Message322061
| Author |
terry.reedy |
| Recipients |
amper, bethard, rhettinger, serhiy.storchaka, terry.reedy |
| Date |
2018-07-21.02:17:49 |
| SpamBayes Score |
-1.0 |
| Marked as misclassified |
Yes |
| Message-id |
<1532139469.9.0.56676864532.issue34149@psf.upfronthosting.co.za> |
| In-reply-to |
|
| Content |
In 2.x, map(None, 'abc', 'zyz') == [('a', 'z'), ('b', 'y'), ('c', 'z')], but with the addition of zip, so that zip('abc', 'xyz') has the same result, we deprecated that use of None to mean identity function.
For python-coded functions, a default is needed to make a keyword-only argument optional, and preferred over use of *args for making positional arguments optional. Unless I am missing something, a function can special-case 'key is identity', to avoid overhead, just as well as it can special-case 'key is None'. So rather than extend 'key=None', to me a kludge, I would rather replace it with 'key=identity'. Both can be accepted during a deprecation period.
For instance, after adding identity,
def nsmallest(n, iterable, key=identity):
...
if key is identity or key is None: # key in (identity, None)
result = min(it, default=sentinel)
...
Since no one need ever write key=None, explicit passing should be rare. It seems to me that the main reason for the type declaration of key to include None is so that the def statement itself passes a consistency check with the None default. Once that is changed, most people should be able to use a simple callable declaration. I am considering this for python-ideas.
Since the weekly issues list came out just 10 hours ago, I will not close this yet, but I will if still open in couple of days and no coredev objections. |
|