Python 3.0, rich comparisons and sorting order

[Andrew Dalke]

Carlos Ribeiro wrote:

> I know it, and that's exactly my question -- if this is going to be
> the way to do it in Python 3000. Today (2.4) there are FOUR ways to
> make it work: passing a compare function as an argument to sort(),
> passing a key funciton as an argument to sort(), implementing a
> __cmp__ function, or implementing the rich comparison methods. If the
> main goal of Py3K  is to have only one obvious way to do it, what one
> is it going to be? I think that the best way should be the simplest
> one: make sort() work whatever is passed to it in a reasonably way,
> and decide to have one preferred way to extend its behavior.

Get rid of __cmp__.

Then there will be:
   one native way to compare two objects to each other,
   one way to override how to compare two objects
   one way to do a Schwartzian transform

And regarding the last, "practicality beats purity."

> Now that's a harder question :-) Any ordering for multiple types is
> going to be arbitrary -- but it still may be considered reasonable if
> it reflects our common sense when it comes to ordering. In some cases,
> it's still a matter of choice, but once decided, it's deterministic,
> and that's what it takes for sorting.

So long as you allow user-defined comparisons then there's
no guarantee that it's consistent, much less deterministic.

I can define __cmp__ to do whatever I want.

 >>> class Strange:
...   def __cmp__(self, other):
...     if isinstance(other, float): return -1
...     else: return 1
...
 >>> x = Strange()
 >>> 10 < x < 5.0
True
 >>>

or for something non-deterministic

class WaitCompare:
   def __init__(self, val): self.val = val
   def __cmp__(self, other):
     while 1:
       page = urllib.urlopen("http://whitehouse.gov/").read()
       if page.find(self.val):
         break
     return cmp(self.val, other)

L = [WaitCompare("Andrew Dalke"),
      WaitCompare("America's New Communist Leaders"),
      WaitCompare("Life discovered on Mercury")]
L.sort()


				Andrew
				dalke at dalkescientific.com