Issue1482

The problem shows when scripts are compiled repeatedly on the fly.
Slowly but surely the PermGen space will fill up with no way to recover.

I can provide a patch against Jython 2.2.1 that uses a Weak Map instead.
From what I googled this is even supposed to be more efficient than
String.intern.

This resource leak -- which should be applicable to 2.5, I don't believe anything has changed in the use of intern usage in String -- has escaped attention. We need to fix it before 2.5.2 beta 2 too. (2.5.2b2 is really trying to address all serious resource leaks like this one.)

Fortunately it should be an easy fix, and it's even something that will clean up our code in a bit in PyString.

I was unable to reproduce this problem on trunk. Using a canonical map (using weak keys) instead of intern seems to make sense, but monitoring with JMX doesn't show any difference.

Attached is a patch to use a canonical map that passes regrtest; and my test program that should intern everything, but also will GC since nothing is held - the globals map is clean each time, that is we do

exec stmt in {}

I can see this patch being useful in 2.6, but without seeing benefit here, it seems too late to apply for 2.5. Most importantly, there are still other uses of intern in our codebase, and usage of the API might also assume that.

Also see the irc discussion on #jython, http://www.extreme.st/jython.extreme.st/irclogs/index.php/?date=2010-09-04

Please reopen if you can submit a reproducible test case - it's important we get this right.