Issue2573

I have observed test_weakref fail a few times while running ant regrtest. The output is:

     [exec] test test_weakref failed -- Traceback (most recent call last):
     [exec]   File "/home/james/git/jython/dist/Lib/test/test_weakref.py", line 1029, in test_weak_values
     [exec]     self.assertEqual(len(list(dict.iterkeys())), 0,
     [exec] AssertionError: deleting the values did not clear the dictionary

The failure appears to be intermittent, I haven't reproduced it when running the test individually, and I have never observed it on Windows.

So our assumption on these sorts of testing bugs is that multiple phases of GC are often required to clear out weak references; and GC in general is never quite as deterministic as we would like for such testing. (CPython's model in contrast does give us deterministic GC, so it's a bit harder to use the same tests. In practice, it would not matter - one should not rely on determinism in GC even in CPython, given the possibility of cycles.) You can see the specific code used in the extra_collect function.

In particular, test_weak_values does not use the extra_collect function; it instead uses gc.collect directly. It would be worthwhile revisiting with extra_collect here to see if it makes a difference.

Note that if the weakref is never collected, that would be a bug!