Question

When I define a list and try to change a single item like this:

list_of_lists = [['a', 'a', 'a'], ['a', 'a', 'a'], ['a', 'a', 'a']]
list_of_lists[1][1] = 'b'
for row in list_of_lists:
    print row

It works as intended. But when I try to use list comprehension to create the list:

row = ['a' for range in xrange(3)]
list_of_lists = [row for range in xrange(3)]
list_of_lists[1][1] = 'b'
for row in list_of_lists:
    print row

It results in an entire column of items in the list being changed. Why is this? How can I achieve the desired effect with list comprehension?

Answer 1

Think about if you do this:

>>> row = ['a' for range in xrange(3)]
>>> row2 = row
>>> row2[0] = 'b'
>>> row
['b', 'a', 'a']

This happens because row and row2 are two different names for the same list (you have row is row2) – your example with nested lists only obscures this a little.

To make them different lists, you can cause it to re-run the list-creation code each time instead of doing a variable assignment:

list_of_lists = [['a' for range in xrange(3)] for _ in xrange(3)]

or, create a new list each time by using a slice of the full old list:

list_of_lists = [row[:] for range in xrange(3)]

Although this isn’t guaranteed to work in general for all sequences – it just happens that list slicing makes a new list for the slice. This doesn’t happen for, eg, numpy arrays – a slice in those is a view of part of the array rather than a copy. If you need to work more generally than just lists, use the copy module:

from copy import copy
list_of_lists = [copy(row) for range in xrange(3)]

Also, note that range isn’t the best name for a variable, since it shadows the builtin – for a throwaway like this, _ is reasonably common.