Memory leak in _insert when using rank 0 input (Trac #572) #1170

Activity

1. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   Attachment added by @cournape on 2007-08-24: insert.2.fix
2. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   Milestone changed to 1.0.4 Release by @alberts on 2007-08-26
3. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   @stefanv wrote on 2007-08-28

   David, I'd like to write a regression test, but I don't see the same behaviour. Currently, I have:

   import numpy as N
   from numpy.lib._compiled_base import _insert
   x = N.array(1)
   _insert(x,[True],1)
   

   Would you please send me a snippet that replicates the problem?
4. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   @cournape wrote on 2007-09-05

   Well, the memory leak is obvious when looking at the code, since the function arr_insert create a new reference of avals in any case, and this is dereferenced for all cases but rank 0 input.

   Your example does not show any leak, indeed... Since the memory leak is obvious, I think we would better apply the patch anyway. I will try to get the test code I used to detect the leak in the first place.
5. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   @cournape wrote on 2007-09-05

   Ok, the following code shows the problem:

   #!python
   from numpy.lib import place
   
   while 1:
       place(1, 1, 1)
   

   Running it a few seconds under massif basically shows a linear increase of memory in the while loop.
6. numpy-gitbot commented on Oct 19, 2012

   

   numpy-gitbot

   on Oct 19, 2012

   Author

   More actions

   @stefanv wrote on 2007-09-20

   Fixed by Travis in r4049. Added test in r4053.
7. numpy-gitbot

   closed this as completedon Oct 19, 2012