[Python-3000] Need closure on __cmp__ removal

[Lars Immisch]

Mike Klaas wrote:
> On 18-Jan-08, at 1:37 PM, Lars Immisch wrote:
>>
>> I like cmp, too. I've looked through my code, and I've only used it in
>> script-ish circumstances, but here is an example that sorts a list of
>> files by modification date:
>>
>> def cmp_mtime(f, g):
>>      """Too much for a lambda for my tastes."""
>>      return cmp(os.stat(f).st_mtime, os.stat(g).st_mtime)
>>
>> files = os.listdir('.')
>> files.sort(cmp_mtime)
> 
> The use case of sort-function delegating to cmp() of part of an object 
> is covered by the key= parameter, which is faster and more succinct:
> 
> files = os.listdir('.')
> files.sort(key=lambda f: os.stat(f).st_mtime)

You are right.

> Grepping though my team's codebase, I see some uses similar to yours 
> above, but all can be replaced with key=.  Personally, I don't see the 
> attraction of writing sort functions this way: I fell in love with key= 
> the moment I discovered it.
> 
> The relative frequency is 84 key-based sorts to 9 cmp()-based sorts 
> (admittedly that is more due to my influence than to a natural 
> distribution).

I know off hand how cmp should look like, but with 'key' I think 
'attribute access', even if it isn't. My head is stuck with 2.3, it seems.

- Lars