I apparently replied to soon. Removing /usr/lib/python2.4 from PYTHONPATH did not solve the problem. I think I may have had a launcher-started gedit running somewhere in the background while testing. Any subsequent terminal-launches would then just create new windows for the existing (non-bugged) process, rather than starting a new process and tripping the bug (and by bug I mean configuration error on my part, most likely).
However, I went further along the sys.path diff and found a path to a second python2.6 installation. Apparently, this one shipped with a standalone middleware client for distributed computing, and was insinuated into my PYTHONPATH via a call to its startup script in my .bashrc. Removing the call to the startup script solved the problem again (!) I still can't explain the traceback though. But fwiw, this solution made the problem stay solved past a reboot, so I have high hopes this time. /Joel -- http://mail.python.org/mailman/listinfo/python-list