Consider the following code:

for z in zip([1]):pass

2to3 does not convert the zip in this code to list(zip(...)); it does not change this code at all.

That can be an (obscure) bug because the zip in Python 2 has different semantics from the zip in Python 3.

The output of this program

from __future__ import print_function
S = []

def l(c):
  for i in [0,1]:
    S.append(c)
    yield i
  S.append(c.upper())

la = l('a')
lb = l('b')

for a,b in zip(la, lb):
  S.append("#")
print(''.join(S))

is different in Python 2 and Python 3 (when converted with 2to3, which doesn't change the program).

In Python 2 the output is:

ababA##

In Python 3 the output is:

ab#ab#A


Obviously this example is somewhat contrived, but I have a non-contrived example involving decoding PNG images (if anyone is interested).

Hm, I would expect that in 99 times out of 100 the extra list(...) would be removed in a manual step following the automated conversion.

I'd really like to see the non-contrived example with a justified use of this evil side effect ;)

As far as a non contrived example, see http://bugs.python.org/issue21628, which was marked as a duplicate of this bug.

This bug is the main thing preventing me from using 2to3, so I think it is a real issue.