In Python 2.x the loop variable is part of the containing scope:

  >>> a
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  NameError: name 'a' is not defined
  >>> [a*2 for a in (1,2,3)]
  [2, 4, 6]
  >>> a
  3

In Python 3.x, it is not:

  >>> a
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  NameError: name 'a' is not defined
  >>> [a*2 for a in (1,2,3)]
  [2, 4, 6]
  >>> a
  Traceback (most recent call last):
    File "<stdin>", line 1, in <module>
  NameError: name 'a' is not defined

It is sometimes useful to know why something failed. The following Python 2 code will not work in Python 3:

  >>> import math
  >>> try:
  ...   values = [math.sqrt(x) for x in (1, 2, -3, 4)]
  ... except ValueError:
  ...   print("cannot compute sqrt(%r)" % x)
  ...
  cannot compute sqrt(-3)

I have mostly found this feature of Python 2 to be useful on the interactive shell, when trying to diagnose what caused an error in my list comprehension. Eg.

  >>> with open("tmp.dat") as input_file:
  ...   fields = [line.split()[3] for line in input_file]
  ...
  Traceback (most recent call last):
    File "<stdin>", line 2, in <module>
  IndexError: list index out of range
  >>> line
  'one two three\n'

Oops, there are only 3 columns, not 4.