Or is it the implementation that is peculiar?

I'm using PythonWin 2.1 and I tried specifying a user compare function
to the list sort() function (so that I could sort file names based on
the numerical ordering instead of alphabetical) and I had difficulty
getting it working. I assumed without reading the docs carefully (if you
search the PythonWin docs for sort you get sparse information anyway)
that I could just return the result of a comparison, thusly:

> >> mylist = [1, 2, 3, 5, 4, 6]
> >> def compare1(x, y):
...  return x < y
...
> >> mylist.sort(compare1)
> >> mylist
[1, 2, 3, 5, 4, 6]

but it doesn't work. It leaves the list unchanged. Bummer.

A few days later (after writing a bubble sort - good thing I had a small
list) I was typing in PythonWin when the tooltip for sort() popped up
and it said that the return value for the compare function should be 0
or *negative* 1. So I tried this:

> >> mylist = [1, 2, 3, 5, 4, 6]
> >> def compare2(x, y):
...  return -(x < y)
...
> >> mylist.sort(compare2)
> >> mylist
[1, 2, 3, 4, 5, 6]

and it works!

On closer inspection the documentation says that the compare function is
supposed to return -1, 0 or 1. However this seems to be either
unnecessary, inaccurate, or both, since returning zero and one had no
effect, and returning 0 and -1 sorts the list perfectly.

Next time I'll have to read the docs more carefully, but somehow it
feels like this could be done better.

The behaviour certainly *feels* curious - having to return -(x < y) is a
trifle inelegant, and the 'correct' return value of:

if x >  y:
    return 1
return -(x < y

is cumbersome. In qsort() in the C library the preferred return method
is obviously something like:

return x - y;

but according to both sets of documentation that is illegal in Python,
since they specify -1, 0, 1 explicitly.

For what it's worth...


-- 
http://mail.python.org/mailman/listinfo/python-list