailey
2016-02-01 22:41:09	nickmbailey	set	messageid: <1454366469.11.0.128909592643.issue2457@psf.upfronthosting.co.za>
2016-02-01 22:41:09	nickmbailey	link	issue2457 messages
2016-02-01 22:41:08	nickmbailey	create