While we're talking about annoyances

[Alex Martelli]

Right. Using sort(key=keyfunc) is supposed to be faster than
decorate-sort-undecorate. And I think it is clearer too.

Right, and speed comparisons are pretty easy (just make sure to copy the
list within the loop to compare like with like...:

brain:~ alex$ python -mtimeit -s'L=range(-3,5)' 'X=L[:]; X.sort(lambda
x, y: cmp(abs(x), abs(y)))'
100000 loops, best of 3: 13 usec per loop

brain:~ alex$ python -mtimeit -s'L=range(-3,5)' 'X=[(abs(x),x) for x in
L]; X.sort(); X=[x for _,x in X]'
100000 loops, best of 3: 7.85 usec per loop

brain:~ alex$ python -mtimeit -s'L=range(-3,5)' 'X=L[:];
X.sort(key=abs)'
100000 loops, best of 3: 4.23 usec per loop

The "intrinsic" DSU done by key= is clearly superior on all scores.

The semantics are subtly different: it guarantees to never compare
anything _except_ the key (because that's typically what one wants), so,
make sure they key includes everything you may ever want compared.


Alex