There's a leak on compiling code like 'lambda a,a:0'. 
TrackRefs says the following are being leaked:

<type 'list'>                                         
        4       16
<type 'str'>                                          
        2       10
<type 'symtable entry'>                               
        2        8
<type 'dict'>                                         
        2        8
<type 'int'>                                          
        0        2

It doesn't seem to leak for, e.g. 'def f(a,a):pass'.

I had a look at the code, but got very confused, I'm
afraid.

Logged In: YES 
user_id=6656

Well, this was easier to fix than I expected (one line patch
attached, which I'll check in soon).

The mystery has transformed itself into "how the hell did
the code work in the first place?" and "why didn't the
function version leak?".

I think I know the answer to the first one, but I'm not
telling :-) I may go as far as posting it to c.l.py as a
Python Mystery Theatre style poser...

Logged In: YES 
user_id=31392

I changed the code, but I think I still see a leak.  At the
interpreter prompt,
f = lambda a,a: None
shows refs growing by one every time.

Logged In: YES 
user_id=31392

But a more careful test shows that totalrefcount doesn't go
up if I compile this code in a loop.  So it looks like it's
fixed.