[Tutor] can someone explain to me how this works [comparison functions]

[Danny Yoo]

On Wed, 22 Oct 2003, Kirk Bailey wrote:

> Not inverts the logical vallue of a returned logical value. Again,
> a[1]=12 and b[1]=4. with this test, and those values, the return is 0.
> Were the values reversed, this would be a 1 return, which is true, as is
> a -1 return value. IF the value HAS to be 0 or -1, we need to define the
> function with customized values. And a simple way to do it is:
>
> def compare(a,b):
> 	return (a[1]>b[1])-1


Hi Kirk,

True, it's simpler for the computer, but it's depending on the fact that
True and False are represented as 1 and 0.  Personally, I like:

###
def compare(a,b):
    if a[1] > b[1]:
        return 0
    else:
        return -1
###

better.  Even though it's more wordy, it won't surprise anyone who is
coming from a language with stricter boolean types.


But even so, I feel:

###
def compare(a,b):
    if a[1] < b[1]:
        return -1
    elif a[1] == b[1]:
        return 0
    else:
        return 1
###

is clearer as a comparison function.  Comparison functions are meant to be
triple-valued.  The original function obscured this point because it only
returned either -1 or 0, which isn't quite right.  By making three
distinct cases, we can convince anyone reading this that it is
triple-valued.

(By the way, the shortest definition I can think that's equivalent to the
above example is:  "def compare(a,b): return cmp(a[1], b[1])".)



It's very important to know that comparison functions must not be boolean
functions.  We can even see sort() malfunction if we accidently make it
return either 0 or 1!

###
>>> def brokenCompare(a, b):
...     if a > b: return 1
...     return 0
...
>>> numbers = [5,4,3,2,1]
>>> numbers.sort(brokenCompare)
>>> numbers
[5, 4, 3, 2, 1]
###

It's only by a coincidence of implementation that sort() works on a
comparison function that returns either 0 or -1.  Even so, we should
strive to make a comparison function return either -1, 0, or 1.



One last point: I've been fudging.  *grin* Comparison functions don't have
to return -1, 0, or 1: they can return negatives, zero, or positives.  If
we know this, then it's tempting to avoid writing:

###
def compare(a,b):
    if a[1] < b[1]:
        return -1
    elif a[1] == b[1]:
        return 0
    else:
        return 1
###

and instead say something like

###
def compare(a, b):
    return a[1] - b[1]
###

to avoid all that wordy comparison logic.  It's fast, it involves a single
numeric operation, so what can be the problem with it?  Actually, it is
dangerous.  It's not so bad in Python, since we have long ints, but in a
language that doesn't automagically use bignums, this definition is
broken: subtraction can overflow!

That's one reason why I personally remind myself that the range of a
comparison function is (-1, 0, 1), and not (negative, zero, positive) ---
it pretty much forces me to do the explicit comparisons.  *grin*


Talk to you later!