[Patches] [ python-Patches-661796 ] BZ2File leaking fd and memory

[SourceForge.net]

Patches item #661796, was opened at 2003-01-03 19:22
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=661796&group_id=5470

Category: Modules
Group: Python 2.3
Status: Open
Resolution: None
Priority: 5
Submitted By: Neal Norwitz (nnorwitz)
Assigned to: Gustavo Niemeyer (niemeyer)
Summary: BZ2File leaking fd and memory

Initial Comment:
The attached patch fixes BZ2File() leaking the fd and
PyFileObject* info (name, read ahead buffer) when an
object is deleted.

I'm not sure the patch fixes these problems in the best
way.  It exposes most of the implementation from
fileobject.c::file_dealloc() through a private API
_PyFile_Dealloc().
BZ2File derives from PyFileObject.

Make sure the file: 'empty' exists and do: 

  from bz2 import *

A simple test which demonstrates the problem is:

  for i in range(100000): o = BZ2File('empty') ; del o

Without the patch, you quickly get:

IOError: [Errno 24] Too many open files: 'empty'

You can modify this to:

  for i in range(100000): o = BZ2File('empty') ;
o.close() ; del o

Now you can see the memory leaks.


----------------------------------------------------------------------

>Comment By: Gustavo Niemeyer (niemeyer)
Date: 2003-01-22 15:39

Message:
Logged In: YES 
user_id=7887

Ok. Do you want me to work on this for 2.3?

----------------------------------------------------------------------

Comment By: Guido van Rossum (gvanrossum)
Date: 2003-01-21 19:24

Message:
Logged In: YES 
user_id=6380

FileObject sharing: I see.  But I still think it's a bad
idea, and you're better off using stdio methods directly (as
the evolution of the FileObject implementation has already
shown).

----------------------------------------------------------------------

Comment By: Gustavo Niemeyer (niemeyer)
Date: 2003-01-20 20:44

Message:
Logged In: YES 
user_id=7887

Sorry about the delay. I was on a not-so-short vacation.

nnorwitz:

Thanks for the patch. I'll check the problem (for my own
understanding) and most certainly apply it.

gvanrossum:

BZ2File used to share a lot of code with FileObject.
Unfortunately (for the BZ2File side), FileObject changed
considerably post 2.2, and some of the code had to be moved
to BZ2File itself. It still
uses some code from FileObject, though (open, close, seek,
part of the softspace handling, access to attributes and
some common methods).


----------------------------------------------------------------------

Comment By: Guido van Rossum (gvanrossum)
Date: 2003-01-05 21:11

Message:
Logged In: YES 
user_id=6380

Um, I don't understand why the BZ2File class inherits from
FileObject. It doesn't seem to be inheriting any code, and
there's no reason to inherit from FileObject just to make a
file-like object. Maybe it was a misunderstanding?

----------------------------------------------------------------------

Comment By: Neal Norwitz (nnorwitz)
Date: 2003-01-05 20:09

Message:
Logged In: YES 
user_id=33168

I've found a better solution: change the last line of
BZ2File_dealloc() from: 
  ((PyObject*)self)->ob_type->tp_free((PyObject *)self);

to
  PyFile_Type.tp_dealloc((PyObject *)self);
Updated patch attached.

----------------------------------------------------------------------

Comment By: Martin v. Löwis (loewis)
Date: 2003-01-03 19:48

Message:
Logged In: YES 
user_id=21627

Assigning to the author of the module for review. Gustavo,
if you can't find the time for a review, feel free to
unassign it.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=661796&group_id=5470