[Python-ideas] Make max() stable

[Steven D'Aprano]

On Fri, Jan 17, 2014 at 11:40:49PM -0800, Devin Jeanpierre wrote:
> On Fri, Jan 17, 2014 at 5:35 PM, Ned Batchelder <ned at nedbatchelder.com> wrote:
> > Is there an example of an actual problem that stability of min and max would
> > make easier to solve?
> 
> In a language like C++, you if min and max had the property specified
> by the OP, you might do:
> 
> x = min(a, b);
> y = max(a, b);
> 
> And then x is the smallest, and y is the other one, and it's simple and
> easy and less code than an if statement.

But that's how max and min work right now, modulo that object identity 
is not important. If you care about object identity, you're probably 
doing something underhanded *wink*

Given the case that a and b are *equal* (as measured by the key 
function, if given) then it shouldn't matter whether you get

smallest = a
biggest = b

or

smallest = b
biggest = a

or

smallest = biggest = a

or 

smallest = biggest = b


These variations only are meaningful if a and b are different types 
with the same value, or the same type but different identities. Even if 
these variations are important, I don't think there is any inherent 
benefit to one over the other.

Personally, I'd either keep the current behaviour, or purely for the 
symmetry, pick the so-called "stable" behaviour. But I don't see any 
rational reason for preferring one over the other. Now that Python 3 
documents the specific behaviour, it's not worth changing.


> I suspect this is where the desire comes from.
> 
> In Python, of course, you do x, y = sorted([a, b])

Now the interesting thing here is that sorted *is* stable, so if a and b 
are equal, sorted([a, b]) is guaranteed to return [a, b]. Which gives 
the behaviour requested.



-- 
Steven