We have a object method that returns a city/state tuple, i.e.
('Boston', 'MA'). Under some valid circumstances, there is no valid city/state to return. Stylistically, does it make more sense to return
None, or a two element tuple containing
(None, None) in that case?
I would return
None. If there is no result, why return something that looks like a result?
It is also easier to test:
result = getCity() if result: # do something
I would only return
(None, None) if it were possible that only one of the two values is
('Boston', None)). It would be more consistent in this case.
By only returning one value in exceptional circumstances, you risk breaking the tuple unpacking idiom. Some of your callers might issue:
city, state = getCityStateTuple("something")
In that case, returning
None will break the caller with the error:
TypeError: 'NoneType' object is not iterable
So, I personally would return
(None, None) in your situation. Then again, your mileage may vary, and it depends on the pattern used by your callers.
As others have noted, a tuple with items in it does not test as
False, which is one reason you might want to return
None rather than
(None, None). However, it is possible to write a tuple subclass that tests as
False even when it has items in it by overriding its
class falsetuple(tuple): def __nonzero__(self): return False
Then you could return
falsetuple((None, None)) when there is no value available. In fact, you could always return the same
I’m not necessarily recommending that you do this, in fact I have serious misgivings about flouting this convention, I’m just saying that the truthiness of non-empty tuples is not necessarily in itself a reason to not return a tuple.
(None, None) does not evaluate to
False in Python. In addition, building a tuple requires more work than, well, not building a tuple. So I would prefer
If your routine normally returns a tuple, then a tuple is what it should keep returning. The real choice is between returning
(None, None), or raising an exception, and we don’t have enough information to offer good advice on that.
If it were me, and I chose the tuple over the exception, I would go with the FalseTuple that kindall suggests, and also realize that the calling code (which is using tuple unpacking) can also test
if city is None:
to see if a valid result was obtained. This way you are supporting tuple extraction across all possible return values, and still allowing the pythonic idiom of asking the object, “Do you evaluate as True?” (Here is kindall’s again for completeness):
class FalseTuple(tuple): def __nonzero__(self): return False
why not making State a property of City? That way your function would return always one value: a City or None.
Returning (None, None) is bad for all the reasons stated in the other answers and serves only to support tuple unpacking.
None is the best value to return to state that no valid city can be returned, but having a function returning 1 or 2 values is not that good, again because of tuple unpacking.
To me, returning (None, None) would imply that (None, State) or (City, None) would also be valid return values.
If that is the case, go with (None, None), otherwise, Felix and Brent provide very good arguments for simply returning None.
I would implement a public method for object that returned, let say
isValidLocation() that returns true if location is valid and false if location is none.
If you return None, it will be much easier for you to check the return value.