Is it legitimate to use items() instead of iteritems() in all places? Why was iteritems() removed from Python 3? Seems like a terrific and useful method. What's the reasoning behind it?
Edit: To clarify, I want to know what is the correct idiom for iterating over a dictionary in a generator-like way (one item at a time, not all into memory) in a way that is compatible with both Python 2 and Python 3?
In Python 2.x - .items() returned a list of (key, value) pairs. In Python 3.x, .items() is now an itemview object, which behaves differently - so it has to be iterated over, or materialised... So, list(dict.items()) is required for what was dict.items() in Python 2.x.
Python 2.7 also has a bit of a back-port for key handling, in that you have viewkeys, viewitems and viewvalues methods, the most useful being viewkeys which behaves more like a set (which you'd expect from a dict).
Simple example:
common_keys = list(dict_a.viewkeys() & dict_b.viewkeys()) Will give you a list of the common keys, but again, in Python 3.x - just use .keys() instead.
Python 3.x has generally been made to be more "lazy" - i.e. map is now effectively itertools.imap, zip is itertools.izip, etc.
dict.iteritems was removed because dict.items now does the thing dict.iteritems did in python 2.x and even improved it a bit by making it an itemview.
The six library helps with writing code that is compatible with both python 2.5+ and python 3. It has an iteritems method that will work in both python 2 and 3. Example:
import six d = dict( foo=1, bar=2 ) for k, v in six.iteritems(d): print(k, v) As the dictionary documentation for python 2 and python 3 would tell you, in python 2 items returns a list, while iteritems returns a iterator.
In python 3, items returns a view, which is pretty much the same as an iterator.
If you are using python 2, you may want to user iteritems if you are dealing with large dictionaries and all you want to do is iterate over the items (not necessarily copy them to a list)
for key in some_dict works on both - and the 2to3 tool will translate iteritems() to items() anyway...Just as @Wessie noted, dict.iteritems, dict.iterkeys and dict.itervalues (which return an iterator in Python2.x) as well as dict.viewitems, dict.viewkeys and dict.viewvalues (which return view objects in Python2.x) were all removed in Python3.x
And dict.items, dict.keys and dict.values used to return a copy of the dictionary's list in Python2.x now return view objects in Python3.x, but they are still not the same as iterator.
If you want to return an iterator in Python3.x, use iter(dictview) :
$ python3.3 >>> d = {'one':'1', 'two':'2'} >>> type(d.items()) <class 'dict_items'> >>> >>> type(d.keys()) <class 'dict_keys'> >>> >>> >>> ii = iter(d.items()) >>> type(ii) <class 'dict_itemiterator'> >>> >>> ik = iter(d.keys()) >>> type(ik) <class 'dict_keyiterator'> 
You cannot use items instead iteritems in all places in Python. For example, the following code:
class C: def __init__(self, a): self.a = a def __iter__(self): return self.a.iteritems() >>> c = C(dict(a=1, b=2, c=3)) >>> [v for v in c] [('a', 1), ('c', 3), ('b', 2)] will break if you use items:
class D: def __init__(self, a): self.a = a def __iter__(self): return self.a.items() >>> d = D(dict(a=1, b=2, c=3)) >>> [v for v in d] Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: __iter__ returned non-iterator of type 'list' The same is true for viewitems, which is available in Python 3.
Also, since items returns a copy of the dictionary’s list of (key, value) pairs, it is less efficient, unless you want to create a copy anyway.
In Python 2, it is best to use iteritems for iteration. The 2to3 tool can replace it with items if you ever decide to upgrade to Python 3.
future.utils allows for python 2 and 3 compatibility.
# Python 2 and 3: option 3 from future.utils import iteritems heights = {'man': 185,'lady': 165} for (key, value) in iteritems(heights): print(key,value) >>> ('lady', 165) >>> ('man', 185) See this link: https://python-future.org/compatible_idioms.html
If for some reason someone still needs to do this and doesn't want to use a compatibility library, a simple approach is to make a top-level callable that calls the correct one by version:
from operator import methodcaller try: {}.viewitems() except AttributeError: # Py3, use items viewitems = methodcaller('items') else: viewitems = methodcaller('viewitems') mydict = {...} for k, v in viewitems(mydict): ... This will behave identically on 2.7 and 3+ (no copying, always gets a live view). You could change all references to view from view to iter to support pre-2.7 Python, though the behavior would differ subtly in some cases (views are iterbles, but not iterators; you can call next on an iterator, but not a non-iterator iterable).
In practice though:
__future__ imports, explicit u and b prefixes for text vs. bytes literals, using io.open to replace open, etc.) and use 2to3 for the little things that you can't directly write portably.