[Python-ideas] sum(...) limitation

[Steven D'Aprano]

On Wed, Aug 13, 2014 at 09:59:52AM -0700, Chris Kaynor wrote:

> The basic form I would use for an updated sum, which would often, but not
> always, perform better would be something like (preferably, written in C,
> not pure Python):
> 
> def sum(items, start=0):
>     value = copy.copy(start) # Make sure that start is not mutated.
>     for item in items:
>         value += item
>     return value

That's a semantic change from the existing sum: start does not have 
to be copyable. Currently I can write this:

py> class A:
...     def __copy__(self):
...             raise TypeError("I am unique!")
...
py> a = A()
py> class B:
...     def __radd__(self, other):
...             return 23
...
py> b = B()
py> sum([b, 1000, 2000], a)
3023


By changing the semantics of sum() to require start to be copyable, 
your revision has just broken my application.

 
> To avoid needing to access copy.copy, something like the following would
> also work, but with a bit more local complexity:
> 
> def sum(items, start=0):
>     count = len(items)

That's another semantic change: sum() currently accepts iterators 
(although not *infinite* iterators). This revision has now broken tens 
of thousands of applications.



-- 
Steven