something Evil happens when large hashes destroyed

guuge guuge at localhost.localhost
Mon Nov 19 16:23:19 CET 2001


On Sun, 18 Nov 2001 18:39:52 -0500, Tim Peters <tim.one at home.com> wrote:
> [guuge]
>> I was trying to sort about 100,000 items by splitting them into
>> groups (using a python dictionary type) and then successively splitting
>> these groups until they were small enough to use a brute force method.
> 
> Try list.sort().
> 
>> My test program, which used only two or three keys to create the first
>> split, worked fine.  When I tried the real thing, using 256 keys, the
>> program slowed to a crawl.  The python interpreter took forever to
>> destroy the dictionaries.
>>
>> Here's a little program ...
> 
> I tried that w/ 2.2b2 under Win98SE, 866MHz, 256MB RAM, also with 100,000
> elements, and the number of bins didn't matter:  for assorted values from 1
> through 100,000, creation time was 2 seconds and destruction time 0
> (time.time() only has 1-second resolution on Windows).
> 
> So best guess is that you're fighting your platform C library's
> malloc()/free().  Which OS?  Which C compiler and library?  

Linux (2.4.2 kernel)
libc-2.2.4
gcc 2.95.4

> If you build Python yourself, try rebuilding 2.2b2 --with-pymalloc, 
> and see whether that helps.  

The problem is fixed using "--with-pymalloc".  Thank you!

> It would be interesting to know one way or the other.



More information about the Python-list mailing list