Message12914

I have stumbled into a test that reasonably-well reproduces what I think is the original problem. I have a test resembling:

from javax.swing.text.Utilities import *
f = getNextWord # May raise NameError
 
If I invoke this simultaneously from 100 threads, it fails just occasionally.

I attempted to disturb the timing, hoping for more frequent failures, by adding a sleep:

from java.lang import Thread
Thread.sleep(instance*10)  # May raise AttributeError
from javax.swing.text.Utilities import *
f = getNextWord            # May raise NameError

This second one fails about one time in five, claiming Thread has no sleep. I suspect it is a cache problem, not a construction one, but I don't understand how it arises.

The problem I found 27/12 (that the cache produces classes that originate in the wrong class loader) is real, and I have a test that fails every time, but I think it is not the one reported.