On Mar 11, 2005, at 17:28, Guido van Rossum wrote:
PS in the blog responses, a problem with sum() was pointed out -- unless you use the second argument, it will only work for numbers. Now
Why is that a *problem*? It solves the "end case (if the sequence is empty" which you mention for any() and all() [[indeed, I think a similar approach to any and all might be best: have them return an item from the sequence, an empty sequence uses the optional second item defaulting to something reasonable -- 0 for sum, False for any, True for all, for example]] in a neatly explicit way. As I recall, I had tried to handle non-empty sequences by picking the first item and doing following + on it (with internal implementation specialcasing for strings, to avoid an "attractive nuisance" situation), and you cut the gordian knot by pronouncing about the second argument with a default of 0 (I immediately thought your pronouncement was excellent and I still can't see why it wouldn't be).
Forcing the user to provide a 2nd arg when summing a sequence of non-numbers (say, datetime.timedelta instances, that's one example that ended up in the 2nd ed Cookbook) is good because it ensures a good return value when the sequence is empty (say, timedelta(0) rather than 0 as a number).
If you're considering revisions to sum's design, my suggestion would be to find a way to let the user tell sum to use a more accurate approach when summing floats. Summing a million floats with a loop of total+=item loses a LOT of accuracy (several digits' worth!) -- but doing the summation the right way takes O(N) auxiliary temporary space, so both approaches (the naive, performance-optimized accuracy disaster, and the expensive but accurate one) should ideally be available. An optional use_partials keyword argument defaulting to False, for example, might allow that... (again, I've hopefully clarified the issues in another 2nd ed Cookbook recipe, I guess I can post that if it helps).
Alex
_______________________________________________ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
