The timeit code disables the garbage collector while running. I ran
the tests on my system after adding the following lines to t1() and
t2().
del base
gc.collect()
Original timetest.py:
Using proxy
100 loops, best of 3: 4.45 msec per loop
Call to dummy proxy
100 loops, best of 3: 2.24 msec per loop
Without proxy call
100 loops, best of 3: 3.01 msec per loop
With garbage collection:
Using proxy
100 loops, best of 3: 5.17 msec per loop
Call to dummy proxy
100 loops, best of 3: 4.27 msec per loop
Without proxy call
100 loops, best of 3: 3.99 msec per loop
The overhead of weakref still doesn't outweigh the overhead of cyclic
garbage collection, but they narrow the gap.
I also saw a bit of noise while running this benchmark. Different runs
were giving me varying times. Large enough to see about 20%
discrepencies.
--
http://mail.python.org/mailman/listinfo/python-list
