[Python-Dev] Python GC/type()/weakref mystery

[Aahz]

On Thu, Mar 18, 2004, Kevin Jacobs wrote:
> Aahz wrote:
>>
>>This looks to me like standard boundary conditions for GC; GC normally
>>only gets provoked when more than gc.get_threshold() objects have been
>>created than deleted.  What makes you think something unusual is
>>happening?
> 
> I realize that this does look like standard GC threshold behavior
> However, it /shouldn't/, because I manually run gc.collect() both
> before and after the test suite.  The other thing that is strange is
> that all of these dead weakrefs disappear when I run gc.collect() as
> part of each iteration of the test suite, but not afterwards.
>
> Please read over my original mail again, since the magnitude of
> weirdness may not have been obvious the first time through.

Hmmmm...  First of all, not all.  When I tried your test with N=1, I did
still get one leaked ref.  I tried running with an extra gc.collect() at
the end of each run, and that didn't help.  I've verified that
gc.collect() immediately after the running test_func() does reap a bunch
of stuff.  And I've verified that the sum of len(gc.getobjects()) for
each run equals that of running it at the start and end of the whole
program.

I'll let someone else who can look at the C code take over....
-- 
Aahz (aahz at pythoncraft.com)           <*>         http://www.pythoncraft.com/

"usenet imitates usenet"  --Darkhawk