I have included in the downloadable file a full installation of Metabase and also a session handler that uses plain old MySQL calls to save the session information (this does *not* crash the server and is there to help with your testing -- it is called mysql_sessions.php and can be tested just by including it in the nabsession_test.php and nabsession_test2.php scripts in the place of metabase_sessions.php.) 

It's interesting that in the last week this bug report has not gotten a single reply. It is an easily reproducable bug that Manuel Lemos (author of the Metabase database abstraction layer) believes is a problem with PHP. He has assured me that the problem is not with Metabase so, accordingly:

There must be a bug with custom session handlers called 
using

 session_set_save_handler
("on_session_start", "on_session_end",
 "on_session_read", "on_session_write",
 "on_session_destroy", "on_session_gc");

that is making it crash when Metabase calls are used in the start/end/read/write etc. functions.

As Metabase is one of the best solutions out there for database abstraction with PHP (are there any others that allow database schema in XML and the range of type conversion options, etc? Or are as well documented?) I believe that this bug at least deserves a reply from the developer community. (Even if it is along the lines of: 'We don't care, fix it yourself' just so I know!) 

I have included a link to all code necessary to reproduce the crash in my original bug report and I've streamlined the code so that only logic necessary for the bug to be seen is present.

Most likely, none of the developers are actually USING
Metabase, so this bug is simply getting glossed over.

Perhaps a reproducible test case that does not require
usage or knowledge of Metabase would help...

IE, while we really appreciate all the work you have
gone through to document this bug, and make these scripts
available, until we can see the bug OUTSIDE of the Metabase
package, it probably won't get a lot of attention.

I have isolated the bug but did not find the cause.  It makes strtok()
crash when attempting to free memory that has been trashed.

It only happens when strtok is called from session on read or on write
handles.  I could not find what is wrong in strtok but I suspect there is
inconsistent use of PHP internal global variables (strtok_string) inside
session handle functions.  So it seems to be a serious PHP bug that may
also crash scripts that use strtok or other functions from inside session
handle functions that use PHP internal global variables

Metabase is no longer affected by this PHP bug because I have banned all
the uses of strtok function.  A new version of Metabase was uploaded to
http://phpclasses.UpperDesign.com/browse.html/package/20 .  If you use
Metabase for session handling your are strongly encouraged to download this
version.

For reproducing the PHP strtok bug without Metabase, try the script below.

<?php

function on_session_start ($save_path, $session_name) 
{
	return true;
}

function on_session_end()
{
	return true;
}

function on_session_read ($key)
{
	return true;
}

function on_session_write ($key, $val)
{
	$query="SELECT * FROM sessions";
	$select=(strtolower(strtok($query," "))=="select");
	return true;
}

function on_session_destroy ($key)
{
	return true;
}

function on_session_gc ($max_lifetime)
{
	return true;
}

// Set the save handlers
session_set_save_handler("on_session_start", "on_session_end",
						 "on_session_read", "on_session_write",
						 "on_session_destroy", "on_session_gc");

session_start();

// Register the $counter variable as part of the sesssion
session_register('counter');
$counter = 1;

echo 'Session test started';
?>

Your test handler doesn't crash PHP for me with the latest CVS version on Linux.

Please try the latest CVS from http://snaps.php.net/
or for Windows: http://www.zend.com/snapshots/

--Jani

Sascha's patch does not fix the problem, which is inconsistent memory
allocation handling by strtok function, but suppresses the sympthom which
is the crash.

Now, it just leaks memory which is good enough for normal use, but I
suppose that somebody with time and patience ought to double-check all
memory allocations that are stored in global variables like strtok_string
when made from session handler functions.

no feedback