Bugs item #924301, was opened at 2004-03-27 01:28 Message generated for change (Settings changed) made by svenil You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=105470&aid=924301&group_id=5470

Category: Python Library Group: Python 2.3 Status: Open Resolution: None Priority: 5 Submitted By: Sverker Nilsson (svenil) Assigned to: Nobody/Anonymous (nobody)

Initial Comment: A leak to do with readline & cmd, in Python 2.3.

I found out what hold on to my interactive objects too long ('for ever') in certain circumstances. The circumstance had to do with an exception being raised in Cmd.cmdloop and handled (or not handled) outside of Cmd.cmdloop.

In cmd.py, class Cmd, in cmdloop(), if an exception is raised and propagated out from the interior of cmdloop, the function postloop() is not called. The default function of this, (in 2.3) when the readline library is present, is to restore the completer, via:

readline.set_completer(self.old_completer)

If this is not done, the newly (by preloop) inserted completer will remain. Even if the loop is called again and run without exception, the new completer will remain, because then in postloop the old completer will be set to our new completer. When we exit, the completer will remain the one we set. This will hold on to our object, aka 'leak'. - In cmd.py in 2.2 no attempt was made to restore the completer, so this was also a kind of leak, but it was replaced the next time a Cmd instance was initialized. Now, however, the next time we will not replace the old completer, but both of them will remain in memory. The old one will be stored as self.old_completer. If we terminate with an exception, bad luck... the stored completer retains both of the instances. If we terminate normally, the old one will be retained. In no case do we restore the space of the first instance. The only way that would happen, would be if the second instance first exited the loop with an exception, and then entered the loop again an exited normally. But then, the second instance is retained instead! If each instance happens to terminate with an exception, it seems well possible that an ever increasing chain of leaking instances will be accumulated.

My fix is to always call the postloop, given the preloop succeeded. This is done via a try:finally clause.

def cmdloop(self, intro=None): ...

self.preloop() try: ... finally: # Make sure postloop called self.postloop()

I am attaching my patched version of cmd.py. It was originally from the tarball of Python 2.3.3 downloaded from Python.org some month or so ago in which cmd.py had this size & date:

14504 Feb 19 2003 cmd.py

Best regards,

Sverker Nilsson

----------------------------------------------------------------------