ID: 33070
User updated by: lindsay at bitleap dot com
Reported By: lindsay at bitleap dot com
Status: Open
Bug Type: Performance problem
Operating System: Linux 2.6.10 kernel
PHP Version: 5.0.3
New Comment:
I noticed the new code was crashing. You can reliably reproduce the
results by creating a file which compresses well. I created my data
sample with the following commands:
dd if=/dev/zero bs=1024 count=65536 of=large
bzip2 large
When I run the script on this, I get the following crash:
file length 256K ran in: 0 seconds
file length 512K ran in: 1 seconds
file length 1M ran in: 1 seconds
file length 2M ran in: 2 seconds
file length 3M ran in: 2 seconds
file length 4M ran in: 3 seconds
file length large ran in: *** glibc detected *** double free or
corruption (!prev): 0x0853f418 ***
Previous Comments:
------------------------------------------------------------------------
[2005-05-23 20:09:43] lindsay at bitleap dot com
I applied the patch suggested by jeff at vertexdev dot com. I also
added another test for 64M data chunks. Just to be sure, I verified
php's bzdecompress w/ the patch produces data creating the same md5sum
as bzunzip.
file length 256K ran in: 0 seconds
file length 512K ran in: 0 seconds
file length 1M ran in: 2 seconds
file length 2M ran in: 1 seconds
file length 3M ran in: 3 seconds
file length 4M ran in: 3 seconds
file length 64M ran in: 38 seconds
------------------------------------------------------------------------
[2005-05-23 16:50:02] lindsay at bitleap dot com
It looks like BZ2_bzRead() requires a file to read from. Would:
BZ2_bzDecompressInit
loop BZ2_bzDecompress //decompress in chunks
BZ2_bzDecompressEnd
work?
The source code for BZ2_bzBuffToBuffDecompress seems like it could
almost be used if the BZ2_bzDecompress was looped through 'source' in
chunks.
------------------------------------------------------------------------
[2005-05-23 10:02:56] [EMAIL PROTECTED]
jeff at vertexdev dot com:
Yes. And now look at the code.
------------------------------------------------------------------------
[2005-05-23 00:40:16] jeff at vertexdev dot com
The fix is really easy. In function bzdecompress, you need to do the
following:
1. Before the 'do' loop, initialize size to some reasonable value:
size = PHP_BZ_DECOMPRESS_SIZE;
2. Inside of the loop, double the size each time (replace the existing
'size = dest_len * iter;' statement with this):
size *= 2;
This will temporarily use up a little bit more memory than stricly
necessary, but it will make the function usable.
Drop me an email if you need more information.
------------------------------------------------------------------------
[2005-05-22 22:04:49] [EMAIL PROTECTED]
Yes, this is possibel with the BZ2_bzRead() api call.
------------------------------------------------------------------------
The remainder of the comments for this report are too long. To view
the rest of the comments, please view the bug report online at
http://bugs.php.net/33070
--
Edit this bug report at http://bugs.php.net/?id=33070&edit=1
