Memory leak (or circular reference) when slicing Column

After slicing a Column and discarding the reference to the original Column, the memory of the Column cannot be released. It seems that the original column is still referenced somewhere invisible.

Here’s the code to reproduce the issue. When using top to see memory usage, one finds that it climbs up quickly (and eventually runs out of memory) even though it shouldn’t.

# WARNING: following code results in excessive memory usage

import gc
import numpy as np
from astropy.table import Column

col = Column(data=np.random.rand(10000000), name='random')
for _ in range(1000):
    col = col[col > 0.5]
    gc.collect()

Below are the versions that I used, installed with conda, and are running on a 64-bit linux machine:

astropy 3.0.4
numpy   1.15.1
python  3.6.6

This issue might be related to #6276 (but the fix is in 3.0.4) and/or #6346, but I am not entirely sure.

[mhvk]

Numpy PR with fix at https://github.com/numpy/numpy/pull/11989

[mhvk]

Thanks for reporting! I tried to reproduce and think this may be a numpy problem: I could not reproduce your problem with astropy 3.0.4, numpy 1.14.5 (Debian testing), or with current astropy but still numpy 1.14.5, but I do reproduce it when I use numpy current master, independent of astropy version; given your result, I suspect it started in numpy 1.15.

With a quick test, I found this is a problem for even the simplest possible ndarray subclass, so this is an upstream bug: See https://github.com/numpy/numpy/issues/11983