The bug can be demonstrated as follows

(In some Bourne-tpye Unix shell) type 

$ mkdir d1 d2 && 
echo 'import sys
print sys.path[0]' > d1/script.py &&
ln -s ../d1/script.py d2 && 
ln -s d2/script.py .

Now see what the  three invocations of  script.py give us:

$ python d1/script.py 
d1
$ python d2/script.py 
d2/../d1
$ python script.py 
d2

--  if script.py imported modules residing in d1, at
the third invocation the modules would not be found.

Logged In: YES 
user_id=539787

Which version of Python?  With a recent 2.4 (built on Aug 21, 
sources of Aug 20), on a 2.4.19 SMP kernel:

$ mkdir d1 d2
$ echo 'import sys
> print sys.path[0]' >d1/script.py
$ ln -s ../d1/script.py d2
$ ln -s d2/script.py .
$ python d1/script.py
/tmp/tzot/d1
$ python d2/script.py
/tmp/tzot/d1
$ python script.py
/tmp/tzot/d1

Please specify OS too.

Logged In: YES 
user_id=539787

My mistake, I missed the Pyhton 2.2.2 part, sorry.

Logged In: YES 
user_id=754125

I hope the startup message is exact enough:

Python 2.2.2 (#1, May 19 2003, 03:15:27) 
[GCC 3.1] on linux2

Logged In: YES 
user_id=666873

I confirm this behaviour on Linux 2.4.17 (glibc 2.2.4) with Python 
2.2.2 and 2.2.3; but not with Python 2.3.

The relevant change seems to be revision 2.114 of 
Python/sysmodule.c, which implemented SF patch #664376 (q.v.). 
 The change appeared in 2.3a2.

Should this be backported to the 2.2 line?